id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0005/cond-mat0005343.html
|
ar5iv
|
text
|
# Effects of ac-field amplitude on the dielectric susceptibility of relaxors
## Abstract
The thermally activated flips of the local spontaneous polarization in relaxors were simulated to investigate the effects of the applied-ac-field amplitude on the dielectric susceptibility. It was observed that the susceptibility increases with increasing the amplitude at low temperatures. At high temperatures, the susceptibility experiences a plateau and then drops. The maximum in the temperature dependence of susceptibility shifts to lower temperatures when the amplitude increases. A similarity was found between the effects of the amplitude and frequency on the susceptibility.
Relaxor ferroelectrics (relaxors) have been studied for nearly 40 years since Pb(Mg<sub>1/3</sub>Nb$`{}_{2/3}{}^{})`$O<sub>3</sub> (PMN) was first synthesized by Smolenski et al.. The dielectric response of relaxors is characterized by the diffuse phase transition (DPT) and a strong frequency dispersion. Various models, such as the compositional heterogeneity model, the superparaelectric model, and the glasslike model, et al., were proposed to rationalize the complicated behaviors of relaxors. It is widely accepted nowadays that the presence of polar microregions in nanoscale is responsible for the relaxor behaviors.
The effects of the applied ac field on relaxors cause great interest since they provide some clue of the relaxation mechanism. Glazounov et al. observed that the dielectric permittivity of PMN increases with increasing amplitude of the applied ac field. A similarity was also found between the effects of the amplitude and frequency on the permittivity. In addition, the ac-drive-enhanced relaxor characteristics and domain breakdown were observed in (PbLa)(ZrTi) (PLZT). There are two possible mechanisms, i. e., domain-wall motion model and superparaelectric model, to explicate the nonlinearity of dielectric permitivity of PMN relaxors. Glazounov et al. suggested that it is related to domain-type process rather than thermally activated flips of the local spontaneous polarization (i.e. superparaelectric model). However, they did not consider the interaction of polar microregions when investigating the superparaelectric model, which is just one of the key points related to response of the external field. In this study, we conduct a Monte Carlo simulation to investigate the influence of measuring field on the dielectric susceptibility of relaxors.
We investigate the thermally activated flipping process of the local spontaneous polarization in relaxors. Following the work of Gui et al., the polar microregions are regarded as point dipoles. Then relaxors are modeled to be a system consisting of Ising-like dipoles with randomly distributed interactions:
$$H=\underset{ij}{}\stackrel{}{J_{ij}}\sigma _i\sigma _jE_{ext}\overline{\mu }\underset{i}{}\frac{|\mu _i\mathrm{cos}\theta _i|}{\overline{\mu }}\sigma _i,$$
(1)
where $`\sigma _i,\sigma _j=\pm 1`$ are dipole spins. When the projection of the $`i`$th dipole moment $`\stackrel{}{\mu }_i`$ on the direction of the external field $`\stackrel{}{E}_{ext}`$ is positive, $`\sigma _i`$ takes value +1, otherwise, $`\sigma _i`$ takes value -1. $`\theta _i`$ is the angle between $`\stackrel{}{\mu }_i`$ and $`\stackrel{}{E}_{ext}`$, and $`\overline{\mu }`$ is the maximal magnitude of the dipole moments. $`\stackrel{}{J_{ij}}`$ is the effective interaction energy between the nearest neighbor dipoles, which has a Gaussian distribution with a width $`\mathrm{\Delta }J`$. $`\stackrel{}{J_{ij}}`$ reflects the correlation between polar microregions, which is essential to the glassy behaviors. In general, the external field contains a measuring ac field and a bias dc field. In this paper, only the ac field is involved, i.e.,
$$E_{ext}=E_0\mathrm{exp}\left(i2\pi \frac{t}{t_L}\right),$$
(2)
where $`t`$ is the real time. $`E_0`$ and $`t_L`$ are the amplitude and the period of the ac field, respectively.
The Monte Carlo simulation is performed on a $`16\times 16\times 16`$ simple cubic lattice with periodic boundary conditions. The details of simulation process can be found in Ref. 15. The dielectric susceptibility is defined as
$$\chi =C\frac{\frac{1}{t_{obs}}\underset{t_0}{\overset{t_0+t_{obs}}{}}p(t)\mathrm{exp}\left(i2\pi \frac{t}{t_L}\right)𝑑t}{E_{ext}},$$
(3)
where $`C`$ is a proportional factor which is chosen to be 1 in this contribution, and $`\mathrm{}`$ denotes the configurational averaging. $`p(t)`$ is the normalized polarization:
$$p(t)=\frac{1}{N}\underset{i}{}\frac{|\mu _i\mathrm{cos}\theta _i|}{\overline{\mu }}\sigma _i.$$
(4)
During the simulation process, $`p(t)`$ is recorded and $`\chi `$ is calculated according to Eq. (3). We choose $`t_0=200`$MCS/dipole to eliminate the influence of the initial state and $`t_{obs}=3000`$MCS/dipole to be the observation time. The simulation is performed in many runs with different initial conditions so that the configurational averaging can be done. Longer observation time was also adopted in test, but no obvious influence on results was observed.
In order to verify the validity of the method, the dielectric susceptibility under a weak field is firstly calculated. The result is shown in Fig. 1. It can be seen that the susceptibility $`\chi `$ reaches its maximum at a certain temperature ($`T_m`$) and changes gradually around $`T_m`$, which is known as the diffuse phase transition (DPT) in relaxors. A strong frequency dispersion can be also observed: $`\chi `$ decreases with increasing field frequency at low temperatures, and $`T_m`$ moves to higher temperatures. All these characteristics are consistent with the experiments and the previous theoretical results.
Now, let us investigate the effects of the field amplitude on the dielectric susceptibility. The susceptibility curves under different ac-field amplitudes $`E_0`$ are depicted in Fig. 2 when the measuring frequency is kept as $`t_L=10`$MCS/dipole. (We express the frequency by $`t_L`$ here and hereafter.) From Fig. 2 one can list the most essential features of the nonlinear effect: (1) the dielectric susceptibility increases with increasing $`E_0`$ at temperatures $`T<T_m`$ where the frequency dispersion is observed; (2) increasing $`E_0`$ will make the maximum in the temperature dependence of $`\chi `$ shifts to lower temperatures, which has the similar effect of decreasing frequency (see also Fig. 1). The change of the imaginary part, $`\chi ^{\prime \prime }`$, shows similar features in the simulation. These features agree with the experiments in PMN very well. The concepts of “slow dipole” and “fast dipole” can help to understand the increasing of the susceptibility. Slow dipoles are those dipoles which flip too slow to keep up with the changing of the ac field and give no or little contribution to the dielectric susceptibility. At low temperatures, there are large amounts of slow dipoles. When $`E_0`$ increases, the driving force on slow dipoles is enhanced. Slow dipoles are forced to flip faster and they give more contribution to the dielectric susceptibility $`\chi `$. For fast dipoles, the contribution changes slightly at low drives (see below). As a result, the susceptibility $`\chi `$ increases with increasing $`E_0`$.
It can be seen in Fig. 2 that the dielectric susceptibility slightly decreases with increasing the external-field amplitude $`E_0`$ at high temperatures. The tendency is weakened at higher frequencies while becomes more evident at lower frequencies. Fig. 3 shows the cases for a lower frequency $`t_L=50`$MCS/dipole. It shows that the dielectric susceptibility decreases at high temperatures and increases at low temperatures when $`E_0`$ increases. These results are similar to the experimental cases in PLZT to some extent. However, the computed maximum in $`\chi (T)`$ decreases with increasing $`E_0`$, which is opposite to the experimental observations. It reflects the defect of the model or/and the method we used.
To get further knowledge of the continuous effects of the ac-field amplitude, we plot in Fig. 4 the curves of $`\chi `$ as functions of the amplitude $`E_0`$ for different temperatures when the measuring frequency is fixed as $`t_L=10`$MCS/dipole. At low temperatures, the dielectric susceptibility increases first, and then drops with increasing $`E_0`$. This means that the applied field speeds up the flipping of dipoles at small $`E_0`$ values so $`\chi `$ increases first, while the system is nearly saturated at large $`E_0`$ values which causes the drop of $`\chi `$. At high temperatures, the dielectric susceptibility experiences a plateau at the beginning and then decreases when the applied field increases. These results are consistent with the experiments in PMN when $`E_0`$ varied in wide range of values. In Ref. 7 and Ref. 8, $`E_0`$ is not large enough, so $`\chi `$ increases at low temperatures and remains steady at high temperatures with increasing $`E_0`$.
Fig. 5 demonstrates the field dependence of $`\chi `$ at different measuring frequencies and a fixed temperature $`T=1.5\mathrm{\Delta }J/k_B`$. It shows that the maximum of the curve shifts to lower field amplitude when decreasing the measuring frequency. The shapes of curves are similar for different frequencies.
Fig. 6 shows the temperature of the susceptibility maximum ($`T_m`$) as a function of the external-ac-field amplitude $`E_0`$. A nonlinear relation can be found between $`T_m`$ and $`E_0`$. It is conflict with the linear law observed in experiments. Perhaps the field used in experiments is not large enough to reveal the high-order effects of the $`T_mE_0`$ curve. Further experiments are needed to testify the theoretical predictions.
The Eq. (3) could be generalized to include the Fourier component at different frequencies than that of $`E_{ext}`$. Fig. 7 gives the curve of $`\chi _{2\omega }/\chi `$, where $`\chi _{2\omega }`$ is the second-order component of the susceptibility. It can be seen that $`\chi _{2\omega }/\chi `$ is stronger at lower $`E_{ext}`$ and $`T`$.
By means of the results above, we can see that the behaviors of the system described by the model Hamiltonian in Eq. (1) are consistent with many aspects of the experiments when the applied-ac-field amplitude varies. There are two points that should be mentioned here. First, the interactions between polar microregions play an important role in the dielectric response. If the interaction does not exist, the dielectric susceptibility will decrease with increasing field amplitude as what is pointed out by Glazounov et al.. Secondly, the model in Eq. (1) is a rather simplified model. It cannot reflect the effects of external field on the crystal structure completely. It describes only the thermally activated flips process of the local polarization. Indeed, there may be more dielectric mechanism in relaxors. For example, It was presented that there may be two kinds of polarization processes in relaxors. Very recently, various types of contributions were found to dominate the dielectric response within different ac-drive amplitude ranges.
In conclusion, the simulation results suggest that the thermally activated flips of the local spontaneous polarization in relaxors plays an important role in producing the relaxation phenomena.
This work was supported by the Chinese National Science Foundation.
* Electronic address: zrliu@phys.tsinghua.edu.cn
|
warning/0005/hep-th0005196.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
In recent developments of string theory, the deep connection between supergravity and super Yang-Mills theory has clarified. One remarkable example is the AdS/CFT correspondence . In this researches, many supergravity theories which can be obtained by dimensional reduction from 11-dimensional supergravity have played important roles .
On the other hand, some years ago, T. Banks, W. Fischler, S. H. Shenker and L. Susskind (BFSS) proposed that Matrix theory gives a complete description of light-front M-theory . It had been proposed as a theory of D0-branes by E. Witten . The candidate of its extension on curved backgrounds is the supermembrane theory. It is described as nonlinear sigma model and couples to 11-dimensional superspace backgrounds that satisfy a number of constraints which are equivalent to 11-dimensional on-shell supergravity .
Thus, it is important that getting much knowledge of 11-dimensional superspace structure. Nevertheless we have little knowledge of it. By E. Cremmer, S. Ferrara, L. Brink and P. Howe the on-shell conditions and components of superfields up to first order in anticommuting coordinates was investigated . By B. de Wit, K. Peeters and J. Plefka the components of 3-form superfield and part of components of vielbein superfields up to second order in anticommuting coordinates was investigated . In the previous paper, the remaining components of vielbein superfields up to second order in anticommuting coordinates was investigated .
In this paper, we compute part of components of connection superfields and the components of Lorentz superparameter at second order in anticommuting coordinates in terms of the component fields of 11-dimensional on-shell supergravity by using ‘Gauge completion’. This configuration of superspace holds the $`\kappa `$-symmetry for supermembrane Lagrangian and represents 11-dimensional on-shell supergravity.
The paper is organyized as follows. In section 2, we explain ‘gauge completion’. In section 3, we compute parts of the superfields. Our conventions are summarized in Appendix.
## 2 Gauge Completion
‘Gauge completion’ was introduced to identify superspace representation as on-shell supergravity . In this section we review this method.
‘Gauge completion’ is searching for structures of the superfields and superparameters which are compatible with ordinary supergravity. That is to say, supertransformations (2.5) - (2.7) are identified as transformations in 11-dimensional spacetime (2.1) and the $`\theta =0`$ components of superfields and superparameters are identified as the fields and parameters of ordinary supergravity.
### 2.1 Supersymmetry algebra
Supersymmetry transformations in components formalism are as follows,
$`\delta _se_m^a`$ $`=`$ $`2\overline{ϵ}\mathrm{\Gamma }^a\psi _m,`$
$`\delta _s\psi _m`$ $`=`$ $`D_m(\widehat{\omega })ϵ+T_m^{rstu}ϵ\widehat{F}_{rstu}\widehat{D}_m(\widehat{\omega })ϵ,`$
$`\delta _sC_{klm}`$ $`=`$ $`6\overline{ϵ}\mathrm{\Gamma }_{[kl}\psi _{m]},`$ (2.1)
$`\text{with},T_m^{rstu}`$ $``$ $`{\displaystyle \frac{1}{288}}(\mathrm{\Gamma }_m^{rstu}8\delta _m^{[r}\mathrm{\Gamma }^{stu]}),`$ (2.2)
where $`\widehat{F}(=F_{klmn}+12\overline{\psi }_{[k}\mathrm{\Gamma }_{lm}\psi _{n]})`$ is the supercovariant field strength, and $`\widehat{\omega }(=\omega _{mb}^a+\frac{1}{2}\overline{\psi _n}\mathrm{\Gamma }_{mb}^{anp}\psi _p)`$ is the supercovariant spin connection. And other notation is the same as that in .
Its algebra is as follows,
$`[\delta _{susy}(ϵ_1),\delta _{susy}(ϵ_2)]=\delta _g(\xi _3)+\delta _s(ϵ_3)+\delta _l(\lambda _3)+\delta _c(\xi _{3mn}),`$ (2.3)
where
$`\xi _3^m`$ $`=`$ $`\overline{ϵ}_2\mathrm{\Gamma }^mϵ_1(12),`$
$`ϵ_3`$ $`=`$ $`\overline{ϵ}_2\mathrm{\Gamma }^nϵ_1\psi _n(12),`$
$`\lambda _{3b}^a`$ $`=`$ $`\overline{ϵ}_2\mathrm{\Gamma }^nϵ_1\widehat{\omega }_{nb}^a+{\displaystyle \frac{1}{144}}\overline{ϵ}_2(\mathrm{\Gamma }_b^{arstu}\widehat{F}_{rstu}+24\mathrm{\Gamma }_{rs}\widehat{F}_b^{ars})ϵ_1(12),`$
$`\xi _{3mn}`$ $`=`$ $`\overline{ϵ}_2\mathrm{\Gamma }^kϵ_1C_{kmn}\overline{ϵ}_2\mathrm{\Gamma }_{mn}ϵ_1(12).`$ (2.4)
On the other hand, transformations in superspace formalism are as follows. The supertransformation is equal to
$`\delta _TX_{M_p\mathrm{}M_1}=\mathrm{\Xi }^K_KX_{M_p\mathrm{}M_1}+p_{[M_p}\mathrm{\Xi }^KX_{|K|M_{p1}\mathrm{}M_1]}`$ (2.5)
for p-form’s components. The local Lorentz transformations are equal to
$`\delta _LE^A`$ $`=`$ $`E^B\mathrm{\Lambda }_B^A,`$
$`\delta _L\mathrm{\Omega }_B^A`$ $`=`$ $`\mathrm{\Lambda }_B^C\mathrm{\Omega }_C^A+\mathrm{\Omega }_B^C\mathrm{\Lambda }_C^Ad\mathrm{\Lambda }_B^A.`$ (2.6)
The supergauge transformations are equal to
$`\delta _GB_{LMN}=3_{[L}\mathrm{\Xi }_{MN]}.`$ (2.7)
We obtain the full algebra of these transformations as follows
$`[\delta _T(\mathrm{\Xi }_1)+\delta _L(\mathrm{\Lambda }_1)+\delta _G(\mathrm{\Xi }_{1MN}),\delta _T(\mathrm{\Xi }_2)+\delta _L(\mathrm{\Lambda }_2)+\delta _G(\mathrm{\Xi }_{2MN})]`$
$`=\delta _T(\mathrm{\Xi }_3)+\delta _L(\mathrm{\Lambda }_3)+\delta _G(\mathrm{\Xi }_{3MN}),`$ (2.8)
where,
$`\mathrm{\Xi }_3^K`$ $`=`$ $`\mathrm{\Xi }_2^L_L\mathrm{\Xi }_1^K+\delta _1\mathrm{\Xi }_2^K(12),`$
$`\mathrm{\Lambda }_{3A}^B`$ $`=`$ $`\mathrm{\Xi }_1^K_K\mathrm{\Lambda }_{2A}^B+\delta _1\mathrm{\Lambda }_{2A}^B+\mathrm{\Lambda }_{1A}^C\mathrm{\Lambda }_{2C}^B(12),`$
$`\mathrm{\Xi }_{3MN}`$ $`=`$ $`\delta _1\mathrm{\Xi }_{2MN}\mathrm{\Xi }_1^K_K\mathrm{\Xi }_{2MN}2_{[M}\mathrm{\Xi }_{2N]K}\mathrm{\Xi }_1^K(12).`$ (2.9)
### 2.2 Gauge completion
Firstly, we choose the input data as follows
$`E_m^{a(0)}`$ $`=`$ $`e_m^a,`$
$`E_m^{\alpha (0)}`$ $`=`$ $`\psi _m^\alpha ,`$
$`\mathrm{\Omega }_{mb}^{a(0)}`$ $`=`$ $`\widehat{\omega }_{mb}^a,`$
$`\mathrm{\Xi }^{m(0)}`$ $`=`$ $`\xi ^m,`$
$`\mathrm{\Xi }^{\mu (0)}`$ $`=`$ $`ϵ^\mu ,`$
$`\mathrm{\Xi }_{mn}^{(0)}`$ $`=`$ $`\xi _{mn},`$
$`B_{mnl}^{(0)}`$ $`=`$ $`C_{mnl}.`$ (2.10)
From (2.1), we obtain
$`\mathrm{\Lambda }_b^{a(0)}=\lambda _b^a.`$ (2.11)
Moreover we introduce the assumption that superparameters do not include the derivative of $`ϵ`$. Then, the higher order components in anticommuting coordinates can be obtained by requiring consistency between the algebra of superspace supergravity and that of ordinary supergravity.
If we can represent $`\mathrm{\Xi }_{MN}=2_{[M}\mathrm{\Phi }_{N]}`$, we can choose the gauge as $`\mathrm{\Xi }_{MN}=0`$ because this superparameters do not change the 3-form superfields (2.7) and the algebra (2.1). Thus we can choose the gauge as follows,
$`\mathrm{\Xi }_{\mu N}^{(0)}`$ $`=`$ $`0.`$ (2.12)
To obtain the higher order components of superparameters which depend on $`ϵ`$, we must calculate the commutation of two supersymmetry transformation.
According to (2.1),(2.1) and (2.2),
$`[\delta _{s1},\delta _{s2}]E_m^{a(0)}`$ $`=`$ $`(\mathrm{\Xi }_3^K_KE_m^a+_m\mathrm{\Xi }_3^KE_K^a+E_m^b\mathrm{\Lambda }_{3b}^a)|_{\theta =0}`$ (2.13)
$`=`$ $`(\delta _g(2\overline{ϵ}_2\mathrm{\Gamma }^mϵ_1)+\delta _s(2\overline{ϵ}_2\mathrm{\Gamma }^nϵ_1\psi _n)+\delta _c(2\overline{ϵ}_2\mathrm{\Gamma }^kϵ_1C_{kmn}2\overline{ϵ}_2\mathrm{\Gamma }_{mn}ϵ_1)`$
$`+\delta _l(2\overline{ϵ}_2\mathrm{\Gamma }^nϵ_1\widehat{\omega }_{nb}^a+{\displaystyle \frac{1}{72}}\overline{ϵ}_2(\mathrm{\Gamma }_b^{arstu}\widehat{F}_{rstu}+24\mathrm{\Gamma }_{rs}\widehat{F}_b^{ars})ϵ_1))e_m^a.`$
Thus one obtains
$`\mathrm{\Xi }^{k(1)}(susy)=\overline{\theta }\mathrm{\Gamma }^kϵ.`$ (2.14)
In the same way, to obtain the higher order components of superparameters which depend on $`\lambda `$ we must calculate the commutation of supersymmetry transformation and Lorentz transformation. To obtain the higher order components of superparameters which depend on $`\xi _{mn}`$ we must calculate the commutation of supersymmetry transformation and gauge transformation. To obtain the higher order components of superparameters which depend on $`\xi ^m`$ we must calculate the commutation of supersymmetry transformation and general coordinate transformation. According to superspace algebra,
$`\delta _{susy}E_m^a|_{\theta =0}`$ $`=`$ $`(\mathrm{\Xi }^K(susy)_KE_m^a+_m\mathrm{\Xi }^K(susy)E_K^a+E_m^b\mathrm{\Lambda }_b^a(susy))|_{\theta =0}`$ (2.15)
$`=`$ $`ϵ^\nu _\nu (E_m^{a(1)})+_mϵ^\nu E_\nu ^{a(0)},`$
while in ordinary supergravity
$`\delta _{susy}e_m^a=2\overline{ϵ}\mathrm{\Gamma }^a\psi _m.`$ (2.16)
Thus, one obtains
$`E_\nu ^{a(0)}`$ $`=`$ $`0,`$
$`E_m^{a(1)}`$ $`=`$ $`2\overline{\theta }\mathrm{\Gamma }^a\psi _m.`$ (2.17)
By this procedure, the following results had been known .
$`\mathrm{\Xi }^m`$ $`=`$ $`\xi ^m+\overline{\theta }\mathrm{\Gamma }^mϵ\overline{\theta }\mathrm{\Gamma }^nϵ\overline{\theta }\mathrm{\Gamma }^m\psi _n+𝒪(\theta ^3),`$ (2.18)
$`\mathrm{\Xi }^\mu `$ $`=`$ $`ϵ^\mu {\displaystyle \frac{1}{4}}\lambda _{cd}(\mathrm{\Gamma }^{cd}\theta )^\mu \overline{\theta }\mathrm{\Gamma }^nϵ\psi _n^\mu +\overline{\theta }\mathrm{\Gamma }^nϵ\overline{\theta }\mathrm{\Gamma }^k\psi _n\psi _k^\mu +{\displaystyle \frac{1}{4}}\overline{\theta }\mathrm{\Gamma }^nϵ\widehat{\omega }_{nab}(\mathrm{\Gamma }^{ab}\theta )^\mu `$ (2.19)
$`{\displaystyle \frac{1}{3}}\overline{\theta }\mathrm{\Gamma }^kϵ(T_k^{abcd}\theta )^\mu \widehat{F}_{abcd}{\displaystyle \frac{1}{864}}\overline{\theta }(\mathrm{\Gamma }_{ab}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{abcd})ϵ(\mathrm{\Gamma }^{ab}\theta )^\mu `$
$`+𝒪(\theta ^3),`$
$`\mathrm{\Lambda }_b^a`$ $`=`$ $`\lambda _b^a\overline{\theta }\mathrm{\Gamma }^nϵ\widehat{\omega }_{nb}^a+{\displaystyle \frac{1}{144}}\overline{\theta }(\mathrm{\Gamma }_b^{arstu}\widehat{F}_{rstu}+24\mathrm{\Gamma }_{rs}\widehat{F}_b^{ars})ϵ+𝒪(\theta ^2),`$ (2.20)
$`\mathrm{\Xi }_{mn}`$ $`=`$ $`\xi _{mn}(\overline{\theta }\mathrm{\Gamma }^pϵC_{pmn}+\overline{\theta }\mathrm{\Gamma }_{mn}ϵ)+\overline{\theta }\mathrm{\Gamma }^kϵ\overline{\theta }\mathrm{\Gamma }^l\psi _kC_{lmn}+\overline{\theta }\mathrm{\Gamma }^kϵ\overline{\theta }\mathrm{\Gamma }_{mn}\psi _k`$ (2.21)
$`+{\displaystyle \frac{4}{3}}\overline{\theta }\mathrm{\Gamma }^lϵ\overline{\theta }\mathrm{\Gamma }_{l[m}\psi _{n]}+{\displaystyle \frac{4}{3}}\overline{\theta }\mathrm{\Gamma }^l\psi _{[n}\overline{\theta }\mathrm{\Gamma }_{|l|m]}ϵ+𝒪(\theta ^3),`$
$`\mathrm{\Xi }_{m\mu }`$ $`=`$ $`{\displaystyle \frac{1}{6}}\overline{\theta }\mathrm{\Gamma }^nϵ(\overline{\theta }\mathrm{\Gamma }_{mn})_\mu +{\displaystyle \frac{1}{6}}(\overline{\theta }\mathrm{\Gamma }^n)_\mu \overline{\theta }\mathrm{\Gamma }_{mn}ϵ+𝒪(\theta ^3),`$ (2.22)
$`\mathrm{\Xi }_{\mu \nu }`$ $`=`$ $`𝒪(\theta ^3),`$ (2.23)
$`E_m^a`$ $`=`$ $`e_m^a+2\overline{\theta }\mathrm{\Gamma }^a\psi _m{\displaystyle \frac{1}{4}}\overline{\theta }\mathrm{\Gamma }^{acd}\theta \widehat{\omega }_{mcd}+{\displaystyle \frac{1}{72}}\overline{\theta }\mathrm{\Gamma }_m^{rst}\theta \widehat{F}_{rst}^a`$ (2.24)
$`+{\displaystyle \frac{1}{288}}\overline{\theta }\mathrm{\Gamma }^{rstu}\theta \widehat{F}_{rstu}e_m^a{\displaystyle \frac{1}{36}}\overline{\theta }\mathrm{\Gamma }^{astu}\theta \widehat{F}_{mstu}+𝒪(\theta ^3),`$
$`E_m^\alpha `$ $`=`$ $`\psi _m^\alpha {\displaystyle \frac{1}{4}}\widehat{\omega }_{mab}(\mathrm{\Gamma }^{ab}\theta )^\alpha +(T_m^{rstu}\theta )^\alpha \widehat{F}_{rstu}`$ (2.25)
$`+\overline{\theta }\mathrm{\Gamma }^k\psi _m(T_k^{abcd}\theta )^\alpha \widehat{F}_{abcd}{\displaystyle \frac{1}{576}}\overline{\theta }(\mathrm{\Gamma }_{ab}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{abcd})\psi _m(\mathrm{\Gamma }^{ab}\theta )^\alpha `$
$`12(T_k^{abcd}\theta )^\alpha \overline{\theta }\mathrm{\Gamma }_{[ab}\widehat{D}_c\psi _{d]}{\displaystyle \frac{1}{4}}(\overline{\theta }\mathrm{\Gamma }_a\widehat{D}_m\psi _b\overline{\theta }\mathrm{\Gamma }_b\widehat{D}_m\psi _a+\overline{\theta }\mathrm{\Gamma }_m\widehat{D}_a\psi _b)(\mathrm{\Gamma }^{ab}\theta )^\alpha `$
$`+𝒪(\theta ^3),`$
$`E_\mu ^a`$ $`=`$ $`(\mathrm{\Gamma }^a\theta )_\mu +𝒪(\theta ^3),`$ (2.26)
$`E_\mu ^\alpha `$ $`=`$ $`\delta _\mu ^\alpha `$ (2.27)
$`{\displaystyle \frac{1}{3}}(\mathrm{\Gamma }^k\theta )_\mu (T_k^{abcd}\theta )^\alpha \widehat{F}_{abcd}+{\displaystyle \frac{1}{1728}}((\mathrm{\Gamma }_{ab}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{abcd})\theta )_\mu (\mathrm{\Gamma }^{ab}\theta )^\alpha `$
$`+𝒪(\theta ^3),`$
$`\mathrm{\Omega }_{\mu b}^a`$ $`=`$ $`{\displaystyle \frac{1}{144}}\{(\mathrm{\Gamma }_b^{arstu}\theta )_\mu \widehat{F}_{rstu}+24(\mathrm{\Gamma }_{rs}\theta )_\mu \widehat{F}_b^{ars}\}+𝒪(\theta ^2),`$ (2.28)
$`\mathrm{\Omega }_{mab}`$ $`=`$ $`\widehat{\omega }_{mab}+2\overline{\theta }\{e_a^ne_b^k(\mathrm{\Gamma }_kD_{[m}\psi _{n]}+\mathrm{\Gamma }_nD_{[m}\psi _{k]}+\mathrm{\Gamma }_mD_{[n}\psi _{k]})\}`$ (2.29)
$`+{\displaystyle \frac{1}{72}}\overline{\theta }(\mathrm{\Gamma }_{ab}^{rstu}\widehat{F}_{rstu}+24\mathrm{\Gamma }_{rs}\widehat{F}_{ab}^{rs})\psi _m+𝒪(\theta ^2),`$
$`B_{mnl}`$ $`=`$ $`C_{mnl}6\overline{\theta }\mathrm{\Gamma }_{[mn}\psi _{l]}+{\displaystyle \frac{3}{4}}\widehat{\omega }_{[l}^{cd}\overline{\theta }\mathrm{\Gamma }_{mn]cd}\theta {\displaystyle \frac{3}{2}}\widehat{\omega }_{[lmn]}\theta ^2`$ (2.30)
$`{\displaystyle \frac{1}{96}}\overline{\theta }\mathrm{\Gamma }_{mnl}^{rstu}\theta \widehat{F}_{rstu}{\displaystyle \frac{3}{8}}\overline{\theta }\mathrm{\Gamma }_{[l}^{rs}\theta \widehat{F}_{|rs|mn]}12\overline{\theta }\mathrm{\Gamma }_a\psi _{[m}\overline{\theta }\mathrm{\Gamma }_n^a\psi _{l]}`$
$`+𝒪(\theta ^3),`$
$`B_{mn\mu }`$ $`=`$ $`(\overline{\theta }\mathrm{\Gamma }_{mn})_\mu +{\displaystyle \frac{8}{3}}\overline{\theta }\mathrm{\Gamma }^k\psi _{[m}(\overline{\theta }\mathrm{\Gamma }_{|k|n]})_\mu +{\displaystyle \frac{4}{3}}(\overline{\theta }\mathrm{\Gamma }^k)_\mu \overline{\theta }\mathrm{\Gamma }_{k[m}\psi _{n]}+𝒪(\theta ^3),`$ (2.31)
$`B_{m\mu \nu }`$ $`=`$ $`(\overline{\theta }\mathrm{\Gamma }_{mn})_{(\mu }(\overline{\theta }\mathrm{\Gamma }^n)_{\nu )}+𝒪(\theta ^3),`$ (2.32)
$`B_{\mu \nu \rho }`$ $`=`$ $`(\overline{\theta }\mathrm{\Gamma }_{mn})_{(\mu }(\overline{\theta }\mathrm{\Gamma }^m)_\nu (\overline{\theta }\mathrm{\Gamma }^n)_{\rho )}+𝒪(\theta ^3).`$ (2.33)
Because the flat geometry had been known, we include the $`\theta ^3`$ term in $`B_{\mu \nu \rho }`$ for completeness.
Up to first order in anticommuting coordinates, the superfield components was investigated by E. Cremmer and S. Ferrara . $`\mathrm{\Xi }^{k(2)},E_M^{k(2)},\mathrm{\Xi }_{MN}^{(2)},B_{LMN}^{(2)}`$ was investigated by B. de Wit, K. Peeters and J. Plefka . $`E_M^{\alpha (2)},\mathrm{\Xi }^{\mu (2)}`$ was investigated in ref. .
## 3 Computation
$`\mathrm{\Lambda }_{ab}^{(2)}`$ is subject to the following equations,
$`ϵ_1^\nu _\mu _\nu \mathrm{\Lambda }_{2ab}^{(2)}(12)`$ $`=`$ $`\overline{ϵ_2}\mathrm{\Gamma }^nϵ_1(\mathrm{\Gamma }^k\psi _n)_\mu \widehat{\omega }_{kba}(\mathrm{\Gamma }^nϵ_1)_\mu \overline{\psi _n}\mathrm{\Gamma }^kϵ_2\widehat{\omega }_{kba}`$ (3.1)
$`{\displaystyle \frac{1}{144}}\overline{ϵ_2}\mathrm{\Gamma }^nϵ_1\{(\mathrm{\Gamma }_{ba}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{bacd})\psi _n\}_\mu `$
$`+{\displaystyle \frac{1}{144}}(\mathrm{\Gamma }^nϵ_1)_\mu \overline{\psi _n}(\mathrm{\Gamma }_{ba}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{bacd})ϵ_2`$
$`+{\displaystyle \frac{1}{6}}(\mathrm{\Gamma }_{ba}^{cdef}ϵ_2)_\mu \overline{ϵ_1}\mathrm{\Gamma }_{cd}\widehat{D}_e\psi _f+4(\mathrm{\Gamma }^{cd}ϵ_2)_\mu \overline{ϵ_1}\mathrm{\Gamma }_{[ba}\widehat{D}_c\psi _{d]}`$
$`+(\mathrm{\Gamma }^kϵ_2)_\mu 2\overline{ϵ_1}(\mathrm{\Gamma }_a\widehat{D}_{[k}\psi _{b]}+\mathrm{\Gamma }_k\widehat{D}_{[b}\psi _{a]}+\mathrm{\Gamma }_b\widehat{D}_{[k}\psi _{a]})`$
$`(12).`$
However, if simply we drive the equation,
$`ϵ_1^\nu _\mu _\nu \mathrm{\Lambda }_{2ab}^{(2)}`$ $`=`$ $`(\overline{ϵ_2}\mathrm{\Gamma }^n)_\nu ϵ_1^\nu (\mathrm{\Gamma }^k\psi _n)_\mu \widehat{\omega }_{kba}(\mathrm{\Gamma }^n)_{\mu \nu }ϵ_1^\nu \overline{\psi _n}\mathrm{\Gamma }^kϵ_2\widehat{\omega }_{kba}`$ (3.2)
$`{\displaystyle \frac{1}{144}}(\overline{ϵ_2}\mathrm{\Gamma }^n)_\nu ϵ_1^\nu \{(\mathrm{\Gamma }_{ba}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{bacd})\psi _n\}_\mu `$
$`+{\displaystyle \frac{1}{144}}(\mathrm{\Gamma }^n)_{\mu \nu }ϵ_1^\nu \overline{\psi _n}(\mathrm{\Gamma }_{ba}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{bacd})ϵ_2`$
$`+{\displaystyle \frac{1}{6}}(\mathrm{\Gamma }_{ba}^{cdef}ϵ_2)_\mu ϵ_1^\nu (\mathrm{\Gamma }_{cd}\widehat{D}_e\psi _f)_\nu +4(\mathrm{\Gamma }^{cd}ϵ_2)_\mu ϵ_1^\nu (\mathrm{\Gamma }_{[ba}\widehat{D}_c\psi _{d]})_\nu `$
$`+(\mathrm{\Gamma }^kϵ_2)_\mu 2ϵ_1^\nu (\mathrm{\Gamma }_a\widehat{D}_{[k}\psi _{b]}+\mathrm{\Gamma }_k\widehat{D}_{[b}\psi _{a]}+\mathrm{\Gamma }_b\widehat{D}_{[k}\psi _{a]})_\nu ,`$
this equation is inconsistent because $`\mu `$ and $`\nu `$ in the left-hand side of it are antisymmetric but these in the right-hand side of it are not antisymmetric. Thus we must add terms which are symmetric under interchanging the indices 1 and 2 in the right-hand side of this equation.
$`ϵ_1^\nu _\mu _\nu \mathrm{\Lambda }_{2ab}^{(2)}`$ $`=`$ $`\overline{ϵ_2}\mathrm{\Gamma }^nϵ_1(\mathrm{\Gamma }^k\psi _n)_\mu \widehat{\omega }_{kba}(\mathrm{\Gamma }^nϵ_1)_\mu \overline{\psi _n}\mathrm{\Gamma }^kϵ_2\widehat{\omega }_{kba}`$ (3.3)
$`{\displaystyle \frac{1}{144}}\overline{ϵ_2}\mathrm{\Gamma }^nϵ_1\{(\mathrm{\Gamma }_{ba}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{bacd})\psi _n\}_\mu `$
$`+{\displaystyle \frac{1}{144}}(\mathrm{\Gamma }^nϵ_1)_\mu \overline{\psi _n}(\mathrm{\Gamma }_{ba}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{bacd})ϵ_2`$
$`+{\displaystyle \frac{1}{6}}(\mathrm{\Gamma }_{ba}^{cdef}ϵ_2)_\mu \overline{ϵ_1}\mathrm{\Gamma }_{cd}\widehat{D}_e\psi _f+4(\mathrm{\Gamma }^{cd}ϵ_2)_\mu \overline{ϵ_1}\mathrm{\Gamma }_{[ba}\widehat{D}_c\psi _{d]}`$
$`+(\mathrm{\Gamma }^kϵ_2)_\mu 2\overline{ϵ_1}(\mathrm{\Gamma }_a\widehat{D}_{[k}\psi _{b]}+\mathrm{\Gamma }_k\widehat{D}_{[b}\psi _{a]}+\mathrm{\Gamma }_b\widehat{D}_{[k}\psi _{a]})+\overline{ϵ_2}ϵ_1{\displaystyle \frac{438}{7}}(\widehat{D}_{[b}\psi _{a]})_\mu `$
$`+\overline{ϵ_2}\mathrm{\Gamma }^{ijk}ϵ_1\{{\displaystyle \frac{1}{28}}(\mathrm{\Gamma }_{bkj}\widehat{D}_{[a}\psi _{i]}(ab))_\mu {\displaystyle \frac{31}{42}}(\mathrm{\Gamma }_{kji}\widehat{D}_{[b}\psi _{a]})_\mu \}`$
$`+\overline{ϵ_2}\mathrm{\Gamma }^{ijkl}ϵ_1\{{\displaystyle \frac{1}{6}}(\mathrm{\Gamma }_{balk}\widehat{D}_j\psi _i)_\mu +{\displaystyle \frac{1}{21}}(\mathrm{\Gamma }_{blkj}\widehat{D}_{[a}\psi _{i]}(ab))_\mu `$
$`{\displaystyle \frac{1}{6}}(\mathrm{\Gamma }_{lkji}\widehat{D}_{[b}\psi _{a]})_\mu {\displaystyle \frac{17}{84}}(\delta _{la}\mathrm{\Gamma }_{kj}\widehat{D}_{[b}\psi _{i]}(ab))_\mu `$
$`+{\displaystyle \frac{1}{3}}(\delta _{la}\delta _{kb}(ab))(\widehat{D}_j\psi _i)_\mu \}.`$
Thus we obtain
$`\mathrm{\Lambda }_{ab}^{(2)}`$ $`=`$ $`\overline{\theta }\mathrm{\Gamma }^nϵ\overline{\theta }\mathrm{\Gamma }^k\psi _n\widehat{\omega }_{kba}{\displaystyle \frac{1}{144}}\overline{\theta }\mathrm{\Gamma }^nϵ\overline{\theta }(\mathrm{\Gamma }_{ba}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{bacd})\psi _n`$ (3.4)
$`+{\displaystyle \frac{1}{64}}[\overline{\theta }\theta ({\displaystyle \frac{14016}{7}}\overline{ϵ}\widehat{D}_{[b}\psi _{a]})`$
$`+{\displaystyle \frac{1}{6}}\overline{\theta }\mathrm{\Gamma }^{xyz}\theta \{{\displaystyle \frac{48}{7}}\overline{ϵ}(\mathrm{\Gamma }_{bzy}\widehat{D}_{[a}\psi _{x]}(ab))+{\displaystyle \frac{992}{7}}\overline{ϵ}(\mathrm{\Gamma }_{zyx}\widehat{D}_{[b}\psi _{a]})\}`$
$`+{\displaystyle \frac{1}{24}}\overline{\theta }\mathrm{\Gamma }^{wxyz}\theta \{128\overline{ϵ}\mathrm{\Gamma }_{bazy}\widehat{D}_x\psi _w{\displaystyle \frac{256}{7}}\overline{ϵ}(\mathrm{\Gamma }_{bzyx}\widehat{D}_{[a}\psi _{w]}(ab))`$
$`+128\overline{ϵ}\mathrm{\Gamma }_{zyxw}\widehat{D}_{[b}\psi _{a]}+{\displaystyle \frac{1088}{7}}\overline{ϵ}(\delta _{za}\mathrm{\Gamma }_{yx}\widehat{D}_{[b}\psi _{w]}(ab))`$
$`256\overline{ϵ}(\delta _{za}\delta _{yb}\widehat{D}_x\psi _w(ab))\}].`$
$`\mathrm{\Omega }_{\mu ab}^{(2)}`$ is subject to the following equation,
$`ϵ^\nu _\nu \mathrm{\Omega }_{\mu ab}^{(2)}`$ $`=`$ $`{\displaystyle \frac{1}{144}}\overline{\theta }\mathrm{\Gamma }^nϵ\overline{\psi _n}(\mathrm{\Gamma }_{ba}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{bacd})_\mu \overline{\theta }\mathrm{\Gamma }^nϵ(\mathrm{\Gamma }^k\psi _n)_\mu \widehat{\omega }_{kba}`$ (3.5)
$`(\mathrm{\Gamma }^nϵ)_\mu \overline{\theta }\mathrm{\Gamma }^k\psi _n\widehat{\omega }_{kba}(\mathrm{\Gamma }^nϵ)_\mu 2\overline{\theta }(\mathrm{\Gamma }_b\widehat{D}_{[a}\psi _{n]}\mathrm{\Gamma }_a\widehat{D}_{[b}\psi _{n]}\mathrm{\Gamma }_n\widehat{D}_{[b}\psi _{a]})`$
$`{\displaystyle \frac{1}{144}}(\mathrm{\Gamma }^nϵ)_\mu \overline{\theta }(\mathrm{\Gamma }_{ab}^{cdef}\widehat{F}_{cdef}+24\mathrm{\Gamma }^{cd}\widehat{F}_{abcd})\psi _n+_\mu \mathrm{\Lambda }_{ab}^{(2)}.`$
Thus we obtain
$`\mathrm{\Omega }_{\mu ab}^{(2)}`$ $`=`$ $`{\displaystyle \frac{1}{64}}[\overline{\theta }\theta {\displaystyle \frac{14212}{7}}(\widehat{D}_{[b}\psi _{a]})_\mu `$ (3.6)
$`+{\displaystyle \frac{1}{6}}\overline{\theta }\mathrm{\Gamma }^{xyz}\theta \{{\displaystyle \frac{120}{7}}(\mathrm{\Gamma }_{bzy}\widehat{D}_{[a}\psi _{x]}(ab))_\mu {\displaystyle \frac{1020}{7}}(\mathrm{\Gamma }_{zyx}\widehat{D}_{[b}\psi _{a]})_\mu `$
$`192(\delta _{za}\mathrm{\Gamma }_b\widehat{D}_y\psi _x(ab))_\mu 48(\delta _{za}\mathrm{\Gamma }_y\widehat{D}_{[b}\psi _{x]}(ab))_\mu \}`$
$`+{\displaystyle \frac{1}{24}}\overline{\theta }\mathrm{\Gamma }^{wxyz}\theta \{{\displaystyle \frac{32}{7}}(\mathrm{\Gamma }_{bzyx}\widehat{D}_{[a}\psi _{w]}(ab))_\mu 132(\mathrm{\Gamma }_{zyxw}\widehat{D}_{[b}\psi _{a]})_\mu `$
$`256(\delta _{za}\mathrm{\Gamma }_{by}\widehat{D}_x\psi _w(ab))_\mu +{\displaystyle \frac{32}{7}}(\delta _{za}\mathrm{\Gamma }_{yx}\widehat{D}_{[a}\psi _{w]}(ab))_\mu \}].`$
## 4 Discussion
We have obtained $`\mathrm{\Lambda }_{ab}^{(2)},\mathrm{\Omega }_{\mu a}^{b(2)}`$. Up to second order in anticommuting coordinates, only $`\mathrm{\Omega }_{ma}^{b(2)}`$ remains. This component is complicated. However, it contains $`\delta _{susy}\widehat{D}_{[m}\psi _{n]}`$ thus it is expected to contain curvature terms. From Bianchi identity, curvature terms should appeared in vielbein superfields at third and the higher order in anticommuting coordinates. This gives interaction terms coupled to curvature in Matrix model and higher curvature corrections to Einstein gravity in low energy effective actions. Thus it is important that we investigate $`\mathrm{\Omega }_{ma}^{b(2)}`$ . These terms and terms which are required to obtain terms of Matrix theory which are third order in anticommuting coordinates is under considerations.
However, $`\mathrm{\Omega }_{\mu a}^{b(2)}`$ is also important in study of superspace structure and curved membrane action. $`\kappa `$-symmetry constraints act on torsion fields and curvature fields. Torsion are defined as $`T^A=DE^A=dE^A+E^B\mathrm{\Omega }_B^A`$. Thus $`\mathrm{\Omega }_{\mu a}^{b(2)}`$ has information about more higher components of vielbein than $`\mathrm{\Omega }_{ma}^{b(2)}`$. Moreover, $`\mathrm{\Omega }_{\mu a}^{b(2)}`$ contains $`\widehat{D}_{[a}\psi _{b]}`$ which is nonlinearly exact supersymmetry field strength which information is important in study of supersymmetry.
## Acknowledgments
I would like to thank Y.Matsuo for valuable suggestions.
## Appendix
## Appendix A Conventions
### A.1 Indices
We use Greek indices for spinorial components and Latin indices for vector components. And we use former alphabet for the tangent space indices and later for general coordinates indices: $`a,b,c,\mathrm{}`$ for tangent vector indices and $`k,l,m,\mathrm{}`$ for general vector indices, and $`\alpha ,\beta ,\mathrm{}`$ for tangent spinorial indices and $`\mu ,\nu ,\mathrm{}`$ for general spinorial indices.
Superspace coordinates $`(x^m,\theta ^\mu )`$ are designated $`Z^M`$ , where later capital Latin alphabet $`M,N,..`$ are collective designations for general coordinate indices. While former capital Latin alphabet $`A,B,..`$ are collective designations for tangent space indices.
### A.2 p-form superfield
We introduce p-form superfields as follows,
$`X`$ $``$ $`{\displaystyle \frac{1}{p!}}dz^{M_p}\mathrm{}dz^{M_1}X_{M_p\mathrm{}M_1}`$ (A.1)
$``$ $`{\displaystyle \frac{1}{p!}}E^{A_p}\mathrm{}E^{A_1}X_{A_p\mathrm{}A_1},`$
$`X_{A_p\mathrm{}A_1}`$ $``$ $`{\displaystyle \underset{i=1}{\overset{32}{}}}X_{A_p\mathrm{}A_1}^{(i)}.`$ (A.2)
$`X_{A_p\mathrm{}A_1}^{(i)}`$ is component at i-th order in anticommuting coordinates.
### A.3 Brackets
Symmetrization bracket $`()`$ and antisymmetrization bracket $`[]`$ is defined as follows,
$`[M_1\mathrm{}M_N]`$ $`=`$ $`{\displaystyle \frac{1}{N!}}(M_1\mathrm{}M_N+\text{antisymmetric terms}),`$
$`(M_1\mathrm{}M_N)`$ $`=`$ $`{\displaystyle \frac{1}{N!}}(M_1\mathrm{}M_N+\text{symmetric terms}).`$ (A.3)
### A.4 Gamma matrices(11-dimensional)
Since we use the Majorana representation, all components are real.
Gamma matrix $`\mathrm{\Gamma }_\beta ^{a\alpha }`$ is defined as follows,
$`\{\mathrm{\Gamma }^a,\mathrm{\Gamma }^b\}=2\eta ^{ab}.`$ (A.4)
We use the mostly plus metric; $`\eta _{ab}(+\mathrm{}+)`$. We lower the spinorial indices by charge conjugation matrix $`C_{\alpha \beta }`$.
$`\overline{\psi }_\beta =\psi ^\alpha C_{\alpha \beta },`$
$`\mathrm{\Gamma }_{\alpha \beta }^a=C_{\alpha \gamma }\mathrm{\Gamma }_\beta ^{a\gamma }.`$ (A.5)
$`\mathrm{\Gamma }_{\alpha \beta }^{a_1..a_n}(n=1,2,5,6,9,10)`$ are symmetric matrices and $`\mathrm{\Gamma }_{\alpha \beta }^{a_1..a_n}(n=0,3,4,7,8,11)`$ are antisymmetric matrices.
|
warning/0005/math0005016.html
|
ar5iv
|
text
|
# On some aspects of the Deligne-Simpson problem 11footnote 1Research partially supported by INTAS grant 97-1644
## 1 Introduction
### 1.1 Formulation of the problem
In the present paper we consider the multiplicative version of the Deligne-Simpson problem:
Give necessary and sufficient conditions for the choice of the conjugacy classes $`C_jSL(n,𝐂)`$ so that there exist irreducible $`(p+1)`$-tuples of matrices $`M_jC_j`$ such that
$$M_1\mathrm{}M_{p+1}=I$$
(1)
In the additive version the conjugacy classes $`c_j`$ belong to $`sl(n,𝐂)`$ and the matrices $`A_jc_j`$ satisfy the condition
$$A_1+\mathrm{}+A_{p+1}=0$$
(2)
The matrices $`A_j`$ and $`M_j`$ are interpreted respectively as matrices-residua and as monodromy operators of fuchsian linear systems, see the next section. Both versions of the problem were considered in \[Ko1\] which is the first part of this paper.
The problem (in the multiplicative version) was stated by P.Deligne and C.Simpson obtained the first results towards its solution, see \[Si\].
Further in the text we consider sometimes $`C_j`$ ($`c_j`$) as conjugacy classes from $`GL(n,𝐂)`$ (from $`gl(n,𝐂)`$) instead of $`SL(n,𝐂)`$ (instead of $`sl(n,𝐂)`$) because when solving the problem there appear such matrices. The passage from the problem for $`M_jSL(n,𝐂)`$ (or $`A_jsl(n,𝐂)`$) to the one for $`M_jGL(n,𝐂)`$ (or $`A_jgl(n,𝐂)`$) and vice versa is trivial.
We presume that there hold the necessary conditions $`_{j=1}^{p+1}`$Tr$`(c_j)=0`$ and $`_{j=1}^{p+1}det(C_j)=1`$. This means that the eigenvalues $`\lambda _{k,j}`$ ($`\sigma _{k,j}`$) of the matrices $`A_j`$ ($`M_j`$) satisfy the conditions
$$\underset{j=1}{\overset{p+1}{}}\underset{k=1}{\overset{n}{}}\lambda _{k,j}=0,\underset{j=1}{\overset{p+1}{}}\underset{k=1}{\overset{n}{}}\sigma _{k,j}=1$$
Here $`k=1,\mathrm{},n`$ and for $`j`$ fixed the eigenvalues are not presumed distinct. By definition, the multiplicity of the eigenvalue $`\lambda _{k,j}`$ (or $`\sigma _{k,j}`$) is the number of eigenvalues $`\lambda _{i,j}`$ (or $`\sigma _{i,j}`$; $`j`$ is fixed) equal to it including the eigenvalue itself.
Define as generic any set of eigenvalues $`\lambda _{k,j}`$ or $`\sigma _{k,j}`$ which satisfy none of the equalities
$$\begin{array}{cccccccc}\hfill _{j=1}^{p+1}_{k\mathrm{\Phi }_j}\lambda _{k,j}& =\hfill & 0\hfill & ,& \hfill _{j=1}^{p+1}_{k\mathrm{\Phi }_j}\sigma _{k,j}& =\hfill & 1\hfill & (\gamma )\hfill \end{array}$$
where the sets $`\mathrm{\Phi }_j`$ contain one and the same number $`\kappa `$ of indices ($`1<\kappa <n`$) for all $`j`$. Reducible $`(p+1)`$-tuples exist only for non-generic eigenvalues – if a $`(p+1)`$-tuple is block upper-triangular, then the eigenvalues of each diagonal block satisfy some relation ($`\gamma `$).
Definition. Call Problem (I) the Deligne-Simpson problem like it is formulated above and Problem (TC) the same problem in which the requirement of irreducibility is replaced by the requirement the centralizer of the $`(p+1)`$-tuple to be trivial.
It is clear that Problem (TC) is weaker than Problem (I) and that for generic eigenvalues the answers to both problems coincide. Part of the results from this paper concern Problem (TC).
### 1.2 The quantities $`q`$, $`d`$, $`\xi `$ and $`m_0`$
Definition. A multiplicity vector (MV) is a vector whose components are non-negative integers whose sum equals $`n`$. We always interpret a MV as the vector of the multiplicities of the eigenvalues of a matrix $`M_j`$ or $`A_j`$. A polymultiplicity vector (PMV) is a $`(p+1)`$-tuple or $`(p+2)`$-tuple of MVs (depending on whether we deal with a $`(p+1)`$\- or $`(p+2)`$-tuple of matrices $`A_j`$ or $`M_j`$). A PMV is called simple (resp. non-simple) if the greatest common divisor $`q`$ of all the components of all its MVs equals 1 (if not).
Notation. Denote by $`\mathrm{\Sigma }_{j,m}(\sigma )`$ the number of Jordan blocks of size $`m`$, of a given matrix $`M_j`$ and corresponding to its eigenvalue $`\sigma `$. Denote by $`d`$ the greatest common divisor of all numbers $`\mathrm{\Sigma }_{j,m}(\sigma )`$ (over all $`\sigma `$, $`j`$ and $`m`$).
Remark: The condition $`q=1`$ implies $`d=1`$, and $`d>1`$ implies $`q>1`$ but the inverse implications are not true. It is true that $`d`$ divides $`q`$ and that $`q`$ divides $`n`$.
In \[Ko1\] we considered the case of generic eigenvalues: the additive version was considered completely, in the multiplicative one we considered only the situation when $`d=1`$. In the present second part of \[Ko1\] we consider the multiplicative version for generic eigenvalues and $`d>1`$ (we call it the case $`d>1`$).
Notation. Let $`q>1`$. Denote by $`\xi `$ the product of the eigenvalues $`\sigma _{k,j}`$ when their multiplicities are reduced $`q`$ times. Hence, $`\xi `$ is a root of unity of order $`q`$. Set $`\xi =\mathrm{exp}(2\pi im_0/q)`$. If $`(m_0,q)=1`$, then $`\xi `$ is a primitive root. Another equivalent definition of $`m_0`$ is given in 5) of the remarks in Subsection 2.1.
Example: If $`p=n=2`$ and if each of the three Jordan normal forms consists of one Jordan block of size 2, then there are two possibilities – either $`\sigma _{1,1}\sigma _{1,2}\sigma _{1,3}=1`$ or $`\sigma _{1,1}\sigma _{1,2}\sigma _{1,3}=1`$ (because $`\sigma _{1,j}=\sigma _{2,j}`$ and $`(\sigma _{1,1}\sigma _{1,2}\sigma _{1,3})^2=1`$). In the first (in the second) case there are no (there are) irreducible triples of matrices $`M_j`$ satisfying (1). In the first (in the second) case the eigenvalues are not (are) generic. In both cases one has $`q=2`$, $`d=1`$. In the first case $`\xi =1`$, in the second case $`\xi =1`$.
Remark: If $`q>1`$ and if $`\xi `$ is not a primitive root of unity of order $`q`$, then the eigenvalues $`\sigma _{k,j}`$ are not generic. However, $`\xi `$ can be a primitive root of unity and the eigenvalues can be non-generic.
Example: Let $`n=4`$, $`p=3`$ and let all Jordan normal forms be diagonal. Let for all $`j`$ one have $`\sigma _{1,j}=\sigma _{2,j}`$ $`\sigma _{3,j}=\sigma _{4,j}`$. Hence, $`d=q=2`$. Let $`\sigma _{1,1}=\sigma _{1,2}=\sigma _{1,3}=1`$, $`\sigma _{1,4}=i`$, $`\sigma _{3,1}=\sigma _{3,2}=\mathrm{exp}(\pi i/4)`$, $`\sigma _{3,3}=\mathrm{exp}(\pi i/8)`$, $`\sigma _{3,4}=\mathrm{exp}(\pi i/8)`$. Hence, $`\xi =1`$. One has $`\sigma _{1,1}\sigma _{1,2}\sigma _{3,3}\sigma _{3,4}=1`$ which is a non-genericity relation.
In the additive version, when $`q>1`$, the eigenvalues $`\lambda _{k,j}`$ satisfy a non-genericity relation – if their multiplicities are reduced $`q`$ times, then their sum is 0. Denote this relation by $`(\gamma _1)`$.
In the multiplicative version, when $`\xi `$ is not a primitive root of unity, the eigenvalues $`\sigma _{k,j}`$ satisfy a non-genericity relation $`(\gamma ^{})`$ – when their multiplicities are reduced $`(m_0,q)`$ times, then their product equals 1. The multiples of $`(\gamma ^{})`$ are obtained (by definition) when after the reduction $`(m_0,q)`$ times the multiplicities are increased $`s`$ times with $`1s<(m_0,q)`$. The multiples are denoted by $`s(\gamma ^{})`$. (The multiples of $`(\gamma _1)`$ are defined by analogy.)
Definition. If $`q>1`$, if in the multiplicative version $`(m_0,q)>1`$ and if the only non-genericity relation satisfied by the eigenvalues $`\lambda _{k,j}`$ (resp. $`\sigma _{k,j}`$) is $`(\gamma _1)`$ (resp. $`(\gamma ^{})`$) and its multiples, then the eigenvalues are called relatively generic.
### 1.3 The results obtained up to now
Definition. Call Jordan normal form of size $`n`$ a family $`J^n=\{b_{i,l}\}`$ ($`iI_l`$, $`I_l=\{1,\mathrm{},s_l\}`$, $`lL`$) of positive integers $`b_{i,l}`$ whose sum is $`n`$. Here $`L`$ is the set of eigenvalues (all distinct) and $`I_l`$ is the set of Jordan blocks with eigenvalue $`l`$, $`b_{i,l}`$ is the size of the $`i`$-th block with this eigenvalue. We presume that the following inequalities hold (for each $`l`$ fixed):
$$b_{1,l}b_{2,l}\mathrm{}b_{s_l,l}$$
An $`n\times n`$-matrix $`Y`$ has the Jordan normal form $`J^n`$ (notation: $`J(Y)=J^n`$) if to its distinct eigenvalues $`\lambda _l`$, $`lL`$, there belong Jordan blocks of sizes $`b_{i,l}`$.
For a conjugacy class $`C`$ in $`GL(n,𝐂)`$ or $`gl(n,𝐂)`$ denote by $`d(C)`$ its dimension and for a matrix $`Y`$ from $`C`$ set $`r(C):=\mathrm{min}_{\lambda 𝐂}\mathrm{rk}(Y\lambda I)`$. The integer $`nr(C)`$ is the maximal number of Jordan blocks of $`J(Y)`$ with one and the same eigenvalue. Set $`d_j:=d(C_j)`$ (resp. $`d(c_j)`$), $`r_j:=r(C_j)`$ (resp. $`r(c_j)`$). The quantities $`r(C)`$ and $`d(C)`$ depend only on the Jordan normal form $`J(Y)=J^n`$, not on the eigenvalues, so we write sometimes $`r(J^n)`$ and $`d(J^n)`$.
The following two conditions are necessary for the existence of irreducible $`(p+1)`$-tuples of matrices $`M_j`$ satisfying (1) (proved in \[Si\]):
$$\begin{array}{cccccc}\hfill d_1+\mathrm{}+d_{p+1}& \hfill & 2n^22\hfill & & & (\alpha _n)\hfill \\ \hfill \mathrm{for}\mathrm{all}j(r_1+\mathrm{}+\widehat{r}_j+\mathrm{}+r_{p+1})& \hfill & n\hfill & & & (\beta _n)\hfill \end{array}$$
These conditions are also necessary for the existence of irreducible $`(p+1)`$-tuples of matrices $`A_j`$ satisfying (2), see \[Ko1\].
Recall the basic result from \[Ko1\]. Its formulation depends not on the conjugacy classes $`C_j`$ but only on the Jordan normal forms defined by them (as long as the eigenvalues remain generic).
For a given $`(p+1)`$-tuple $`J^n`$ of Jordan normal forms $`J_j^n`$ (the upper index indicates the size of the matrices) define the $`(p+1)`$-tuples of Jordan normal forms $`J^{n_\nu }`$, $`\nu =0`$, $`\mathrm{}`$, $`s`$ as follows. Set $`n_0=n`$. If $`J^n`$ satisfies the condition
$$\begin{array}{cccccc}\hfill r_1+\mathrm{}+r_{p+1}& \hfill & 2n\hfill & & & (\omega _n)\hfill \end{array}$$
or if it doesn’t satisfy condition $`(\beta _n)`$ or if $`n=1`$, then set $`s=0`$. If not, then set
$$n_1=r_1+\mathrm{}+r_{p+1}n(\mathrm{hence},n_1<n).$$
For $`j=1,\mathrm{},p+1`$ the Jordan normal form $`J_j^{n_1}`$ is obtained from $`J_j^n`$ by finding (one of) the eigenvalue(s) of $`J_j^n`$ with greatest number of Jordan blocks and by decreasing by 1 the sizes of the $`nn_1`$ smallest Jordan blocks with this eigenvalue. (Their number is $`nr_j`$ which is $`nn_1`$ because if $`J^n`$ satisfies condition $`(\beta _n)`$, then $`n_1r_j`$.) Denote symbolically the construction of the Jordan normal forms $`J_j^{n_1}`$ by
$$\mathrm{\Psi }:\{\begin{array}{ccc}n& & n_1\\ (J_1^n,\mathrm{},J_{p+1}^n)& & (J_1^{n_1},\mathrm{},J_{p+1}^{n_1})\end{array}$$
Let the $`(p+1)`$-tuples of Jordan normal forms $`J^{n_0}`$, $`\mathrm{}`$, $`J^{n_{\nu _0}}`$ be constructed. If $`J^{n_{\nu _0}}`$ satisfies condition $`(\omega _{n_{\nu _0}})`$ or if it doesn’t satisfy condition $`(\beta _{n_{\nu _0}})`$ or if $`n_{\nu _0}=1`$, then set $`s=n_{\nu _0}`$. If not, then define $`J^{n_{\nu _0+1}}`$ after $`J^{n_{\nu _0}}`$ in the same way as $`J^{n_1}`$ was defined after $`J^n`$, see $`\mathrm{\Psi }`$. One has $`n=n_0>n_1>\mathrm{}>n_s`$.
In the additive version and in the case $`d=1`$ of the multiplicative one the following theorem is true (see \[Ko1\]):
###### Theorem 1
For given conjugacy classes $`c_j`$ or $`C_j`$ with Jordan normal forms $`J_j^n`$ with generic eigenvalues there exist irreducible $`(p+1)`$-tuples of matrices $`A_j`$ or $`M_j`$ with Jordan normal forms $`J_j^n`$ satisfying respectively (2) or (1) if and only if the following two conditions hold:
i) The $`(p+1)`$-tuple of Jordan normal forms $`J_j^n`$ satisfies the inequalities $`(\alpha _n)`$ and $`(\beta _n)`$;
ii) Either the $`(p+1)`$-tuple of Jordan normal forms $`J_j^{n_s}`$ satisfies the inequality $`(\omega _{n_s})`$, or one has $`n_s=1`$.
Remarks: 1) Set $`d_j^\nu =d(J_j^{n_\nu })`$, $`d_j^0=d_j=d(J_j^n)`$. Set $`d_1+\mathrm{}+d_{p+1}=`$$`2n^22+\kappa `$, $`\kappa 𝐍`$. The number $`n`$ is divisible by $`d`$ and the numbers $`d_j`$ are divisible by $`d^2`$. Hence, if $`d>1`$ and if condition $`(\alpha _n)`$ holds, then one must have $`\kappa 2`$.
The quantity $`2\kappa `$ is called index of rigidity (see \[Ka\]). Irreducible representations with $`\kappa =0`$ are called rigid; they are unique up to conjugacy, see \[Ka\] and \[Si\]. In the last remark of this section we explain how to give the exhaustive list of $`(p+1)`$-tuples of Jordan normal forms defining irreducible representations of a given index of rigidity and with generic eigenvalues.
2) On the other hand, one has (for all $`\nu `$) $`d_1^\nu +\mathrm{}+d_{p+1}^\nu =`$ $`2(n_\nu )^22+\kappa `$. Indeed, for all $`\nu `$ one has $`_{j=1}^{p+1}r(J_j^{n_\nu })=n_\nu +n_{\nu +1}`$ by definition. One has as well $`d_j^{\nu +1}=d_j^\nu 2(n_\nu n_{\nu +1})r(J_j^{n_\nu })`$ (to be checked directly). Hence,
$$\underset{j=1}{\overset{p+1}{}}d_j^{\nu +1}=\underset{j=1}{\overset{p+1}{}}d_j^\nu 2(n_\nu n_{\nu +1})\underset{j=1}{\overset{p+1}{}}r(J_j^{n_\nu })=$$
$$=2(n_\nu )^22+\kappa 2(n_\nu n_{\nu +1})(n_\nu +n_{\nu +1})=2(n_{\nu +1})^22+\kappa $$
Thus in the case $`d>1`$ one can never have $`n_s=1`$ because for $`n_s=1`$ one has $`\kappa =0`$.
3) It is shown in \[Ko2\] that if the $`(p+1)`$-tuple of Jordan normal forms $`J_j^n`$ satisfies condition $`(\omega _n)`$, then one has $`d_1+\mathrm{}+d_{p+2}2n^2`$. Hence, if the Jordan normal forms $`J_j^{n_\nu }`$ satisfy condition $`(\omega _{n_\nu })`$, then $`d_1^\nu +\mathrm{}+d_{p+1}^\nu 2(n_\nu )^2`$. This together with 1) and 2) implies that $`n_s=1`$ if and only if $`(\alpha _n)`$ is an equality.
Definition. Call a $`(p+1)`$-tuple $`J^n`$ of Jordan normal forms $`J_j^n`$ good if it satisfies conditions i) and ii) of the theorem.
###### Corollary 2
Let $`n>1`$. Then the $`(p+1)`$-tuple $`J^n`$ of Jordan normal forms $`J_j^n`$ is good when either it satisfies condition $`(\omega _n)`$ or when the $`(p+1)`$-tuple of Jordan normal forms $`J_j^{n_1}`$ is good (and only in these two cases).
This follows from the definition of the Jordan normal forms $`J_j^{n_\nu }`$. Indeed, if $`J^n`$ is good, then it either satisfies condition $`(\omega _n)`$ or the $`(p+1)`$-tuple $`J^{n_s}`$ satisfies condition $`(\omega _{n_s})`$ (recall that if $`d>1`$, the possibility $`n_s=1`$ is excluded). In the second case one proves by induction on $`k=s\nu `$ that all $`(p+1)`$-tuples $`J^{n_\nu }`$ are good.
Theorem 1 can be reformulated as follows:
In the additive version and in the case $`d=1`$ of the multiplicative one there exist for generic eigenvalues irreducible $`(p+1)`$-tuples of matrices $`A_j`$ or $`M_j`$ satisfying respectively condition (2) or (1) if and only if the $`(p+1)`$-tuple of Jordan normal forms $`J_j^n`$ is good.
### 1.4 The basic results of this paper and what still remains to be done
The aim of the present paper is to prove
###### Theorem 3
Conditions i) and ii) from Theorem 1 are necessary and sufficient for the existence of irreducible $`(p+1)`$-tuples of matrices $`M_j`$ satisfying (1) and with generic eigenvalues in the case $`d>1`$.
The sufficiency of conditions i) and ii) from Theorem 1 follows from
###### Theorem 4
Let $`d>1`$ and let $`\xi `$ be a primitive root of unity. Then conditions i) and ii) from Theorem 1 are sufficient for the existence of $`(p+1)`$-tuples of matrices $`M_j`$ with trivial centralizers.
Theorem 4 is stronger than the proof of the sufficiency in Theorem 3 because the eigenvalues are not presumed generic.
###### Theorem 5
If $`d>1`$ and if $`d_1+\mathrm{}+d_{p+1}2n^2+2`$, then conditions i) and ii) from Theorem 1 are necessary and sufficient for the existence of $`(p+1)`$-tuples of matrices $`M_j`$ satisfying (1) and with trivial centralizers. If the eigenvalues are relatively generic, then there exist irreducible such $`(p+1)`$-tuples.
Notice that the theorem does not require primitivity of $`\xi `$. The last two theorems are proved in Section 5.
###### Lemma 6
A) If $`q>1`$, if $`d_1+\mathrm{}+d_{p+1}=2n^2`$, if all $`(p+1)`$ Jordan normal forms are diagonal, if the eigenvalues $`\sigma _{k,j}`$ are relatively generic and if $`\xi `$ is not primitive, then such a $`(p+1)`$-tuple of matrices $`M_j`$ satisfying (1) (if it exists) is with trivial centralizer if and only if it is irreducible.
B) Let $`d_1+\mathrm{}+d_{p+1}2n^2+2`$ and let for the rest the conditions from A) hold. Then if there exist $`(p+1)`$-tuples of matrices $`M_j`$ satisfying (1), then there exist irreducible ones as well.
The lemma is proved in Subsection 3.2.
###### Theorem 7
Conditions i) and ii) from Theorem 1 are necessary for the existence of $`(p+1)`$-tuples of matrices $`M_j`$ or $`A_j`$ satisfying (1) or (2) and with a trivial centralizer.
The theorem claims necessity in all possible situations. It is proved in Section 4.
For generic eigenvalues Problem (I) (and Problem (TC) as well) is completely solved by Theorems 1 and 3.
For non-generic eigenvalues we focus on Problem (TC). The situations in which the answer is known are given by Theorems 4 and 5. It is shown in \[Ko1\] that conditions i) and ii) from Theorem 1 are necessary and sufficient for the solvability of Problem (TC) when $`d=1`$, $`(\alpha _n)`$ being a strict inequality.
The cases in which Problem (TC) remains to be considered for non-generic eigenvalues (in both versions – additive or multiplicative) and the conjectures the author makes are:
1) $`(\alpha _n)`$ is an equality (this implies $`d=1`$, see the remarks after Theorem 1).
Conjecture. Conditions i) and ii) from Theorem 1 are necessary and sufficient if $`q=1`$. For $`q>1`$ there are cases in which they are and cases in which they are not sufficient.
2) $`d>1`$, $`d_1+\mathrm{}+d_{p+1}=2n^2`$ and in the multiplicative version $`\xi `$ is not primitive.
Conjecture. There exist no $`(p+1)`$-tuples of matrices $`A_j`$ or $`M_j`$, satisfying (2) or (1), with trivial centralizers.
Remark: It is possible to give an exhaustive list of the $`(p+1)`$-tuples of Jordan normal forms of a given size $`n`$ admitting generic eigenvalues and defining irreducible representations of a fixed index of rigidity. Explain it first for rigid representations (for them one has $`n_s=1`$), with diagonal Jordan normal forms $`J_j^n`$.
Denote by $`\mathrm{\Xi }(m)`$ the set of all $`(p+1)`$-tuples of diagonal Jordan normal forms satisfying conditions i) and ii) from Theorem 1, of size $`m`$ (scalar Jordan normal forms are allowed). Find then all $`(p+1)`$-tuples of diagonal Jordan normal forms not from $`\mathrm{\Xi }(m)`$ which after applying the map $`\mathrm{\Psi }`$ result in a $`(p+1)`$-tuple from $`\mathrm{\Xi }(m)`$. To this end one has to assume that every diagonal Jordan normal form of a $`(p+1)`$-tuple from $`\mathrm{\Xi }(m)`$ has an eigenvalue of multiplicity 0 (which eventually was of positive multiplicity before applying $`\mathrm{\Psi }`$). Thus one can obtain $`\mathrm{\Xi }(m+1)`$ (one will have to exclude the $`(p+1)`$-tuples of size $`>m+1`$).
Having obtained the set $`\mathrm{\Xi }(n)`$, we leave only the $`(p+1)`$-tuples of size exactly $`n`$. Denote their set by $`\mathrm{\Theta }`$. Explain how to obtain the analog of $`\mathrm{\Theta }`$ obtained when the requirement the Jordan normal forms to be diagonal to be dropped. One constructs the set of $`(p+1)`$-tuples of Jordan normal forms $`J_j^{n,1}`$ such that the $`(p+1)`$-tuple of corresponding diagonal Jordan normal forms belongs to $`\mathrm{\Theta }`$ – this defines the set $`\mathrm{\Phi }`$. After this one excludes from $`\mathrm{\Phi }`$ the $`(p+1)`$-tuples not admitting generic eigenvalues – this gives the necessary list for index of rigidity equal to 2.
For index of rigidity $`h0`$ one has to give first the list $`L_h`$ of $`(p+1)`$-tuples of diagonal Jordan normal forms satisfying condition $`(\omega _n)`$; after this one constructs the sets $`\mathrm{\Xi }(n)`$, $`\mathrm{\Theta }`$ and $`\mathrm{\Phi }`$ by analogy with the case of index of rigidity equal to 2.
For index of rigidity $`h=0`$ the list $`L_0`$ contains only three triples and one quadruple of diagonal Jordan normal forms, with MVs equal to a) (1,1), (1,1), (1,1), (1,1) b) (1,1,1), (1,1,1), (1,1,1) c) (1,1,1,1), (1,1,1,1), (2,2) d) (1,1,1,1,1,1), (2,2,2), (3,3). The reader will be able to deduce this fact from the beginning of the last section and from Lemma 3 from \[Ko2\].
If one wants to obtain the list $`L_h`$ for $`h2`$, then one has to assume that $`L_0`$ contains four series of diagonal Jordan normal forms. They are a) (d,d), (d,d), (d,d), (d,d) b) (d,d,d), (d,d,d), (d,d,d) c) (d,d,d,d), (d,d,d,d), (2d,2d) d) (d,d,d,d,d,d), (2d,2d,2d), (3d,3d) where $`d𝐍`$ is such that the size of the matrices be $`n`$.
For negative indices of rigidity the list $`L_h`$ can be given by using again the ideas of the last section. First of all, we operate instad of with diagonal Jordan normal forms with their corresponding Jordan normal forms with a single eigenvalue. Having found the list $`L_{h2}`$ one obtains the list $`L_h`$ by using operations $`(s,l)`$ (defined in 2 of Subsubsection 6.3.1) and the inverse of the merging (defined in 6 of that subsubsection. The details are left for the reader.
## 2 Definitions and notation
### 2.1 Levelt’s result and its corollaries
In \[L\] Levelt describes the structure of the solution to a regular system at a pole:
###### Theorem 8
In a neighbourhood of a pole the solution to the regular linear system
$$\dot{X}=A(t)X$$
(3)
can be represented in the form
$$X=U_j(ta_j)(ta_j)^{D_j}(ta_j)^{E_j}G_j$$
(4)
where $`U_j`$ is holomorphic in a neighbourhood of the pole $`a_j`$, $`D_j=\text{diag}(\phi _{1,j},\mathrm{},\phi _{n,j})`$, $`\phi _{n,j}𝐙,`$ $`G_jGL(n,𝐂)`$. The matrix $`E_j`$ is in upper-triangular form and the real parts of its eigenvalues belong to $`[0,1)`$ (by definition, $`(ta_j)^{E_j}=e^{E_j\mathrm{ln}(ta_j)}`$). The numbers $`\phi _{k,j}`$ satisfy the condition (6) formulated below.
System (3) is fuchsian at $`a_j`$ if and only if
$$detU_j(0)0$$
(5)
We formulate the condition on $`\phi _{k,j}.`$ Let $`E_j`$ have one and the same eigenvalue in the rows with indices $`s_1<s_2<`$$`\mathrm{}`$$`<s_q`$. Then we have
$$\phi _{s_1,j}\phi _{s_2,j}\mathrm{}\phi _{s_q,j}$$
(6)
Remarks: 1) Denote by $`\beta _{k,j}`$ the diagonal entries (i.e. the eigenvalues) of the matrix $`E_j`$. If the system is fuchsian, then the sums $`\beta _{k,j}+\phi _{k,j}`$ are the eigenvalues $`\lambda _{k,j}`$ of the matrix-residuum $`A_j`$, see \[Bo1\], Corollary 2.1.
2) The numbers $`\phi _{k,j}`$ are defined as valuations in the solution eigensubspace for the eigenvalue $`\mathrm{exp}(2\pi i\beta _{k,j})`$ of the monodromy operator, see the details in \[L\]. These valuations can be defined on each subspace invariant for the monodromy operator.
3) One can assume without loss of generality that equal eigenvalues of $`E_j`$ occupy consecutive positions on the diagonal and that the matrix $`E_j`$ is block-diagonal, with diagonal blocks of sizes equal to their multiplicities. The diagonal blocks themselves are upper-triangular.
4) An improvement of Levelt’s form (4) can be found in \[Ko3\]. More precisely, the fact that all entries of $`E_j`$ above the diagonal can be made equal to 0 or to 1.
5) Let $`q>1`$. The rest of the division of $`_{j=1}^{p+1}_{k=1}^n\beta _{k,j}`$ (which is an integer) by $`q`$ equals $`m_0`$.
6) In \[Bo1\] A.Bolibrukh proves the following lemma (see Lemma 3.6 there):
###### Lemma 9
The sum of the numbers $`\phi _{k,j}+\beta _{k,j}`$ of system (3) corresponding to a subspace of the solution space invariant for all monodromy operators is a non-positive integer.
###### Corollary 10
Let the sum of the eigenvalues $`\lambda _{k,j}`$ of the matrices-residua of a fuchsian system corresponding to a subspace of the solution space of dimension $`m`$ invariant for all monodromy operators be 0. Denote these eigenvalues by $`\lambda _{k,j}^{}`$. Then there exists a change $`XRX`$, $`RGL(n,𝐂)`$ after which the system becomes block upper-triangular, the left upper block being of size $`m`$, the restrictions of the matrices-residua to it having eigenvalues $`\lambda _{k,j}^{}`$.
A proof of the corollary can be found in \[Bo2\].
### 2.2 Subordinate Jordan normal forms, canonical and strongly generic eigenvalues
Definition. Given two conjugacy classes $`c^{}`$, $`c^{\prime \prime }`$ with one and the same eigenvalues, of one and the same multiplicities, we say that $`c^{\prime \prime }`$ is subordinate to $`c^{}`$ if $`c^{\prime \prime }`$ lies in the closure of $`c^{}`$, i.e. for any matrix $`Ac^{\prime \prime }`$ there exists a deformation $`\stackrel{~}{A}(\epsilon )`$, $`\stackrel{~}{A}(0)=A`$ such that for $`0\epsilon (𝐂,0)`$ one has $`\stackrel{~}{A}(\epsilon )c^{}`$. Given two Jordan normal forms $`J^{}`$, $`J^{\prime \prime }`$, we say that $`J^{\prime \prime }`$ is subordinate to $`J^{}`$ if there exist conjugacy classes $`c^{}J^{}`$, $`c^{\prime \prime }J^{\prime \prime }`$ such that $`c^{\prime \prime }`$ is subordinate to $`c^{}`$.
In the text we denote by $`A_j`$ matrices from $`gl(n,𝐂)`$. They are often interpreted as matrices-residua of fuchsian systems on Riemann’s sphere, i.e. systems of the form
$$\dot{X}=(\underset{j=1}{\overset{p+2}{}}A_j/(ta_j))X$$
(7)
If this system has no pole at infinity, then one has $`A_1+\mathrm{}+A_{p+2}=0`$.
Remark. System (7) has $`p+2`$ poles because in what follows we have to realize the monodromy groups by fuchsian systems having an additional singularity with scalar local monodromy. When $`A_{p+2}=0`$ the system has $`p+1`$ poles.
Definition. The eigenvalues $`\lambda _{k,j}`$ of the matrix-residuum $`A_j`$ are called canonical if none of the differences between two of its eigenvalues is a non-zero integer. The eigenvalues of the corresponding monodromy operators $`M_j`$ equal $`\mathrm{exp}(2\pi i\lambda _{k,j})`$. Hence, if the eigenvalues of $`A_j`$ are canonical, then to equal (to different) eigenvalues of the corresponding monodromy operator there correspond equal (different) eigenvalues of $`A_j`$.
###### Proposition 11
If the eigenvalues of the matrix $`A_j`$ are canonical, then one has $`J(A_j)=J(M_j)`$ and $`M_j`$ is conjugate to $`\mathrm{exp}(2\pi iA_j)`$.
The proof can be found in \[Wa\].
Definition. Eigenvalues $`\lambda _{k,j}`$ satisfying none of the equalities ($`\gamma `$) modulo $`𝐙`$ (see the previous section) are called strongly generic; this definition is given only for eigenvalues $`\lambda _{k,j}`$; they are strongly generic if and only if the corresponding eigenvalues $`\sigma _{k,j}=\mathrm{exp}(2\pi i\lambda _{k,j})`$ are generic.
Definition. If $`q>1`$ and if the eigenvalues $`\lambda _{k,j}`$ satisfy none of the equalities ($`\gamma `$) modulo $`𝐙`$ except $`(\gamma ^{})`$ and its multiples (see Subsection 1.2), then the eigenvalues are called strongly relatively generic.
Definition. Let the eigenvalues $`\lambda _{k,j}`$ not be strongly generic. Hence, there holds at least one equality of the form
$$\underset{j=1}{\overset{p+2}{}}\underset{k\mathrm{\Phi }_j}{}\lambda _{k,j}=m,m𝐙,\mathrm{}\mathrm{\Phi }_1=\mathrm{}=\mathrm{}\mathrm{\Phi }_{p+2}=\kappa ,1\kappa <n(\gamma ^{})$$
The number $`|m|`$ is called distance of the set of eigenvalues to the relation $`(\gamma ^{})`$. The minimal of the numbers $`|m|`$ (over all relations ($`\gamma ^{}`$)) is called distance of the set of eigenvalues to the set of non-generic eigenvalues (or just distance). For generic eigenvalues their distance is by definition equal to $`\mathrm{}`$.
###### Lemma 12
A) Let the eigenvalues $`\sigma _{k,j}`$ defined by the conjugacy classes $`C_j`$ be non-generic and either with a simple PMV or with a non-simple one $`\xi `$ being a primitive root of unity, and let at least one of the classes $`C_j`$ (say, $`C_1`$) have at least two different eigenvalues. Then for every $`h𝐍`$ sufficiently large there exist eigenvalues $`\lambda _{k,j}`$ with zero sum such that
1) for all $`k`$, $`j`$ one has $`\mathrm{exp}(2\pi i\lambda _{k,j})=\sigma _{k,j}`$;
2) for $`jp`$ the eigenvalues $`\lambda _{k,j}`$ are canonical;
3) for $`j=p+1`$ if $`\lambda _{k_1,j}\lambda _{k_2,j}𝐙`$, then $`\lambda _{k_1,j}\lambda _{k_2,j}=0`$ or $`\pm 1`$;
4) the distance of the eigenvalues is $`h`$.
B) If $`\xi `$ is not primitive (the rest of the conditions being like in A)), then there exist eigenvalues $`\lambda _{k,j}`$ satisfying conditions 1), 2), 3) and
4’) for every relation $`(\gamma ^{})`$ which is not a multiple of $`(\gamma ^{})`$ its distance is $`h`$.
Before proving the lemma we deduce from it
###### Corollary 13
If the eigenvalues $`\lambda _{k,j}`$ of the matrices-residua $`A_j`$ of system (7) are like in the lemma, with $`h>n`$, and if the $`(p+1)`$-tuple of matrices $`A_j`$ is irreducible, then the monodromy group of the system is with trivial centralizer.
Notice that the irreducibility of the $`(p+1)`$-tuple of matrices $`A_j`$ is not automatic when $`\xi `$ is not primitive because there holds $`(\gamma ^{})`$.
Proof:
$`1^0`$. Suppose first (see $`1^0`$$`3^0`$) that $`\xi `$ is primitive. Assume that the monodromy group of system (7) satisfying the conditions of the corollary is with non-trivial centralizer $`𝒵`$. Then $`𝒵`$ either
a) contains a diagonalizable matrix $`D`$ with exactly two distinct eigenvalues or
b) contains a nilpotent matrix $`N`$ with $`N^2=0`$.
(Indeed, if $`X𝒵`$, then every polynomial of the semi-simple or of the nilpotent part of $`X`$ belongs to $`𝒵`$ which allows to construct such matrices $`D`$ or $`N`$.)
$`2^0`$. In case a) one can assume that $`D`$ is diagonal – $`D=\left(\begin{array}{cc}\alpha I& 0\\ 0& \beta I\end{array}\right)`$. Hence, the monodromy operators are of the form $`M_j=\left(\begin{array}{cc}M_j^{}& 0\\ 0& M_j^{\prime \prime }\end{array}\right)`$. Applying Lemma 9 to the sums $`\lambda ^{}`$, $`\lambda ^{\prime \prime }`$ of the eigenvalues $`\lambda _{k,j}`$ corresponding respectively to the $`(p+1)`$-tuples of blocks $`M_j^{}`$, $`M_j^{\prime \prime }`$, one obtains the inequalities $`\lambda ^{}0`$, $`\lambda ^{\prime \prime }0`$. On the other hand, $`\lambda ^{}+\lambda ^{\prime \prime }=0`$. Hence, $`\lambda ^{}=\lambda ^{\prime \prime }=0`$ which contradicts the condition $`h>0`$.
$`3^0`$. In case b) one can assume that $`N=\left(\begin{array}{ccc}0& 0& I\\ 0& 0& 0\\ 0& 0& 0\end{array}\right)`$ and $`M_j=\left(\begin{array}{ccc}M_j^{}& & \\ 0& M_j^{\prime \prime }& \\ 0& 0& M_j^{}\end{array}\right)`$ (the form of $`N`$ can be achieved by conjugation, the one of $`M_j`$ results from $`[M_j,N]=0`$).
Denote by $`\lambda ^1`$, $`\lambda ^2`$, $`\lambda ^3`$ the sums of the eigenvalues $`\lambda _{k,j}`$ corresponding respectively to the $`(p+1)`$-tuples of upper blocks $`M_j^{}`$, of blocks $`M_j^{\prime \prime }`$, of lower blocks $`M_j^{}`$. One has $`\lambda ^1+\lambda ^2+\lambda ^3=0`$, $`\lambda ^10`$, $`\lambda ^1+\lambda ^20`$, $`\lambda ^2+\lambda ^30`$ $`\lambda ^30`$ (Lemma 9) and $`|\lambda ^1\lambda ^3|<n`$ because the eigenvalues $`\lambda _{k,j}`$ are canonical for $`jp`$ and for $`j=p+1`$ there holds condition 3) of the lemma (and the blocks $`M_j^{}`$, $`M_j^{\prime \prime }`$ are of sizes $`<n`$). Hence, either $`|\lambda ^1+\lambda ^2|<n`$ or $`|\lambda ^2+\lambda ^3|<n`$ which is a contradiction with the distance of the set of eigenvalues to the set of non-generic eigenvalues being $`>n`$.
$`4^0`$. Suppose now that $`\xi `$ is not primitive. If the monodromy group of the system is with non-trivial centralizer, then in case a) one obtains $`\lambda ^{}=\lambda ^{\prime \prime }=0`$. This is still not a contradiction with the condition $`h>0`$ because there remains the relation $`(\gamma ^{})`$.
By Corollary 10, one can make a change $`XRX`$ and block-triangularize the matrices-residua of the system. The change results in $`A_jR^1A_jR`$. This contradicts the condition the $`(p+1)`$-tuple of matrices $`A_j`$ to be irreducible.
In case b) one has $`\lambda ^1+\lambda ^2+\lambda ^3=0`$, $`\lambda ^10`$, $`\lambda ^1+\lambda ^20`$, $`\lambda ^2+\lambda ^30`$ $`\lambda ^30`$ (Lemma 9) and $`|\lambda ^1\lambda ^3|<n`$ (like in $`3^0`$). This is possible only if $`\lambda ^1=\lambda ^2=\lambda ^3=0`$ and each of the last equalities results from $`(\gamma ^{})`$. By Corollary 10, one can block-triangularize the matrices-residua of the system by a change $`XRX`$. This again contradicts the irreducibility of the $`(p+1)`$-tuple of matrices $`A_j`$.
The corollary is proved.
Proof of Lemma 12:
$`1^0`$. Prove A) (see $`1^0`$$`7^0`$). Fix some canonical eigenvalues $`\lambda _{k,j}^0`$ satisfying condition 1) whose sum eventually is non-zero. Denote by $`l`$ the rest of the division of their sum (which is necessarily integer) by $`n`$. Decrease $`l`$ of the eigenvalues $`\lambda _{k,p+1}^0`$ by 1 (and don’t change the other eigenvalues $`\lambda _{k,j}`$) – this defines the eigenvalues $`\lambda _{k,j}^1`$.
$`2^0`$. Fix integers $`g_{k,j}`$ such that
a) to equal eigenvalues $`\lambda _{k,j}^0`$ there correspond equal integers $`g_{k,j}`$ and
b) $`_{k=1}^n_{j=1}^{p+1}(\lambda _{k,j}^1+g_{k,j})=0`$.
Set $`\lambda _{k,j}^2=\lambda _{k,j}^1+g_{k,j}`$.
$`3^0`$. Denote by $`\lambda _j^{}`$, $`\lambda _j^{\prime \prime }`$ two different eigenvalues $`\lambda _{k,j}^0`$, of multiplicities $`m^{}`$, $`m^{\prime \prime }`$. Change their corresponding numbers $`g_{k,j}`$ (denoted by $`g_j^{}`$, $`g_j^{\prime \prime }`$) respectively to $`g_j^{}+um^{\prime \prime }`$, $`g_j^{\prime \prime }um^{}`$, $`u𝐍`$. This defines a new set of eigenvalues $`\lambda _{k,j}^2`$. (The other integers $`g_{k,j}`$ do not change.)
$`4^0`$. For a given relation ($`\gamma ^{}`$) (satisfied by the eigenvalues $`\lambda _{k,j}^0`$ instead of $`\lambda _{k,j}`$) call quasi-multiplicity of an eigenvalue $`\lambda _{k,j}^0`$ the number of eigenvalues $`\lambda _{i,j}^0`$ equal to $`\lambda _{k,j}^0`$ such that $`i\mathrm{\Phi }_j`$ (the sets $`\mathrm{\Phi }_j`$ were defined in Subsection 1.1). For each given relation ($`\gamma ^{})`$ the quasi-multiplicities of the eigenvalues $`\lambda _{k,j}^0`$ are not proportional to their multiplicities because either the PMV is simple or it is not but $`\xi `$ is a primitive root of unity.
$`5^0`$. For each relation ($`\gamma ^{}`$) and for each couple of eigenvalues $`\lambda _j^{}`$, $`\lambda _j^{\prime \prime }`$ like in $`3^0`$ either
1’) for all values of $`u𝐍`$ the distance of the set of eigenvalues to the relation ($`\gamma ^{})`$ remains the same or
2’) only for finitely many of them this distance is $`<h`$.
$`6^0`$. For a given relation ($`\gamma ^{}`$) there exists a couple of eigenvalues $`\lambda _j^{}`$, $`\lambda _j^{\prime \prime }`$ like in $`3^0`$ for which there holds 2’). Having chosen this couple, denote by $`\mathrm{\Xi }`$ the set of relations ($`\gamma ^{}`$) for which the chosen couple satisfies 2’). Hence, for the chosen couple one can find a finite subset $`N_0`$ of $`𝐍`$ such that for $`u(𝐍\backslash N_0)`$ the distance of the set of eigenvalues $`\lambda _{k,j}^2`$ to each of the relations from $`\mathrm{\Xi }`$ is $`h`$. Fix $`u=u_0(𝐍\backslash N_0)`$. This means that we change the eigenvalues $`\lambda _{k,j}^2`$.
$`7^0`$. If $`\mathrm{\Xi }`$ is not the set of all relations ($`\gamma ^{}`$), then choose such a relation not from it and repeat what was done in $`6^0`$. Every time we do this, the distance of the set of eigenvalues to more and more relations ($`\gamma ^{}`$) becomes $`h`$ and the distance to the rest of them does not change. As there are finitely many relations ($`\gamma ^{}`$), after finitely many steps the distance of the set of eigenvalues $`\lambda _{k,j}^2`$ to the set of non-generic eigenvalues becomes $`h`$, i.e. condition 4) holds. Conditions 1) – 3) hold by construction.
$`8^0`$. The proof of B) is completely analogous. One first divides the multiplicities of all eigenvalues by $`(m_0,q)`$, then constructs the eigenvalues $`\lambda _{k,j}`$ like in case A) and then one multiplies the multiplicities by $`(m_0,q)`$. The distances of $`(\gamma ^{})`$ and of its multiples remain 0, the distances of the other relations are $`(m_0,q)h>h`$.
The lemma is proved.
### 2.3 Corresponding Jordan normal forms and normalized chains
In \[Ko1\] we define correspondence between Jordan normal forms. Namely, for each Jordan normal form $`J_0`$ we define its corresponding diagonal Jordan normal form $`J_1`$ as follows.
Suppose first that $`J_0`$ has a single eigenvalue of multiplicity $`n`$. Replace each Jordan block of $`J_0`$ of size $`q^{}\times q^{}`$ by a diagonal $`q^{}\times q^{}`$-matrix with $`q^{}`$ distinct eigenvalues where the last eigenvalues of the diagonal matrices replacing Jordan blocks of $`J_0`$ are the same, their last but one eigenvalues are the same and different from the last ones and for all $`s0`$ their last but $`s`$ eigenvalues (denoted by $`h_s`$) are the same and different from the last but $`l`$ ones for $`l<s`$.
If $`J_0`$ has several eigenvalues, then this procedure is performed for every eigenvalue with the requirement eigenvalues of $`J_1`$ corresponding to different eigenvalues of $`J_0`$ to be different.
Remark: The definition of $`J_1`$ after $`J_0`$ can be described in equivalent terms like this. Call Jordan normal form (JNF) of size $`n`$ a family $`J^n=\{b_{i,l}\}`$ ($`iI_l`$, $`I_l=\{1,\mathrm{},s_l\}`$, $`lL`$) of positive integers $`b_{i,l}`$ whose sum is $`n`$. Here $`L`$ is the set of eigenvalues (all distinct) and $`I_l`$ is the set of Jordan blocks with eigenvalue $`l`$, $`b_{i,l}`$ is the size of the $`i`$-th block with this eigenvalue. An $`n\times n`$-matrix $`Y`$ has the JNF $`J^n`$ (notation: $`J(Y)=J^n`$) if to its distinct eigenvalues $`\lambda _l`$, $`lL`$, there belong Jordan blocks of sizes $`b_{i,l}`$. For a given JNF $`J^n=\{b_{i,l}\}`$ define its corresponding diagonal JNF $`J_{}^{}{}_{}{}^{n}`$. A diagonal JNF is a partition of $`n`$ defined by the multiplicities of the eigenvalues. For each $`l`$ $`\{b_{i,l}\}`$ is a partition of $`_{iI_l}b_{i,l}`$ and $`J^{}`$ is the disjoint sum of the dual partitions. Thus if for each fixed $`l`$ one has $`b_{1,l}`$$`\mathrm{}`$$`b_{s_l,l}`$, then each eigenvalue $`lL`$ is replaced by $`b_{1,l}`$ new eigenvalues $`h_{1,l}`$, $`\mathrm{}`$, $`h_{b_{1,l},l}`$ (hence, $`J_{}^{}{}_{}{}^{n}`$ has $`_{lL}b_{1,l}`$ distinct eigenvalues). For $`l`$ fixed, set $`g_k`$ for the multiplicity of the eigenvalue $`h_{k,l}`$. Then the first $`b_{s_l,l}`$ numbers $`g_k`$ equal $`s_l`$, the next $`b_{s_{l1},l}b_{s_l,l}`$ equal $`s_l1`$, $`\mathrm{}`$, the last $`b_{1,l}b_{2,l}`$ equal 1.
Definition. Two Jordan normal forms correspond to one another if they correspond to one and the same diagonal Jordan normal form (and thus the correspondence of Jordan normal forms is a relation of equivalence).
Definition. We say that the eigenvalues of a Jordan matrix $`G`$ with Jordan normal form $`J_1`$ form a normalized chain w.r.t. the Jordan normal form $`J_0`$ if for every eigenvalue of $`J_0`$ and for all $`s`$ one has $`h_sh_{s1}𝐍^+`$ and any two eigenvalues of $`G`$ corresponding to different eigenvalues of $`J_0`$ are non-congruent modulo $`𝐙`$.
The following properties of corresponding Jordan normal forms are proved in \[Ko1\]:
###### Theorem 14
1) If the Jordan normal forms $`J_0`$ and $`J_1`$ correspond to one another, then $`r(J_0)=r(J_1)`$ and $`d(J_0)=d(J_1)`$.
2) To each Jordan normal form $`J^{}=\{b_{i,l}^{}\}`$ there corresponds a single Jordan normal form with a single eigenvalue $`J^{\prime \prime }=\{b_{k,1}^{\prime \prime }\}`$. One has $`b_{k,1}^{\prime \prime }=_lb_{k,l}^{}`$, $`k=1,\mathrm{},s_1`$.
3) Let the two Jordan normal forms $`J_{}^{n}{}_{}{}^{}`$ and $`J_{}^{n}{}_{}{}^{\prime \prime }`$ correspond to one another. Choose in each of them an eigenvalue with maximal number of Jordan blocks. By 1) these numbers coincide. Denote them by $`k^{}`$. Decrease by 1 the sizes of the $`k`$ smallest Jordan blocks with these eigenvalues where $`kk^{}`$. Then the two Jordan normal forms of size $`nk`$ obtained in this way correspond to one another.
4) Denote by $`G^0`$, $`G^1`$ two Jordan matrices with Jordan normal forms $`J_0`$, $`J_1`$ corresponding to each other, where $`G^1`$ is diagonal and $`G^0`$ is nilpotent and they are defined like $`J_1`$ and $`J_0`$ from the beginning of the subsection (when $`J_1`$ and $`J_0`$ are considered like matrices ($`J_0`$ being Jordan), not like Jordan normal forms). Then the orbits of the matrices $`\epsilon G^1`$ and $`G^0+\epsilon G^1`$ are the same for $`\epsilon 𝐂^{}`$.
5) If $`G^0`$ is not necessarily nilpotent and if $`G^0`$, $`G^1`$ are defined by analogy with $`J_0`$, $`J_1`$, then the matrix $`G^0+\epsilon G^1`$ is diagonalizable and its Jordan normal form is $`J_1`$ if $`\epsilon 𝐂^{}`$ is small enough.
###### Corollary 15
Consider two $`(p+1)`$-tuples of Jordan normal forms – $`J_j^n`$ and $`J_{j}^{n}{}_{}{}^{}`$ – where for all $`j`$ the two Jordan normal forms $`J_j^n`$ and $`J_{j}^{n}{}_{}{}^{}`$ correspond to one another. Construct for each of the two $`(p+1)`$-tuples the $`(p+1)`$-tuples of Jordan normal forms $`J_j^{n_\nu }`$ and $`J_j^{n_\nu ^{}}`$ like explained before Theorem 1. Then one has $`n_\nu =n_\nu ^{}`$ for all $`\nu `$, and for all $`\nu `$ and all $`j`$ the Jordan normal forms $`J_j^{n_\nu }`$ and $`J_j^{n_\nu ^{}}`$ correspond to one another.
The corollary follows from 1) and 3) of the above theorem.
Denote by $`J`$ and $`J^{}`$ an arbitrary Jordan normal form and its corresponding Jordan normal form with a single eigenvalue. Consider a couple $`D`$, $`D^{}`$ of Jordan matrices with these Jordan normal forms. Suppose that the Jordan blocks of $`D`$ of sizes $`b_{k,l}`$ for $`k`$ fixed are situated in the same rows where is situated the Jordan block of size $`_lb_{k,l}`$ of $`D^{}`$ (see 2) of the above theorem). Denote by $`D_0`$ the diagonal (i.e. semi-simple) part of the matrix $`D`$.
###### Proposition 16
For all $`\epsilon 0`$ the matrix $`\epsilon D_0+D^{}`$ is conjugate to $`\epsilon D`$.
Remark: Permute the diagonal entries of $`D_0`$ so that before and after the permutation each entry remains in one of the rows of one and the same Jordan block of $`D^{}`$. Then the proposition holds again and the proof is the same, see Proposition 20 from \[Ko1\].
Notation. We denote by $`a_j`$ the poles of the fuchsian system (7) and by $`\alpha _j`$ the quantities $`1/(a_ja_{p+2})`$. We often use the matrix $`=`$diag$`(1,\mathrm{},1,0,\mathrm{},0)`$ ($`m_0`$ times $`1`$ and $`nm_0`$ times 0). The matrix with a single entry (equal to 1) in position $`(i,j)`$ is denoted by $`E_{i,j}`$. In all other cases double subscripts denote matrix entries.
### 2.4 Plan of the paper
The main difficulty in the case $`d>1`$ is the impossibility to realize the monodromy groups by fuchsian systems with canonical eigenvalues. If the eigenvalues $`\lambda _{k,j}`$ are represented in the form $`\beta _{k,j}+\phi _{k,j}`$ with $`\phi _{k,j}𝐙`$ and with Re$`(\beta _{k,j})[0,1)`$, then the sum $`_{j=1}^{p+1}\beta _{k,j}`$ is an integer (the sum of all eigenvalues $`\lambda _{k,j}`$ is 0, see (2)). If this integer is not divisible by $`d`$, then there exists no set of canonical eigenvalues $`\lambda _{k,j}`$ (canonical for each index $`j`$).
Therefore we realize such monodromy groups by fuchsian systems having $`(p+2)`$-nd singularities with scalar local monodromy. More exactly – equal to id (such singularities are called apparent. (This is why in the formulation of the problem we speak about $`(p+1)`$ matrices $`M_j`$ and here we consider $`(p+2)`$-tuples of matrices $`A_j`$.) The eigenvalues at the first $`(p+1)`$ singular points are canonical. The ones at the $`(p+2)`$-nd singularity compensate the rest $`m_0`$ of the division by $`d`$ of the above sum ($`1m_0d1`$).
The necessity to operate with systems with additional singularities explains why we begin the present second part of \[Ko1\] by the study of $`(p+2)`$-tuples of matrices $`A_jgl(n,𝐂)`$ with zero sum and satisfying certain linear equalities (see Subsection 3.3). For a fuchsian system defined by such a $`(p+2)`$-tuple of matrices for fixed poles these equalities provide the necessary and sufficient condition the monodromy around one of the singularities of the fuchsian system (7) to be scalar.
Section 3 contains also the modification of the basic technical tool used in \[Ko1\]. It is used to construct analytic deformations of $`(p+2)`$-tuples of matrices $`A_j`$ with trivial centralizer and satisfying certain linear equalities.
Section 4 contains the proof of the necessity of conditions i) and ii) for the existence of irreducible $`(p+1)`$-tuples of matrices $`M_j`$ satisfying (1) in the case $`d>1`$; Section 5 contains the proof of their sufficiency. In the proof of the sufficiency we use some results concerning the existence of irreducible $`(p+1)`$-tuples of nilpotent matrices with zero sum, see Section 6.
## 3 Some preliminaries
### 3.1 The basic technical tool
The basic technical tool is a way to deform a $`(p+1)`$-tuple of matrices $`M_j`$ or $`A_j`$ with a trivial centralizer into one with either new eigenvalues and the same Jordan normal forms of the respective matrices, or into one in which some of these Jordan normal forms $`J_j^n`$ are replaced by other Jordan normal forms $`J_{j}^{n}{}_{}{}^{}`$ where $`J_j^n`$ is subordinate to $`J_{j}^{n}{}_{}{}^{}`$, or where $`J_{j}^{n}{}_{}{}^{}`$ corresponds to $`J_j^n`$. Explain how it works in the multiplicative version first.
Given a $`(p+1)`$-tuple of matrices $`M_j^1`$ satisfying condition (1) and whose centralizer is trivial, look for $`M_j`$ of the form
$$M_j=(I+\epsilon X_j(\epsilon ))^1(M_j^1+\epsilon N_j(\epsilon ))(I+\epsilon X_j(\epsilon ))$$
where the given matrices $`N_j`$ depend analytically on $`\epsilon (𝐂,0)`$ and one looks for $`X_j`$ analytic in $`\epsilon `$. (One can set $`M_j^1=Q_j^1G_jQ_j`$, $`N_j=Q_j^1V_j(\epsilon )Q_j`$ where $`G_j`$ are Jordan matrices and $`Q_jGL(n,𝐂)`$.)
The matrices $`M_j`$ must satisfy equality (1). In first approximation w.r.t. $`\epsilon `$ this yields
$$\underset{j=1}{\overset{p+1}{}}M_1^1\mathrm{}M_{j1}^1([M_j^1,X_j(0)]+N_j(0))M_{j+1}^1\mathrm{}M_{p+1}^1=0$$
or
$$\underset{j=1}{\overset{p+1}{}}P_{j1}([M_j^1,X_j(0)(M_j^1)^1]+N_j(0)(M_j^1)^1)P_{j1}^1=0$$
(8)
with $`P_j=M_1^1\mathrm{}M_j^1`$, $`P_1=I`$ (recall that there holds (1), therefore $`M_j^1M_{j+1}^1\mathrm{}M_{p+1}^1=P_{j1}^1`$).
Condition (1) implies that $`detM_1\mathrm{}detM_{p+1}=1`$. There holds $`detM_j=detM_j^1det(I+\epsilon (M_j^1)^1N_j)`$ $`=(detM_j^1)(1+\epsilon `$tr$`((M_j^1)^1N_j(0))+o(\epsilon ))`$. As $`detM_1^1\mathrm{}detM_{p+1}^1=1`$, one has tr$`(_{j=1}^{p+1}(M_j^1)^1N_j(0))`$$`=0`$ (term of first order w.r.t. $`\epsilon `$ in $`detM_1\mathrm{}detM_{p+1}`$).
Equation (8) admits the following equivalent form:
$$\underset{j=1}{\overset{p+1}{}}([S_j,Z_j]+T_j)=0$$
(9)
with $`S_j=P_{j1}M_j^1P_{j1}^1`$, $`Z_j=P_{j1}X_j(0)(M_j^1)^1P_{j1}^1`$, $`T_j=P_{j1}N_j(0)(M_j^1)^1P_{j1}^1`$.
The centralizers of the $`(p+1)`$-tuples of matrices $`M_j^1`$ and $`S_j`$ coincide (to be checked directly), i.e. they are both trivial. There holds
###### Proposition 17
The $`(p+1)`$-tuple of matrices $`A_j`$ is with trivial centralizer if and only if the mapping $`(sl(n,𝐂))^{p+1}sl(n,𝐂)`$, $`(X_1,\mathrm{},X_{p+1})_{j=1}^{p+1}[A_j,X_j]`$ is surjective.
Proof: The mapping is not surjective if and only if the images of all mappings $`X_j[A_j,X_j]`$ belong to one and the same proper linear subspace of $`sl(n,𝐂)`$ which can be defined by a condition of the form tr$`(D[A_j,X_j])=0`$ for all $`X_jsl(n,𝐂)`$ where $`0Dsl(n,𝐂)`$. This amounts to tr$`([D,A_j]X_j)=0`$ for all $`X_jsl(n,𝐂)`$, i.e. $`[D,A_j]=0`$ for $`j=1,\mathrm{},p+1`$.
The proposition is proved.
The mapping
$$(sl(n,𝐂))^{p+1}sl(n,𝐂),(Z_1,\mathrm{},Z_{p+1})\underset{j=1}{\overset{p+1}{}}[S_j,Z_j]$$
is surjective (Proposition 17). Recall that tr$`(_{j=1}^{p+1}(M_j^1)^1N_j(0))=0`$, i.e. tr$`(_{j=1}^{p+1}T_j)=0`$. Hence, equation (9) is solvable w.r.t. the unknown matrices $`Z_j`$ and, hence, equation (8) is solvable w.r.t. the matrices $`X_j(0)`$. The implicit function theorem implies (we use the surjectivity here) that one can find $`X_j`$ analytic in $`\epsilon (𝐂,0)`$, i.e. one can find the necessary matrices $`M_j`$.
In the additive version one has matrices $`A_j^1=Q_j^1G_jQ_j`$ instead of $`M_j^1`$ and one sets
$$\stackrel{~}{A}_j=(I+\epsilon X_j(\epsilon ))^1Q_j^1(G_j+\epsilon V_j(\epsilon ))Q_j(I+\epsilon X_j(\epsilon ))$$
where $`V_j(\epsilon )`$ are given matrices analytic in $`\epsilon `$; one has tr$`(_{j=1}^{p+1}V_j(\epsilon ))0`$. The matrices $`X_j(0)`$ satisfy the equation (which is of the form (9))
$$\underset{j=1}{\overset{p+1}{}}[A_j^1,X_j(0)]=\underset{j=1}{\overset{p+1}{}}Q_j^1V_jQ_j.$$
The existence of $`X_j`$ analytic in $`\epsilon `$ is justified like in the multiplicative version.
###### Lemma 18
If for given Jordan normal forms $`J_j^n`$ of the matrices $`M_j`$ and generic eigenvalues there exists irreducible $`(p+1)`$-tuples of such matrices (satisfying (1)), then there exist such $`(p+1)`$-tuples for $`M_j`$ from the corresponding diagonal Jordan normal forms (and with generic eigenvalues).
Proof: $`1^0`$. Denote by $`J_j^{}`$ and $`J_j^{\prime \prime }`$ an arbitrary and its corresponding diagonal Jordan normal form. Let the $`(p+1)`$-tuple of matrices $`M_j^0`$ be irreducible, with generic eigenvalues and satisfying (1). Set $`M_j^0=Q_j^1G_jQ_j`$ where $`G_j`$ are Jordan matrices and $`J(G_j)=J_j^{}`$.
$`2^0`$. Construct a deformation of the $`(p+1)`$-tuple of the form
$$M_j=(I+\epsilon X_j(\epsilon ))^1Q_j^1(G_j+\epsilon L_j)Q_j(I+\epsilon X_j(\epsilon ))$$
where $`X_j`$ are analytic in $`\epsilon (𝐂,0)`$ and $`L_j`$ are diagonal matrices with $`J(L_j)=J_j^{\prime \prime }`$. More exactly, assume that $`G_j`$ and $`L_j`$ are defined respectively like $`G^0`$ and $`G^1`$ from 4) of Theorem 14. By 5) of that theorem, one has $`J(M_j)=J(L_j)`$ for $`\epsilon 0`$ small enough.
$`3^0`$. The basic technical tool provides the existence of irreducible $`(p+1)`$-tuples of matrices $`M_j`$ for $`\epsilon 0`$ small enough. Their Jordan normal forms are $`J_j^{\prime \prime }`$ and their eigenvalues are generic.
The lemma is proved.
### 3.2 Proof of Lemma 6
$`1^0`$. Prove A) (see $`1^0`$$`5^0`$). Suppose that there exists such a $`(p+1)`$-tuple of matrices $`M_j`$ with trivial centralizer but not irreducible. Then it can be conjugated to a block upper-triangular form with irreducible diagonal blocks whose eigenvalue satisfy the only non-genericity relation $`(\gamma ^{})`$ (defined in Subsection 1.2) and eventually some of its multiples.
$`2^0`$. Consider two diagonal blocks and the representations $`\mathrm{\Phi }_1`$, $`\mathrm{\Phi }_2`$ (of sizes $`m_1`$, $`m_2`$) which they define. The Jordan normal forms of each of the matrices $`M_j`$ restricted to each of these blocks is a multiple of one and the same diagonal Jordan normal form and the ratio of the multiplicities of one and the same eigenvalues of $`M_j`$ as eigenvalues of $`\mathrm{\Phi }_1`$ and $`\mathrm{\Phi }_2`$ equals $`m_1/m_2`$ for every eigenvalue.
$`3^0`$. For the dimensions $`d_{j,i}`$ of the conjugacy classes of the restrictions of $`M_j`$ to the two diagonal blocks one has $`d_{1,i}+\mathrm{}+d_{p+1,i}=2(m_i)^2`$. A direct computation shows that dim Ext$`{}_{}{}^{1}(\mathrm{\Phi }_1,\mathrm{\Phi }_2)=0`$.
Indeed, consider the case when there are only two diagonal blocks and $`M_j=\left(\begin{array}{cc}M_j^1& F_j\\ 0& M_j^2\end{array}\right)`$. Hence, there exist matrices $`G_j`$ such that $`F_j=M_j^1G_jG_jM_j^2`$.
One has dim$`E=2m_1m_2`$ where $`E=\{(F_1,\mathrm{},F_{p+1})|F_j=M_j^1G_jG_jM_j^2\}`$ (to be checked by the reader). The subspace $`E^{}`$ of $`E`$ defined by
$$\underset{j=1}{\overset{p+1}{}}P_j=0,P_j:=M_1^1\mathrm{}M_{j1}^1F_jM_{j+1}^2\mathrm{}M_{p+1}^2\}$$
(condition resulting from (1)) is of dimension $`m_1m_2`$. These conditions are linearly independent (the change of variables $`(F_1,\mathrm{},F_{p+1})`$$`(P_1,\mathrm{},P_{p+1})`$ is bijective due to det$`M_j^i0`$).
$`4^0`$. One has dim$`E^{}=`$dim$`E^{\prime \prime }`$ where $`E^{\prime \prime }=\{(F_1,\mathrm{}F_{p+1})|F_j=M_j^1GGM_j^2,GM_{m_1,m_2}(𝐂)\}`$. This is the space of blocks $`F`$ of size $`m_1\times m_2`$ resulting from the simultaneous congugation of the $`(p+1)`$-tuple of matrices $`M_j^0=\left(\begin{array}{cc}M_j^1& 0\\ 0& M_j^2\end{array}\right)`$ by a matrix $`\left(\begin{array}{cc}I& G\\ 0& I\end{array}\right)`$.
Thus dim$`(E^{}/E^{\prime \prime })=`$dim Ext$`{}_{}{}^{1}(\mathrm{\Phi }_1,\mathrm{\Phi }_2)=0`$.
$`5^0`$. This is true for every couple of diagonal blocks $`\mathrm{\Phi }_1`$, $`\mathrm{\Phi }_2`$. Hence, it is possible to conjugate the $`(p+1)`$-tuple to a block-diagonal form which contradicts the triviality of the centralizer.
$`6^0`$. Prove B). Let $`\mathrm{\Phi }_1`$ and $`\mathrm{\Phi }_2`$ have the same meaning as above. Then dim Ext$`{}_{}{}^{1}(\mathrm{\Phi }_1,\mathrm{\Phi }_2)2`$ and one can construct a semi-direct sum of $`\mathrm{\Phi }_1`$, $`\mathrm{\Phi }_2`$ which is not a direct one. Suppose that it is defined by matrices $`M_j`$ like in $`3^0`$. One can assume that the representations $`\mathrm{\Phi }_1`$ and $`\mathrm{\Phi }_2`$ are not equivalent (even if $`m_1=m_2`$) because for neither of them neither of conditions $`(\alpha _{m_1})`$, $`(\alpha _{m_2})`$ (which they satisfy) is an equality and there exist small deformations of the representations into nearby non-equivalent ones; when $`(\alpha _n)`$ is an equality, then such an irreducible representation is said to be rigid; it is unique up to conjugacy, see \[Si\] and \[Ka\].
$`7^0`$. There exist infinitesimal conjugations of the matrices $`M_j`$ of the form
$$\stackrel{~}{M}_j=(I+\epsilon X_j)^1M_j(I+\epsilon X_j)$$
such that in first approximation w.r.t. $`\epsilon `$ one has $`\stackrel{~}{M}_1\mathrm{}\stackrel{~}{M}_{p+1}=I`$ and the conjugations do not result from a simultaneous infinitesimal conjugation of the matrices $`M_j`$. This follows from $`d_1+\mathrm{}+d_{p+1}2n^2+2`$.
The details look like this: set $`X_j=\left(\begin{array}{cc}V_j& W_j\\ U_j& S_j\end{array}\right)`$. One can assume that $`W_j=0`$ because infinitesimal conjugations of $`M_j`$ with the matrices $`\left(\begin{array}{cc}I& \epsilon W_j\\ 0& I\end{array}\right)`$ do not change the block upper-triangular form of the $`(p+1)`$-tuple. Hence,
$$\stackrel{~}{M}_j=M_j+\epsilon \left(\begin{array}{cc}[M_j^1,V_j]+F_jU_j& F_jS_jV_jM_j^2\\ M_j^2U_jU_jM_j^1& [M_j^2,S_j]U_jF_j\end{array}\right)=o(\epsilon )$$
Set
$$T=\{(U_1,\mathrm{},U_{p+1})|U_jM_{m_2,m_1}(𝐂),\underset{j=1}{\overset{p+1}{}}Y_j=0\}$$
where $`Y_j=M_1^2\mathrm{}M_{j1}^2U_jM_{j+1}^1\mathrm{}M_{p+1}^1`$ (the condition $`_{j=1}^{p+1}Y_j=0`$ is the condition $`\stackrel{~}{M}_1\mathrm{}\stackrel{~}{M}_{p+1}=I`$ restricted to the left lower $`m_2\times m_1`$-block and considered in first approximation w.r.t. $`\epsilon `$). One has dim$`Tm_1m_2+2`$ (this results from $`d_1+\mathrm{}+d_{p+1}2n^2+2`$).
The subspace $`T^{}`$ of $`T`$ defined by the condition Tr$`_{j=1}^{p+1}F_jU_j=0`$ is of codimension 1 in it, i.e. of dimension $`m_1m_2+1`$. This is more than the size of the block $`U`$, i.e. more than the dimension of simultaneous infinitesimal conjugations with matrices $`\left(\begin{array}{cc}I& 0\\ \epsilon U& I\end{array}\right)`$ (this is the subspace of $`T`$ of the form $`\{(U,\mathrm{},U)\}`$). Hence, one can choose $`U_j`$ from $`T^{}/\{(U,\mathrm{},U)\}`$ and after this choose $`V_j`$ and $`S_j`$ such that
$$\underset{j=1}{\overset{p+1}{}}([M_j^1,V_j]+F_jU_j)=0,\underset{j=1}{\overset{p+1}{}}([M_j^2,S_j]U_jF_j)=0$$
The matrices $`X_j`$ define the infinitesimal deformations $`\stackrel{~}{M}_j`$.
$`8^0`$. There exists also a true deformation of the form
$$\stackrel{~}{M}_j=(I+\epsilon X_j+\epsilon ^2Y_j(\epsilon ))^1M_j(I+\epsilon X_j+\epsilon ^2Y_j(\epsilon ))$$
with $`Y_j`$ analytic in $`\epsilon `$. The existence is justified by analogy with the basic technical tool and we leave the details for the reader. The triviality of the centralizer of the $`(p+1)`$-tuple of matrices $`M_j`$ makes the implicit function theorem applicable.
Hence, for $`\epsilon 0`$ small enough the $`(p+1)`$-tuple of matrices $`\stackrel{~}{M}_j`$ is irreducible.
The lemma is proved.
### 3.3 The set $`𝒮`$
Fix the distinct complex numbers $`a_1`$, $`\mathrm{}`$, $`a_{p+2}`$. Consider the set $`𝒮`$ of $`(p+2)`$-tuples $`A^{}`$ of matrices $`(A_1,\mathrm{},A_{p+2})`$ such that
1) $`A_{p+2}=`$, $``$ was defined at the end of Subsection 2.3;
2) $`A_1+\mathrm{}+A_{p+2}=0`$;
3) $`(_{j=1}^{p+1}\alpha _jA_j)|_{\kappa ,\nu }=0`$, $`\kappa =m_0+1,\mathrm{},n`$; $`\nu =1,\mathrm{},m_0`$;
4) $`n=dn^{}`$, $`n^{}𝐍`$, $`d>1`$ and $`(d,m_0)=1`$, see Subsection 1.2;
5) the Jordan normal forms of the matrices $`A_j`$ are fixed.
###### Lemma 19
Let the matrices $`A_j`$ satisfy conditions 1) and 2). Then the monodromy operator at $`a_{p+2}`$ of the fuchsian system (7) is scalar if and only if condition 3) holds (in which case it equals $`I`$).
Proof:
Represent system (7) locally, at $`a_{p+2}`$, by its Laurent series
$$\dot{X}=(A_{p+2}/(ta_{p+2})+B+o(1))X$$
where $`B=_{j=1}^{p+1}\alpha _jA_j`$ (to be checked directly). The change of variables (local, at $`a_{p+2}`$) $`X`$diag$`((ta_{p+2})^1,\mathrm{},(ta_{p+2})^1,1,\mathrm{},1)X`$ ($`nm_0`$ units) brings system (7) to the form
$$\dot{X}=(A_{p+2}^{}/(ta_{p+2})+O(1))X$$
with $`A_{p+2}^{}=B^{}`$ where the restriction of the matrix $`B^{}`$ to the left lower $`((nm_0)\times m_0`$-block equals the one of $`B`$ to it and all other entries of $`B^{}`$ are 0. By Proposition 11, the monodromy operator $`M_{p+2}`$ is scalar if and only if $`B^{}=0`$.
The lemma is proved.
Definition. Call canonical change of the eigenvalues of the matrices $`A_j`$ (CCE) a change under which each eigenvalue changes by an integer, equal eigenvalues remain equal (hence, canonical or strongly generic eigenvalues remain such), the eigenvalues of $`A_{p+2}`$ (which are not canonical) do not change and the sum of all eigenvalues remains 0.
###### Proposition 20
Let for given Jordan normal forms of the matrices $`A_j`$ and given set $`\stackrel{~}{\lambda }`$ of generic eigenvalues there exist $`(p+2)`$-tuples from $`𝒮`$. Then for all generic eigenvalues sufficiently close to $`\stackrel{~}{\lambda }`$ there exist $`(p+2)`$-tuples from $`𝒮`$ with the same Jordan normal forms of the matrices $`A_j`$.
The proposition is proved in the next subsection. Its proof describes a way to construct $`(p+2)`$-tuples from $`𝒮`$ for nearby eigenvalues by deforming given $`(p+2)`$-tuples from $`𝒮`$ (with given eigenvalues). This way is called the modified basic technical tool.
Denote by $`𝐂^{}`$ the space of eigenvalues of the matrices $`A_1`$, $`\mathrm{}`$, $`A_{p+1}`$ when their Jordan normal forms are fixed.
###### Corollary 21
If for given Jordan normal forms of the matrices $`A_j`$ the set $`𝒮`$ is not empty, then for all eigenvalues from some Zariski open dense subset of $`𝐂^{}`$ there exist $`(p+2)`$-tuples from $`𝒮`$.
The corollary follows directly from the proposition.
Denote by $`\lambda ^0`$ a point from $`𝐂^{}`$ defining for $`jp+1`$ canonical eigenvalues. Consider all points from $`𝐂^{}`$ obtained from $`\lambda ^0`$ as a result of a CCE. Denote their set by $`\mathrm{\Sigma }(\lambda ^0)`$.
Choose $`\lambda ^0`$ such that
a) for $`jp+1`$ (one of) the eigenvalue(s) of $`A_j`$ of greatest multiplicity is integer; denote it by $`\lambda _j^0`$; all other eigenvalues are non-integer (for $`jp+1`$);
b) the only non-genericity relations modulo $`𝐙`$ satisfied by the eigenvalues $`\lambda _{k,j}`$ are of the form
$$k(\underset{j=1}{\overset{p+1}{}}\lambda _j^0)+\delta =0,\delta 𝐙$$
(the eigenvalues of $`A_{p+2}`$ are integer, so we include them in $`\delta `$; $`k𝐍`$ does not exceed the smallest of the multiplicities of the eigenvalues $`\lambda _j^0`$).
###### Corollary 22
If $`𝒮`$ is non-empty, then the set $`\mathrm{\Sigma }(\lambda ^0)`$ contains a point $`\lambda ^1`$ for which the sum $`_{j=1}^{p+1}\lambda _j^1`$ ($`\lambda _j^1`$ being an integer eigenvalue of $`A_j`$) is $`>1`$.
Proof:
$`1^0`$. Call a point from $`\mathrm{\Sigma }(\lambda ^0)`$ good (bad) if $`𝒮`$ projects on this point (if not). If a line $`l_1𝐂^{}`$ passing through two points from $`\mathrm{\Sigma }(\lambda ^0)`$ contains infinitely many bad points, then it contains only bad points from $`\mathrm{\Sigma }(\lambda ^0)`$ (see the proposition and the corollary and remember that $`𝒮`$ is constructible).
$`2^0`$. Suppose that no line parallel to $`l_1`$ and passing through two points from $`\mathrm{\Sigma }(\lambda ^0)`$ contains only finitely many bad points. Hence, $`\mathrm{\Sigma }(\lambda ^0)`$ must contain only bad points. The above proposition and corollary imply that $`𝒮`$ does not project on any line passing through two points of $`\mathrm{\Sigma }(\lambda ^0)`$. Hence, it does not project on any point of any affine subspace of $`𝐂^{}`$ of dimension $`k`$ passing through $`k`$ points from $`\mathrm{\Sigma }(\lambda ^0)`$ for $`k=2,\mathrm{}`$, dim$`𝐂^{}`$ (proved by induction on $`k`$). Hence, $`𝒮`$ must be empty – a contradiction.
$`3^0`$. This means that for every line $`l_1`$ passing through two points from $`\mathrm{\Sigma }(\lambda ^0)`$ there exists a line $`l_1^{}`$ parallel to it, passing through two points from $`\mathrm{\Sigma }(\lambda ^0)`$ and containing only finitely many bad points. Choose an index $`j`$ such that $`A_j`$ has at least two different eigenvalues ($`\lambda _j^{}`$ and $`\lambda _j^{\prime \prime }`$, of multiplicities $`m^{}`$ and $`m^{\prime \prime }`$); one of the two eigenvalues is the integer eigenvalue of $`A_j`$.
$`4^0`$. Denote by $`t_1`$, $`t_2`$ two points from $`𝐂^{}`$ such that the CCE which changes $`t_1`$ to $`t_2`$ is of the form $`\lambda _j^{}\lambda _j^{}m^{\prime \prime }`$, $`\lambda _j^{\prime \prime }\lambda _j^{\prime \prime }+m^{}`$ (all other eigenvalues remaining the same). Hence, there exists a line $`l`$ in $`𝐂^{}`$ parallel to the one passing through $`t_1`$ and $`t_2`$, also passing through two points from $`\mathrm{\Sigma }(\lambda ^0)`$ and such that $`l`$ contains only finitely many bad points. Hence, the line $`l`$ contains a point for which the sum $`_{j=1}^{p+1}\lambda _j^1`$ is a positive integer.
The corollary is proved.
### 3.4 The modified basic technical tool
The modified basic technical tool is used to prove the existence of deformations (depending analytically on $`\epsilon (𝐂,0)`$) of $`(p+2)`$-tuples $`A^{}𝒮`$ with trivial centralizers. It will be used in different contexts and we explain it here in one of them (namely, the proof of Proposition 20). The basic points in the reasoning in all other contexts will be the same, there will be differences only in the technical details.
Proof of Proposition 20:
$`1^0`$. Denote by $`A_j^0=Q_j^1G_jQ_j`$ the matrices from an irreducible $`(p+2)`$-tuple $`A^{}𝒮`$, $`G_j`$ being Jordan matrices. We look for matrices of the form $`A_{p+2}A_{p+2}^0=H`$ ($`H`$ was defined in Section 2), and for $`jp+1`$
$$A_j(\epsilon )=(I+\epsilon X_j(\epsilon ))^1Q_j^1(G_j+\epsilon L_j)Q_j(I+\epsilon X_j(\epsilon )),\epsilon (𝐂,0)$$
where $`L_j`$ are diagonal matrices which are polynomials of the semi-simple parts of the corresponding matrices $`G_j`$. Hence, $`[G_j,L_j]=0`$ and the matrices $`G_j`$ and $`G_j+\epsilon L_j`$ define one and the same Jordan normal form for $`\epsilon `$ small enough, see 4) – 5) from Theorem 14. We want conditions 1) – 5) from the previous subsection to hold.
Obviously, one has $`A_j=A_j^0+\epsilon ([A_j^0,X_j(0)]+Q_j^1L_jQ_j)+o(\epsilon )`$ for $`jp+1`$.
$`2^0`$. Conditions 2) and 3) yield in first approximation w.r.t. $`\epsilon `$ the following system of equations linear in the unknown variables the entries of the matrices $`X_j(0)`$:
$$\underset{j=1}{\overset{p+1}{}}[A_j^0,X_j(0)]=\underset{j=1}{\overset{p+1}{}}Q_j^1L_jQ_j,(\underset{j=1}{\overset{p+1}{}}\alpha _j[A_j^0,X_j(0)])|_{\kappa ,\nu }=(\underset{j=1}{\overset{p+1}{}}\alpha _jQ_j^1L_jQ_j)|_{\kappa ,\nu }$$
(10)
with $`\kappa =m_0+1,\mathrm{},n`$; $`\nu =1,\mathrm{},m_0`$; $`\alpha _j=1/(a_ja_{p+2})`$. The left hand-sides of these equations are linear forms in the entries of the matrices $`X_j(0)`$.
###### Lemma 23
These linear forms are linearly independent.
The proof of this lemma occupies the rest of the proof of the proposition. It implies the existence of $`X_j`$ analytic in $`\epsilon (𝐂,0)`$. Indeed, for $`\epsilon =0`$ equations (10) are solvable and the mapping
$$(X_1(0),\mathrm{},X_{p+1}(0))(\underset{j=1}{\overset{p+1}{}}[A_j^0,X_j(0)],(\underset{j=1}{\overset{p+1}{}}\alpha _j[A_j^0,X_j(0)])|_{\kappa ,\nu })$$
is surjective onto $`sl(n,𝐂)\times 𝐂^{(nm_0)m_0}`$. The existence of $`X_j`$ analytic in $`\epsilon `$ small enough follows from the implicit function theorem.
Proof of the lemma:
$`1^0`$. If the lemma were not true, then there should exist a couple of matrices $`(0,0)(V,W)sl(n,𝐂)^2`$ such that $`W_{i,j}=0`$ if $`i>m_0`$ or if $`jm_0`$ and
$$\mathrm{tr}(V(\underset{j=1}{\overset{p+1}{}}[A_j^0,X_j(0)])+W(\underset{j=1}{\overset{p+1}{}}\alpha _j[A_j^0,X_j(0)]))=0\mathrm{for}\mathrm{all}X_j(0),\mathrm{i}.\mathrm{e}.$$
$$\mathrm{tr}(([V,A_j^0]+\alpha _j[W,A_j^0])X_j(0))=0\mathrm{identically}\mathrm{in}\mathrm{the}\mathrm{entries}\mathrm{of}X_j(0),\mathrm{i}.\mathrm{e}.[V,A_j^0]+\alpha _j[W,A_j^0]=0.$$
Summing up the equalities for $`j=1,\mathrm{}p+1`$ and making use of $`_{j=1}^{p+1}A_j^0=A_{p+2}^0`$, one gets
$$[V,A_{p+2}^0]+[W,\underset{j=1}{\overset{p+1}{}}\alpha _jA_j^0]=0$$
(11)
The left lower $`m_0\times (nm_0)`$-block of the second summand in (11) is 0. Hence, so is the left lower block of the first summand, and, hence, the one of $`V`$ itself (remember that $`A_{p+2}^0=`$).
$`2^0`$. Choose $`\gamma 𝐂`$ such that the matrix $`V^{}=V+\gamma I`$ be non-degenerate. Hence, for all $`t𝐂\backslash \{a_{p+2}\}`$ the matrix $`V^{}+W/(ta_{p+2})`$ is non-degenerate, block upper-triangular and det$`(V^{}+W/(ta_{p+2}))`$det$`V^{}`$. Hence, its inverse is also block upper-triangular, with constant diagonal blocks and with constant non-zero determinant.
$`3^0`$. Consider the fuchsian system $`\dot{X}=A(t)X`$ with $`A(t)=(_{j=1}^{p+2}A_j^0/(ta_j))`$. Perform in it the change of variables $`X(V^{}+W/(ta_{p+2}))X`$:
$$A(t)(V^{}+W/(ta_{p+2}))^1(W/(ta_{p+2})^2)+(V^{}+W/(ta_{p+2}))^1A(t)(V^{}+W/(ta_{p+2}))$$
(gauge transformation). The matrix $`U(t)\stackrel{\mathrm{def}}{=}V^{}+W/(ta_{p+2})`$ being holomorphic and holomorphically invertible for $`ta_{p+2}`$, the system remains fuchsian at $`a_1`$, $`\mathrm{}`$, $`a_{p+1}`$ and has no singularities other than $`a_j`$.
The equalities $`[\gamma I+V,A_j^0]+\alpha _j[W,A_j^0]=0`$ are equivalent to $`[U(a_j),A_j^0]=0`$ and imply that its residua at $`a_j`$, $`jp+1`$, don’t change. (The residua of the new system equal $`(U(a_j))^1A_j^0U(a_j)`$.)
Check that the system remains fuchsian at $`a_{p+2}`$ as well. To this end observe that the matrix $`U^1=(V^{}(I+(V^{})^1W/(ta_{p+2}))^1`$ equals $`(V^{})^1(V^{})^1W(V^{})^1/(ta_{p+2})`$ because $`(I+(V^{})^1W/(ta_{p+2}))^1`$$`=I(V^{})^1W/(ta_{p+2})`$ due to $`((V^{})^1W)^2=0`$ (recall the block upper-triangular form of $`V^{}`$ and $`W`$). This implies that there are no polar terms of order higher than 1 at $`a_{p+2}`$ in the new system.
Indeed, a priori the matrix $`U^1\dot{U}+U^1A(t)U`$ (with $`U`$ and $`U^1`$ as above) can have at $`a_{p+2}`$ a pole of order $`3`$. Its coefficient before $`1/(ta_{p+2})^3`$ equals
$$((V^{})^1W)^2(V^{})^1W(V^{})^1A_{p+2}W$$
where each of the two summands is 0 (this follows from the form of the matrices $`V`$, $`W`$ and $`A_{p+2}`$). The one before $`1/(ta_{p+2})^2`$ equals
$$(V^{})^1W(V^{})^1W(V^{})^1A_{p+2}V^{}+(V^{})^1A_{p+2}W=(V^{})^1PV^{},$$
$$P=W(V^{})^1W(V^{})^1A_{p+2}+A_{p+2}W(V^{})^1$$
where $`P=0`$ for the same reason. Hence, the residuum at $`a_{p+2}`$ of the system also doesn’t change (because the sum of all residua is 0).
$`4^0`$. Hence, the solution $`UX`$ to the new system (which is the old one) equals $`XC`$, $`CGL(n,𝐂)`$ (some solution to the old system). Make the analytic continuations of both solutions along one and the same closed contour. Hence, $`UXM=XMC`$ where $`M`$ is the monodromy operator corresponding to the contour. But $`UXM=XCM`$ which implies $`[C,M]=0`$. This is true for any closed contour.
The monodromy group of the system being irreducible, the matrix $`C`$ must be scalar. Hence one has $`UXCX`$, i.e. $`UC`$ which implies $`V^{}=C`$, $`W=0`$. Recall that $`V^{}=V+\gamma I`$, $`Vsl(n,𝐂)`$. Hence, $`V=0`$.
The lemma is proved, the proposition as well.
## 4 Proof of the necessity
### 4.1 The proof in the case when $`\xi `$ is primitive
Given a $`(p+1)`$-tuple with a trivial centralizer, one can deform it into one also with a trivial centralizer, with relatively generic eigenvalues and with the same Jordan normal forms, therefore we presume the eigenvalues relatively generic.
We make use of Lemma 18 and Corollary 15 and consider only the case when all matrices $`A_j`$ and $`M_j`$ are diagonalizable. Denote by $`\mathrm{\Lambda }^n`$ the PMVs of their eigenvalues; they are defined by the Jordan normal forms $`J_j^n`$. Denote by $`\mathrm{\Lambda }^{n_\nu }`$ the PMVs defined by the Jordan normal forms $`J_j^{n_\nu }`$ (their definition is given before Theorem 1).
Consider the fuchsian system
$$\dot{X}=(\underset{j=1}{\overset{p+2}{}}A_j/(ta_j))X$$
(12)
where the $`(p+2)`$-tuple of matrices $`A_j`$ belongs to $`𝒮`$ and the eigenvalues of $`A_j`$ satisfy the conclusion of Corollary 22. Hence, the $`(p+2)`$-tuple of matrices $`A_j`$ is irreducible and the eigenvalues of its monodromy operators satisfy the only non-genericity relation
$$\sigma _1\mathrm{}\sigma _{p+2}=1(\gamma _0)$$
where $`\sigma _j=1`$ (notice that 1 is the only eigenvalue of $`M_{p+2}`$).
###### Lemma 24
The monodromy group of system (12) with eigenvalues defined as above has a trivial centralizer.
The eigenvalues of system (12) satisfy the conditions of Lemma 12 and Lemma 24 follows from Corollary 13.
###### Lemma 25
The monodromy group of system (12) with eigenvalues defined as in Lemma 24 can be conjugated to the form $`\left(\begin{array}{cc}\mathrm{\Phi }& \\ 0& I\end{array}\right)`$ where $`\mathrm{\Phi }`$ is $`n_1\times n_1`$.
Remark: Notice that the subrepresentation $`\mathrm{\Phi }`$ can be reducible.
###### Lemma 26
The centralizer $`Z(\mathrm{\Phi })`$ of the subrepresentation $`\mathrm{\Phi }`$ is trivial.
The last two lemmas are proved in Subsection 4.3. We prove them also in the case when $`\xi `$ is not primitive because we need this for the next subsection.
The subrepresentation $`\mathrm{\Phi }`$ being of dimension $`n_1<n`$, one can use induction on $`n`$ to prove the necessity. The induction base is the case when condition $`(\omega _n)`$ holds – in this case there is nothing to prove.
The PMV of the matrices $`M_j^{}`$ defining $`\mathrm{\Phi }`$ equals $`\mathrm{\Lambda }^{n_1}`$. It follows from Lemma 26 that for generic eigenvalues close to the ones of the matrices $`M_j^{}`$ defining $`\mathrm{\Phi }`$ there exist irreducible $`(p+1)`$-tuples of diagonalizable matrices $`\stackrel{~}{M}_j^{}GL(n_1,𝐂)`$ with PMV $`\mathrm{\Lambda }^{n_1}`$ and satisfying (1) (this can be proved by using the basic technical tool in the multiplicative version).
Hence, if $`\mathrm{\Lambda }^n`$ is good, then $`\mathrm{\Lambda }^{n_1}`$ is good. Condition $`(\omega _n)`$ doesn’t hold by assumption and conditions $`(\alpha _n)`$ and $`(\beta _n)`$ hold, see the Introduction. Finally, the PMV $`\mathrm{\Lambda }^{n_s}`$ is the same for $`\mathrm{\Lambda }^n`$ and for $`\mathrm{\Lambda }^{n_1}`$ (this follows from the definition of the PMVs $`\mathrm{\Lambda }^{n_\nu }`$). If $`\mathrm{\Lambda }^{n_1}`$ is good, then $`\mathrm{\Lambda }^{n_s}`$ satisfies condition $`(\omega _{n_s})`$ (one can’t have $`n_s=1`$, see the remark after Theorem 1). Hence, if the PMV $`\mathrm{\Lambda }^n`$ is good, then it satisfies conditions i) and ii) of Theorem 1, i.e. they are necessary.
The necessity is proved.
### 4.2 The proof in the case when $`\xi `$ is not primitive and in the additive version
$`1^0`$. If $`\xi `$ is not primitive, then the proof needs only small modifications. The eigenvalues of the matrices $`A_j`$ of system (12) are only relatively generic and satisfy only the non-genericity relation $`(\gamma ^{})`$ defined at the end of Subsection 1.2 (and its multiples).
$`2^0`$. Hence, the $`(p+1)`$-tuple of matrices-residua $`A_j`$ might be reducible. Suppose that it is in block upper-triangular form. Consider instead of the system its restriction to one of the diagonal blocks $`P`$; this restriction is presumed to be irreducible.
For this restriction Lemma 24 holds again. This follows from Corollary 13. The rest of the proof is the same because for all $`j`$ $`J_j^n`$ is a multiple of $`J(A_j|_P)`$. Recall that Lemmas 25 and 26 are proved also in the case when $`\xi `$ is not primitive.
$`3^0`$. In the additive version we proved the necessity in the case of generic eigenvalues in \[Ko1\]. If the eigenvalues are non-generic and if $`q=1`$, then given such an irreducible $`(p+1)`$-tuple of matrices $`A_j`$ one can deform it by means of the basic technical tool into a nearby one with generic eigenvalues and the same Jordan normal forms of the respective matrices. Thus the necessity is proved for $`q=1`$ in the additive version.
$`4^0`$. Consider the additive version with $`q>1`$. Given such an irreducible $`(p+1)`$-tuple of matrices $`A_j`$, one can multiply it by $`c𝐂^{}`$ to make the eigenvalues canonical. Next, by means of the basic technical tool one can deform it into a nearby irreducible one with the same Jordan normal forms of the matrices $`A_j`$ and with canonical relatively strongly generic eigenvalues. By Corollary 10, the monodromy group must be irreducible. By Proposition 11, for each $`j`$ one has $`J(A_j)=J(M_j)`$. Conditions i) and ii) are necessary in the multiplicative version, therefore they will be fulfilled in the additive one as well.
### 4.3 Proofs of Lemmas 25 and 26
Proof of Lemma 25:
We assume that condition $`(\omega _n)`$ does not hold (otherwise there is nothing to prove).
$`1^0`$. Consider first (in $`1^0`$$`5^0`$) the case when $`\xi `$ is a primitive root of unity. The monodromy group can be conjugated to a block upper-triangular form. The diagonal blocks define either irreducible or one-dimensional representations. The eigenvalues of each diagonal block $`1\times 1`$ satisfy the non-genericity relation $`(\gamma _0)`$; it is the only one satisfied by them due to the definition of the eigenvalues. This means that there is a single diagonal block of size $`>1`$.
$`2^0`$. The block in the right lower corner must be of size 1. Indeed, by Lemma 9 the left upper block can’t be of size 1 (because the corresponding sum of eigenvalues $`\lambda _{k,j}`$ is a positive integer). Hence, it must be the only block of size $`>1`$ and the matrices $`M_j`$ look like this:
$$M_j=\left(\begin{array}{cc}M_j^{}& L_j\\ 0& I\end{array}\right)(M)$$
where the size of $`M_j^{}`$ is $`n^{}\times n^{}`$ and the $`(p+1)`$-tuple of matrices $`M_j^{}`$ is irreducible.
$`3^0`$. The block $`M^{}`$ must be of size $`n_1`$. Indeed, if its size $`n^{}`$ is $`>n_1`$, then we show that the columns of the $`(p+1)`$-tuples of matrices $`L_j`$ aren’t linearly independent modulo the space $`𝒲`$ defined below which will imply the non-triviality of the centralizer of the monodromy group.
Denote by $`𝒲𝐂^n^{}`$ the space of $`(p+1)`$-tuples of vector-columns of the form $`(M_jI)X`$, $`X𝐂^n^{}`$. These are the vector-columns (right upper blocks) obtained by conjugating the $`(p+1)`$-tuple of matrices $`\left(\begin{array}{cc}M_j^{}& 0\\ 0& 1\end{array}\right)`$ by $`\left(\begin{array}{cc}I& X\\ 0& 1\end{array}\right)`$.
One has dim$`𝒲=n^{}`$. Indeed, if dim$`𝒲<n^{}`$, then the images of the linear operators $`X(M_jI)X`$ belong to a proper subspace of $`𝐂^n^{}`$. This subspace can be assumed to belong to the space spanned by the first $`n^{}1`$ vectors of the canonical basis of $`𝐂^n^{}`$ (which can be achieved by conjugating the monodromy operators $`M_j`$ by a block-diagonal matrix with diagonal blocks of sizes $`n^{}`$ and $`nn^{}`$, the latter equal to $`I`$).
But then the matrices $`M_j^{}`$ will be of the form $`\left(\begin{array}{cc}M_j^{\prime \prime }& \\ 0& 1\end{array}\right)`$ which contradicts the irreducibility of the matrix group generated by $`M_1^{}`$, $`\mathrm{}`$, $`M_{p+1}^{}`$.
$`4^0`$. The $`(p+1)`$-tuples of columns of the blocks $`L_j`$ belong to a subspace of $`𝐂^{(p+1)n^{}}`$ of dimension $`\mathrm{\Delta }:=r_1+\mathrm{}+r_{p+1}n^{}`$ (each column of $`L_j`$ belongs to a space of dimension $`r_j`$ and there are $`n^{}`$ linear conditions satisfied by these columns; these conditions result from (1) and look like this: $`_{j=1}^{p+1}M_1^{}\mathrm{}M_{j1}^{}L_j=0`$; they are linearly independent because the change of variables $`L_j^{}=M_1^{}\mathrm{}M_{j1}^{}L_j`$ transforms them to $`_{j=1}^{p+1}L_j^{}=0`$ and the independence in this form is evident).
$`5^0`$. The columns of the block $`L`$ must be linearly independent modulo the space $`𝒲`$, otherwise the monodromy group will be a direct sum of the form $`M_j=\left(\begin{array}{cc}M_j^{\prime \prime \prime }& 0\\ 0& 1\end{array}\right)`$, $`M_j^{\prime \prime \prime }GL(n1,𝐂)`$. Hence, $`\mathrm{\Delta }(nn^{})+n^{}`$ (because the block $`L`$ has $`nn^{}`$ columns and dim$`𝒲=n^{}`$). Recall that $`r_1+\mathrm{}+r_{p+1}=n+n_1`$. Hence, $`n+n_1n^{}nn^{}+n^{}`$, i.e. $`n^{}n_1`$.
$`6^0`$. Let now $`\xi `$ be a non-primitive root. The diagonal blocks of the monodromy group can be of two types. The first are of size 1, the eigenvalues satisfying the non-genericity relation $`(\gamma _0)`$.
Describe the second type of diagonal blocks. Their sizes are $`>1`$ and can be different. Define the unitary set of eigenvalues: for each $`j`$ divide by $`(m_0,q)`$ the multiplicities of all eigenvalues $`\sigma _{k,j}`$ of the ones that are equal among themselves and are $`1`$. A block $`F`$ of the second type contains $`h`$ times the unitary set, $`1h(m_0,q)`$, and a certain number of eigenvalues equal to 1. (To different matrices $`M_j`$ there correspond, in general, different numbers of eigenvalues from the unitary set; therefore one must, in general, add some number of eigenvalues 1 for some values of $`j`$ to make the number of eigenvalues of the restrictions of the matrices $`M_j`$ to $`F`$ equal; one then could eventually add one and the same number of eigenvalues equal to 1 to all matrices $`M_j|_F`$.)
The eigenvalues of the blocks of the second type satisfy the non-genericity relation $`(\gamma ^{})`$ (defined in Subsection 1.2) and eventually $`(\gamma _0)`$ as well.
$`7^0`$. Denote by $`\kappa (F)`$ the ratio ”number of eigenvalues $`\sigma _{k,j}`$ equal to 1”/”number of eigenvalues $`\sigma _{k,j}`$ not equal to 1” (eigenvalues of the restriction of the monodromy group to $`F`$), and by $`\kappa _0`$ the same ratio computed for the entire matrices $`M_j`$ (in both ratios one takes into account the eigenvalues of all matrices $`M_j`$). Then one must have $`\kappa (F)<\kappa _0`$.
Indeed, one can’t have $`\kappa (F)\kappa _0`$ because condition $`(\omega _n)`$ does not hold, hence, the restriction of the monodromy group to $`F`$ wouldn’t satisfy this condition either. In the presence of the non-genericity relation $`(\gamma _0)`$ this implies a contradiction with the following
###### Lemma 27
(see \[Ko1\]). The following condition is necessary for the existence of irreducible $`(p+1)`$-tuples of matrices $`M_j`$ from the conjugacy classes $`C_j`$ and satisfying (1):
$$\underset{b_1,\mathrm{},b_{p+1}𝐂^{},b_1\mathrm{}b_{p+1}=1}{\mathrm{min}}(\mathrm{rk}(b_1M_1I)+\mathrm{}+\mathrm{rk}(b_{p+1}M_{p+1}I))2n$$
But then the sum of the eigenvalues $`\lambda _{k,j}`$ corresponding to the eigenvalues $`\sigma _{k,j}`$ from $`F`$ will be negative. If the block $`F`$ is to be in the right lower corner, then this sum must be positive (Lemma 9; it can’t be 0 because the only non-genericity relation satisfied by the eigenvalues $`\lambda _{k,j}`$ is $`(\gamma ^{})`$ and its multiples, see Subsection 1.2; if the sum is 0, then by Corollary 10 the $`(p+1)`$-tuple of matrices-residua would be block upper-triangular up to conjugacy – a contradiction). Hence, the right lower block is of size 1.
$`8^0`$. Denote by $`\mathrm{\Pi }`$ the left upper $`(n1)\times (n1)`$-block. Conjugate it to make all non-zero rows of the restriction of the $`(p+1)`$-tuple $`\stackrel{~}{M}`$ of matrices $`M_jI`$ to $`\mathrm{\Pi }`$ linearly independent. After the conjugation some of the rows of the restriction of $`\stackrel{~}{M}`$ to $`\mathrm{\Pi }`$ might be 0. In this case conjugate the matrices $`M_j`$ by one and the same permutation matrix which places the zero rows of $`M_jI`$ in the last (say, $`nn^{}`$) positions (recall that the last row of $`M_jI`$ is 0, so $`nn^{}1`$). Notice that if the restriction to $`\mathrm{\Pi }`$ of a row of $`M_jI`$ is zero, then the $`n`$-th position of the row is 0 as well, otherwise $`M_j`$ is not diagonalizable.
$`9^0`$. After this conjugation the monodromy matrices have the form (M) from $`2^0`$, the matrix group generated by the matrices $`M_j^{}`$ is not necessarily irreducible, but can be conjugated to a block upper-triangular form, its restrictions to the diagonal blocks (all of sizes $`>1`$) being irreducible. Hence, the mapping
$$(X_1,\mathrm{},X_{p+1})\underset{j=1}{\overset{p+1}{}}(M_j^{}I)X_j,X_j𝐂^n^{}$$
(13)
is surjective onto $`𝐂^n^{}`$. Indeed, the matrix algebra $``$ generated by the matrices $`M_j^{}I`$ is block upper-triangular, its restrictions to each diagonal block (say, of size $`u`$) is irreducible and, hence, is $`gl(u,𝐂)`$ (the Burnside theorem). Thus the algebra contains a non-degenerate matrix $`L`$.
One has $`L=_{j=1}^{p+1}(M_j^{}I)H_j`$, $`H_j`$. For every vector-column $`X𝐂^n^{}`$ there exists a unique $`Y𝐂^n^{}`$ such that $`X=LY`$. Hence, one can set $`X_j=H_jY`$ which proves the surjectivity of the mapping.
$`10^0`$. If one defines the space $`𝒲`$ as above, then one finds that dim$`𝒲=n^{}`$. This is proved like in the case when $`\xi `$ is a primitive root, see $`3^0`$, but the form $`\left(\begin{array}{cc}M_j^{\prime \prime }& \\ 0& 1\end{array}\right)`$ of the matrices $`M_j^{}`$ is forbidden not because the group generated by them must be irreducible (which, in general, is not true) but just because by definition there are no diagonal blocks of size 1. For the rest the proof goes on like in the case when $`\xi `$ is primitive.
This proves the lemma.
Proof of Lemma 26:
$`1^0`$. Consider first the case when $`\xi `$ is a primitive root (in $`1^0`$$`4^0`$). If the lemma is not true, then $`Z(\mathrm{\Phi })`$ either A) contains a diagonalizable matrix $`D`$ with exactly two distinct eigenvalues or B) it contains a nilpotent matrix $`N`$ with $`N^2=0`$, see $`1^0`$ of the proof of Corollary 13.
$`2^0`$. In case A) one conjugates the monodromy group to the form $`\left(\begin{array}{cc}G_j& L_j\\ 0& I\end{array}\right)`$ with $`G_j=\left(\begin{array}{cc}M_j^1& 0\\ 0& M_j^2\end{array}\right)`$. The sizes of $`M_j^1`$, $`M_j^2`$ equal the multiplicities of the two eigenvalues of $`D`$ and one has $`D=\left(\begin{array}{cc}\alpha I& 0\\ 0& \beta I\end{array}\right)`$, $`\alpha \beta `$, $`D`$ and $`G_j`$ are $`n_1\times n_1`$.
At least one of the blocks $`M_j^i`$ must equal $`I`$ (for all $`j`$) because there is a single diagonal block of size $`>1`$. But this would mean that the monodromy group is a direct sum. Indeed, if $`M_j^2=I`$ for all $`j`$, then in the rows of the block $`M_j^2`$ and in the last columns (the ones of the block $`I`$) the entries of $`M_j`$ must be 0, otherwise $`M_j`$ is not diagonalizable. Being a direct sum contradicts Lemma 24.
$`3^0`$. In case B) one can conjugate $`N`$ to the form $`N=\left(\begin{array}{ccc}0& 0& I\\ 0& 0& 0\\ 0& 0& 0\end{array}\right)`$ (or $`N=\left(\begin{array}{cc}0& I\\ 0& 0\end{array}\right)`$) with $`I`$ being $`v\times v`$, $`vn_1/2`$; the second case corresponds to $`v=n_1/2`$.
Hence, $`M_j=\left(\begin{array}{cccc}M_j^1& R_j& T_j& Q_j\\ 0& M_j^2& S_j& H_j\\ 0& 0& M_j^1& P_j\\ 0& 0& 0& I\end{array}\right)`$ where $`M_j^1`$ is $`v\times v`$ and if $`v=n_1/2`$, then the blocks $`R_j`$, $`M_j^2`$, $`S_j`$ and $`H_j`$ are absent.
$`4^0`$. Like in case A) one shows that $`M_j^1=I`$ (this implies that in fact the possibility $`v=n_1/2`$ does not exist because there would be no diagonal block of size $`>1`$ at all). This means that the matrix with a single non-zero entry in position $`(1,n)`$ belongs to the centralizer of the monodromy group which contradicts Lemma 24.
$`5^0`$. Let $`\xi `$ not be a primitive root. Suppose that we are in case A). We showed in the proof of the previous lemma (see $`4^0`$ of its proof) that the $`(p+1)`$-tuple of columns of the blocks $`L_j`$ (see $`2^0`$ of the present proof) belong to a subspace of $`𝐂^{(p+1)n_1}`$ (denote it by $`𝒱`$) of dimension $`\mathrm{\Delta }=n+n_1n_1=n`$. When this subspace is factorized by the space $`𝒲`$ (see $`3^0`$ of the proof of the previous lemma), then the dimension becomes $`nn_1`$.
On the other hand, there are $`nn_1`$ columns of the blocks $`L_j`$. In case A) the space $`𝒱/𝒲`$ splits into a direct sum of two such spaces defined for each of the blocks $`M_j^1`$ and $`M_j^2`$. The sum of their dimensions equals $`nn_1`$, the number of columns of $`L_j`$; this implies that one can conjugate the matrices $`M_j`$ by $`GGL(n,𝐂)`$ to the form $`M_j=\left(\begin{array}{cccc}M_j^1& 0& L_j^{}& 0\\ 0& M_j^2& 0& L_j^{\prime \prime }\\ 0& 0& I& 0\\ 0& 0& 0& I\end{array}\right)`$ which means that the monodromy group is a direct sum (one makes a self-evident permutation of the rows and columns which results from conjugation to achieve a block-diagonal form of the matrices). The matrix $`G`$ is block-diagonal, with diagonal blocks of sizes $`n_1`$ and $`nn_1`$, the former equal to $`I`$.
This is a contradiction with Lemma 24.
$`6^0`$. In case B) a conjugation with a matrix $`G^{}`$ (defined like $`G`$ in $`5^0`$) brings the block $`\left(\begin{array}{c}Q_j\\ H_j\\ P_j\end{array}\right)`$ to the form $`\left(\begin{array}{ccc}U_j^1& 0& 0\\ 0& U_j^2& 0\\ 0& 0& U_j^1\end{array}\right)`$ with heights of the blocks $`U_j^1`$, $`U_j^2`$ the same as the ones of $`M_j^1`$, $`M_j^2`$.
This means that the matrix $`\left(\begin{array}{cccccc}0& 0& I& 0& 0& 0\\ 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& I\\ 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0\end{array}\right)`$ belongs to the centralizer of the monodromy group. This is again a contradiction with Lemma 24.
The lemma is proved.
## 5 Proof of the sufficiency
### 5.1 Plan of the proof
$`1^0`$. Two cases are possible:
Case A) $`d_1+\mathrm{}+d_{p+1}2n^2+2`$
Case B) $`d_1+\mathrm{}+d_{p+1}=2n^2`$
The condition $`d>1`$ excludes the possibility $`d_1+\mathrm{}+d_{p+1}=2n^22`$, see the remarks after Theorem 1. In case B) $`\xi `$ is presumed primitive.
In case A) we construct $`(p+1)`$-tuples of nilpotent matrices $`A_j`$ with trivial centralizers where for each $`j`$ $`J(A_j)`$ corresponds to the necessary Jordan normal form $`J(M_j)`$ of the monodromy operator $`M_j`$, see Lemma 28, part I). Such a construction was already carried out in \[Ko1\].
$`2^0`$. After this we deform the $`(p+1)`$-tuple into a nearby $`(p+2)`$-tuple of matrices $`\stackrel{~}{A}_j`$ with $`J(\stackrel{~}{A}_j)=J(M_j)`$ for $`jp+1`$, with $`A_{p+2}=\epsilon `$ and with
$$\stackrel{~}{A}_j=(I+\epsilon X_j(\epsilon ))^1Q_j^1(G_j+\epsilon L_j)Q_j(I+\epsilon X_j(\epsilon ))$$
for $`j=1,\mathrm{},p+1`$ so that the matrix $`G_j+\epsilon L_j`$ be conjugate to $`\epsilon L_j`$ (see 6) of Theorem 14). We do this like in the modified basic technical tool. The following condition must hold:
$$(\underset{j=1}{\overset{p+1}{}}\stackrel{~}{A}_j/(a_ja_{p+2}))_{\mu ,\nu }=0,\mu =m_0+1,\mathrm{},n,\nu =1,\mathrm{},m_0.$$
(14)
$`3^0`$. Then one multiplies the $`(p+2)`$-tuple by $`1/\epsilon `$; hence, the difference between the two eigenvalues of $`\stackrel{~}{A}_{p+2}`$ becomes equal to 1. Condition (14) implies that the singularity at $`a_{p+2}`$ of the fuchsian system with residua $`\stackrel{~}{A}_j`$ will be apparent (i.e. with local monodromy equal to $`I`$).
$`4^0`$. In case B) one constructs $`(p+2)`$-tuples of nilpotent matrices $`A_j`$ with trivial centralizers, see Lemma 28, part II). The Jordan normal form of the matrix $`A_{p+2}`$ has $`m_0`$ blocks of size 2 and $`n2m_0`$ ones of size 1. One sets $`\stackrel{~}{A}_{p+2}=A_{p+2}+\epsilon `$, the other matrices $`\stackrel{~}{A}_j`$ are defined like in case A). The matrix $`A_{p+2}`$ is such that for $`\epsilon 0`$ the matrix $`\stackrel{~}{A}_{p+2}`$ is conjugate to $``$. One has $`\stackrel{~}{A}_{p+2}|_{\epsilon =0}0`$ to make possible the construction of a $`(p+2)`$-tuple of matrices $`A_j`$ with a trivial centralizer.
Like in case A) one multiplies the residua $`\stackrel{~}{A}_j`$ by $`1/\epsilon `$. Condition (14) holds in case B) as well and the fuchsian system obtained after this multiplication has an apparent singularity at $`a_{p+2}`$.
$`5^0`$. The deformation of $`A_j`$ into $`\stackrel{~}{A}_j`$ is possible to be done when certain transversality conditions make the theorem of the implicit function applicable.
### 5.2 The basic lemma
Denote by $`\stackrel{~}{J}_j^n`$ Jordan normal forms ($`j=1,\mathrm{},p+1`$) satisfying conditions i) and ii) of Theorem 1. Denote by $`J_j^n`$ both their corresponding Jordan normal forms with a single eigenvalue and the nilpotent conjugacy classes defining these Jordan normal forms.
###### Lemma 28
Let $`d>1`$. Let the $`p+1`$ Jordan normal forms $`J_j^n`$ satisfy conditions i) and ii) of Theorem 1.
$`I)`$ Then in case A) there exists a $`(p+1)`$-tuple of nilpotent matrices $`A_j`$ satisfying (2) such that
1) for $`j=1,\mathrm{},p+1`$ the matrix $`A_j`$ belongs either to $`J_j^n`$ or to its closure;
2) the centralizer $`𝒵`$ of the $`(p+1)`$-tuple is trivial;
3) the $`(p+1)`$-tuple is in block upper-triangular form, the diagonal blocks being of sizes $`n_s`$, $`n_{s1}n_s`$, $`\mathrm{}`$, $`nn_1`$; the restriction of the $`(p+1)`$-tuple only to the first of them is non-zero;
4) the restriction to the diagonal block of size $`n_s`$ of the matrix $`B:=_{j=1}^{p+1}\alpha _jA_j`$ is in Jordan normal form and its first $`d`$ eigenvalues are non-zero and simple;
5) the restrictions to this block of the matrices $`A_j`$ define an irreducible representation.
$`II)`$ In case B) there exists a $`(p+2)`$-tuple of nilpotent matrices $`A_j`$ satisfying (2) such that 1), 2) and 3) hold and
6) the Jordan normal form of the matrix $`A_{p+2}`$ consists of $`m_0`$ blocks of size 2 and of $`n2m_0`$ blocks of size 1; its non-zero entries are all in the diagonal block of size $`n_s`$;
7) the restrictions to the diagonal block of size $`n_s`$ of the matrices $`A_j`$ are themselves block upper-triangular (and define a representation with a trivial centralizer), with diagonal blocks of equal size (which is 2, 3, 4 or 6) defining non-equivalent irreducible representations;
8) the restriction to the diagonal block of size $`n_s`$ of the matrix $`B`$ is upper-triangular, with distinct non-zero eigenvalues.
Remark: The Jordan normal forms $`J_j^{n_s}`$ from case B) correspond to a triple or quadruple of matrices from the four so-called special cases, see Subsection 6.1.
###### Corollary 29
Let the $`(p+1)`$-tuple of conjugacy classes $`C_jGL(n,𝐂)`$ define Jordan normal forms $`J_j^n`$ which satisfy conditions i) and ii) of Theorem 1. Let $`d>1`$ and let in case B) $`\xi `$ be primitive. Then there exist $`(p+1)`$-tuples of matrices $`M_jC_j`$ satisfying (1) and with trivial centralizers.
Proof:
$`1^0`$. Construct $`(p+1)`$\- or $`(p+2)`$-tuples of nilpotent matrices $`A_j^0`$ like in the lemma where for $`j=1,\mathrm{},p+1`$ $`A_j^0`$ is nilpotent and $`J(A_j^0)=J_j^n`$. Set $`A_j^0=Q_j^1G_jQ_j`$ where $`G_j`$ are nilpotent Jordan matrices. Then construct matrices
$$\stackrel{~}{A}_j=(I+\epsilon X_j(\epsilon ))^1Q_j^1(G_j+\epsilon L_j)Q_j(I+\epsilon X_j(\epsilon ))$$
such that for $`j=1,\mathrm{},p+1`$ the matrix $`G_j+\epsilon L_j`$ be conjugate to $`\epsilon L_j`$ (see 6) of Theorem 14).
$`2^0`$. Choose the eigenvalues of the matrices $`L_j`$ to be canonical for $`jp+1`$ and to have $`\mathrm{exp}(2\pi iL_j)C_j`$ (see Proposition 11). Set $`L_{p+2}=`$.
###### Lemma 30
There exist matrices $`X_j`$ analytic in $`\epsilon (𝐂,0)`$ such that
a) $`X_{p+2}0`$;
b) $`\stackrel{~}{A}_1+\mathrm{}+\stackrel{~}{A}_{p+2}=0`$;
c) $`(_{j=1}^{p+1}\stackrel{~}{A}_j/(a_ja_{p+2}))_{\mu ,\nu }=0`$ for $`\mu =m_0+1,n`$, $`\nu =1,\mathrm{},m_0`$.
The lemma is proved in the next subsection.
$`3^0`$. Fix $`\epsilon 0`$. Multiply the matrices $`\stackrel{~}{A}_j`$ by $`1/\epsilon `$. Conditions a) and c) imply that the singularity at $`a_{p+2}`$ of the fuchsian system with poles $`a_j`$ and residua $`\stackrel{~}{A}_j`$ is apparent. Hence, the monodromy operators have the necessary eigenvalues and Jordan normal forms.
Let $`\xi `$ be primitive. Choose the eigenvalues of the matrices $`A_j`$ generic and satisfying the conditions of Corollary 13. Hence, the centralizer of the monodromy group is trivial. If in case A) $`\xi `$ is not primitive, then the existence of representations with trivial centralizers follows from Lemma 6.
The corollary is proved.
### 5.3 Proof of Lemma 30
The reader is supposed to have read and understood the proof of Lemma 32 to which we refer.
$`1^0`$. Like in the proof of Proposition 20 one finds out that the existence of the necessary matrices $`X_j`$ follows from the solvability of the system of linear equations (10). So one has to prove Lemma 23. Notice that this time one has $`A_{p+2}=0`$, unlike in the conditions of Lemma 23, so the lemma has to be reproved. This is what we do.
$`2^0`$. In case A) equation (11) is reduced to $`[W,_{j=1}^{p+1}\alpha _jA_j^0]=0`$ which implies $`W=0`$ (recall condition 4) from Lemma 28 and the fact that only the entries in positions $`(\kappa ,\nu )`$ of $`W`$ can be non-zero with $`\kappa =1,\mathrm{},m_0<d`$, $`\nu =m_0+1,\mathrm{},n`$).
Hence, $`[V,A_j^0]=0`$ for $`j=1,\mathrm{},p+1`$ which implies $`V=0`$ (recall condition 2) from Lemma 28).
$`3^0`$. Consider case B). Denote by $`H_{\mu ,\nu }`$ the block in position $`(\mu ,\nu )`$ when the matrices from $`gl(n,𝐂)`$ are block-decomposed, with sizes of the diagonal blocks equal to $`n_s`$, $`n_{s1}n_s`$, $`n_{s2}n_{s1}`$, $`\mathrm{}`$, $`nn_1`$. Recall that $`m_0<d<n_s`$.
Equation (11) implies that $`V_{H_{i,1}}=0`$ for $`i>1`$. This follows from conditions 3) and 8) of Lemma 28 and from $`W_{H_{i,1}}=0`$ for $`i>1`$. The same equation implies that the restriction of $`W`$ only to $`H_{1,1}`$ can be non-zero. Indeed, one has $`[V,A_{p+2}^0]_{H_{1,i}}=0`$ for $`i>1`$, hence, $`[W,_{j=1}^{p+1}\alpha _jA_j^0]_{H_{1,i}}=0`$. Conditions 3) and 8) of Lemma 28 imply that $`W_{H_{1,i}}=0`$. On the other hand, for $`j>1`$ one has $`W_{H_{j,i}}=0`$.
$`4^0`$. We prove (in $`5^0`$$`8^0`$) that $`V_{H_{1,1}}=W_{H_{1,1}}=0`$. By $`3^0`$, this would imply $`W=0`$ and the equations $`[V,A_j^0]=0`$, $`j=1,\mathrm{},p+1`$ would imply $`V=0`$ (recall condition 2) of Lemma 28). Hence, $`V=W=0`$ which proves Lemma 23. As we know that $`V_{H_{i,1}}=W_{H_{i,1}}=0`$ for $`i>1`$, we can replace in the equations
$$[V+\alpha _jW,A_j^0]=0$$
(15)
the matrices $`V`$, $`W`$ and $`A_j^0`$ by their restrictions to $`H_{1,1}`$. This is what we do.
$`5^0`$. Block-decompose the matrices from $`gl(n_s,𝐂)`$, all diagonal blocks being of size 2,3,4 or 6 (recall that the Jordan normal forms $`J_j^{n_s}`$ correspond to one of the special cases from Subsection 6.1). Denote the blocks by $`U_{\mu ,\nu }`$. The form of the matrices $`A_{p+2}`$ and $`W`$ implies that the restriction of equation (15) to $`U_{i,k}`$ with $`ik`$ looks like this:
$$V|_{U_{i,k}}A_j|_{U_{k,k}}A_j|_{U_{i,i}}V|_{U_{i,k}}=0$$
Hence, if $`i>k`$, then $`V|_{U_{i,k}}=0`$, if $`i=k`$, then $`V|_{U_{i,k}}=\gamma _iI`$. This follows from the non-equivalence of the irreducible representations defined by the matrices $`A_j|_{U_{k,k}}`$ and $`A_j|_{U_{i,i}}`$ and from Schur’s lemma.
$`6^0`$. The reader should have read the proof of Lemma 32 to which we refer. In this proof the blocks $`F^{}`$, $`F^{\prime \prime }`$, $`F`$ and $`{}_{}{}^{t}F`$ were defined. Like in the proof of Lemma 32 one shows that $`V_{U_{\mu ,\nu }}=0`$ if $`\mu \nu `$, $`U_{\mu ,\nu }F`$, and $`V_{U_{\mu ,\nu }}=\gamma _\mu I`$ if $`\mu =\nu `$. Let $`U^{}=U_{\mu ,\nu }F`$. Then one has
$$A_j|_{U_{\mu ,\mu }}(V+\alpha _jW)|_U^{}(V+\alpha _jW)|_U^{}A_j|_{U_{\nu ,\nu }}+(\gamma _\mu \gamma _\nu )A_j|_U^{}=0$$
(16)
Consider the fuchsian system with poles at $`a_1`$, $`\mathrm{}`$ ,$`a_{p+1}`$ and residua $`A_j^0`$. Perform in it the change of variables
$$X\stackrel{~}{W}X,\stackrel{~}{W}=\delta I+V+W/(ta_{p+2})$$
(17)
where $`\delta 𝐂`$ is chosen such that $`\mathrm{\Delta }:=`$det$`\stackrel{~}{W}`$ be non-zero (notice that $`\mathrm{\Delta }`$ does not depend on $`t`$ due to the form of $`V`$ and $`W`$). The change (17) transforms the residua as follows: $`A_j^0A_j^1=\stackrel{~}{W}(a_j)^1A_j^0\stackrel{~}{W}(a_j)`$, $`j=1,\mathrm{},p+1`$. At $`a_{p+2}`$ a new singular point appears whose residuum is 0 (from the polar part only the term $`O(1/(ta_{p+2})^2)`$ is non-zero).
$`7^0`$. The block $`U^{}`$ of the residuum $`A_j^1`$ equals up to a non-zero factor $`A_j|_{U_{\mu ,\mu }}(V+\alpha _jW)|_U^{}(V+\alpha _jW)|_U^{}A_j|_{U_{\nu ,\nu }}`$. This non-zero factor depends on $`\delta `$. If $`\gamma _\mu \gamma _\nu `$, then equation (16) shows that the blocks $`A_j|_U^{}`$ are obtained as a result of the change (17) in the fuchsian system with residua $`A_j^0`$.
This, however, is impossible because the sum of the residua of a meromorphic 1-form on $`𝐂P^1`$ is 0, so one should have $`_{j=1}^{p+1}A_j|_U^{}=0`$; on the other hand, one has $`_{j=1}^{p+1}A_j|_U^{}=A_{p+2}|_U^{}0`$ by construction (see the proof of Lemma 32) – a contradiction. Hence, one has $`\gamma _\mu =\gamma _\nu `$ for all $`(\mu ,\nu )`$. As $`Vsl(n_s,𝐂)`$, one must have (for all $`i`$) $`\gamma _i=0`$. Thus only the entries of $`V`$ belonging to the block $`F`$ can be non-zero.
$`8^0`$. Sum up equations (16) from $`1`$ to $`p+1`$. As $`_{j=1}^{p+1}A_j^0=0`$ and $`\alpha _jA_j=B`$, one gets $`[A_{p+2},V]+[B,W]=0`$. The form of the matrices $`A_{p+2}`$ and $`V`$ implies that $`[A_{p+2},V]=0`$. The matrix $`B`$ is upper-triangular and has distinct eigenvalues. This and the form of $`W`$ implies $`W=0`$.
But then one has $`[A_j,V]=0`$ for all $`j`$. Hence, $`A_j|_{U_{\mu ,\mu }}V|_U^{}V|_U^{}A_j|_{U_{\nu ,\nu }}=0`$. The $`(p+1)`$-tuples of matrices $`A_j|_{U_{\mu ,\mu }}`$ and $`A_j|_{U_{\nu ,\nu }}`$ defining non-equivalent irreducible representations, this implies $`V=0`$.
The lemma is proved.
### 5.4 Proof of Lemma 28
$`1^0`$. Block decompose the matrices from $`gl(n,𝐂)`$, the diagonal blocks being square, of sizes $`n_s`$, $`n_{s1}n_s`$, $`n_{s2}n_{s1}`$, $`\mathrm{}`$, $`nn_1`$. Recall that we denote by $`H_{\mu ,\nu }`$ the block in position $`(\mu ,\nu )`$ (it is of size $`(n_{s\mu +1}n_{s\mu +2})\times (n_{s\nu +1}n_{s\nu +2})`$; we set $`n_{s+1}=0`$). Denote by $`L_k`$ the left upper block $`n_{sk+1}\times n_{sk+1}`$ (one has $`L_1=H_{1,1}`$).
$`2^0`$. Construct the matrices $`A_j|_{L_1}`$. For $`j=1,\mathrm{},p+1`$ one has $`J(A_j|_{L_1})=J_j^{n_s}`$, for $`j=p+2`$ the Jordan normal form of $`A_{p+2}|_{L_1}`$ consists of $`m_0`$ blocks of size 2 and of $`n_s2m_0`$ ones of size 1; the sizes $`n_\nu `$ being divisible by $`d>m_0`$, one has $`n_s2d>2m_0`$ (one can’t have $`n_s=d`$ because this would mean that the matrices $`A_j|_{L_1}`$ are scalar – a contradiction with condition $`(\omega _{n_s})`$).
###### Lemma 31
In Case A) there exists an irreducible $`(p+1)`$-tuple of matrices $`A_j^s:=A_j|_{L_1}`$ satisfying condition (14), with zero sum and with $`J(A_j^s)=J_j^{n_s}`$ for $`j=1,\mathrm{},p+1`$. The matrix $`B^s:=_{j=1}^{p+1}\alpha _jA_j^s`$ is in Jordan normal form, its first $`d`$ eigenvalues are simple and non-zero.
The lemma results from Theorem 34, see Section 6.
###### Lemma 32
In Case B) there exists a $`(p+2)`$-tuple of matrices $`A_j^s:=A_j|_{L_1}`$ satisfying condition (14), with zero sum, with trivial centralizer and with $`J(A_j^s)=J_j^{n_s}`$ for $`j=1,\mathrm{},p+1`$; $`J(A_{p+2}^s)`$ consists of $`m_0`$ blocks of size 2 and of $`n_s2m_0`$ blocks of size 1 and the matrix algebra generated by $`A_1^s`$, $`\mathrm{}`$, $`A_{p+2}^s`$ contains a non-degenerate matrix. The matrix $`B^s:=_{j=1}^{p+1}\alpha _jA_j^s`$ is upper-triangular and has distinct non-zero eigenvalues.
The lemma is proved in the next subsection.
$`3^0`$. Suppose that the matrices $`A_j^{sk+1}:=A_j|_{L_k}`$ (whose sum is 0) are constructed such that
a) for $`j=1,\mathrm{},p+1`$ one has $`A_j^{sk+1}J_j^{n_{sk+1}}`$ or $`A_j^{sk+1}`$ belongs to the closure of $`J_j^{n_{sk+1}}`$;
b) in case B) $`J(A_{p+2})`$ consists of $`m_0`$ blocks of size 2 and of $`n_{sk+1}2m_0`$ blocks of size 1; in case A) $`A_{p+2}=0`$;
c) the matrices $`A_j^{sk+1}`$ are block upper-triangular, with $`A_j^{sk+1}|_{H_{\mu ,\nu }}=0`$ for $`\mu >\nu `$ and for $`\mu =\nu >1`$;
d) the columns of the $`(p+1)`$-tuple of matrices $`A_j^{sk+1}`$ $`jp+1`$ are linearly independent.
The last conditions means that if $`(A_j^{sk+1})^i`$ denotes the $`i`$-th column of the matrix $`A_j^{sk+1}`$ and if $`_{i=1}^{n_{sk+1}}\beta _i(A_j^{sk+1})^i=0`$ for some constants $`\beta _i𝐂`$ and for $`j=1,\mathrm{},p+1`$, then $`\beta _1=`$$`\mathrm{}`$$`=\beta _{n_{sk+1}}=0`$.
Remark: The entries of the matrix $`A_{p+2}`$ outside $`L_1`$ are 0. Therefore the construction from this moment on goes on like in \[Ko1\] and we omit some details.
$`4^0`$. Construct the matrices $`A_j^{sk}`$. Set $`A_j^{sk}|_{H_{k+1,\nu }}=0`$ for $`\nu =1,\mathrm{},k+1`$. Hence, condition c) will hold.
Define the linear subspaces $`S_{j,k}𝐂^{n_{sk+1}}`$ of vector-columns as follows. Fix matrices $`Q_jGL(n_{sk+1},𝐂)`$ such that the matrices $`T_j:=Q_j^1A_j^{sk+1}Q_j`$ be in upper-triangular Jordan normal forms. Then for $`j=1,\mathrm{},p+1`$ the space $`S_{j,k}`$ is spanned by the vectors of the form $`Q_jV`$ where the non-zero coordinates of $`V`$ can be only in the rows where $`T_j`$ has units and in the last rows of the smallest $`r(J_j^{n_{sk}})`$rk$`(A_j^{sk+1})`$ Jordan blocks of $`T_j`$. (In some cases one has to choose part of the blocks of given size; there are no conditions imposed on the choice.)
For $`j=1,\mathrm{},p+1`$ one has dim$`(S_{j,k})=`$rk$`(A_j^{sk+1})+r(J_j^{n_{sk}})`$rk$`(A_j^{sk+1})=r(J_j^{n_{sk}})`$.
Denote by $`D_k`$ the union of blocks $`H_{1,k+1}\mathrm{}H_{k,k+1}`$.
###### Lemma 33
Let for $`j=1,\mathrm{},p+1`$ the matrix $`A_j^{sk+1}`$ belong to the conjugacy class $`J_j^{n_{sk+1}}`$ or to its closure. If the columns of $`A_j^{sk}|_{D_k}`$ belong to the space $`S_{j,k}`$, then the matrix $`A_j^{sk}`$ belongs to the conjugacy class $`J_j^{n_{sk}}`$ or to its closure.
The reader can find the proof of this fact in \[Ko1\].
$`5^0`$. Denote by $`\mathrm{\Gamma }_k`$ the linear space of $`(p+1)`$-tuples of vector-columns $`Q_jV_j𝐂^{n_{sk+1}}`$ with zero sum where $`V_jS_{j,k}`$. One has dim$`(\mathrm{\Gamma }_k)n_{sk}`$. Indeed,
$$\mathrm{dim}(\mathrm{\Gamma }_k)\mathrm{dim}(S_{1,k})+\mathrm{}+\mathrm{dim}(S_{p+1,k})n_{sk+1}=$$
$$=r(J_1^{n_{sk}})+\mathrm{}+r(J_{p+1}^{n_{sk}})n_{sk+1}=n_{sk}+n_{sk+1}n_{sk+1}=n_{sk}$$
We admit that the $`n_{sk+1}`$ conditions arizing from $`Q_1V_1+\mathrm{}+Q_{p+1}V_{p+1}=0`$ might not be linearly independent which explains why there is an inequality.
Define $`\mathrm{\Delta }_k𝐂^{n_{sk+1}}`$ as the space spanned by the $`(p+1)`$-tuples of vector-columns of the matrices $`A_j^{sk+1}`$, $`jp+1`$. Hence, dim$`(\mathrm{\Delta }_k)=n_{sk+1}`$ and dim$`(\mathrm{\Gamma }_k/\mathrm{\Delta }_k)n_{sk}n_{sk+1}`$.
$`6^0`$. Hence, one can choose $`n_{sk}n_{sk+1}`$ columns of the $`(p+1)`$-tuple of matrices $`A_j^{sk}|_{D_k}`$ which belong to $`\mathrm{\Gamma }_k`$ and are linearly independent modulo the space $`\mathrm{\Delta }_k`$. With this choice the columns of the $`(p+1)`$-tuple of matrices $`A_j^{sk}`$ will be linearly independent, i.e. condition c) from $`3^0`$ will hold. Condition b) holds by construction. Condition a) follows from Lemma 33. Thus conditions a) – d) hold for every $`k`$ (and, in particular, for $`k=s+1`$).
$`7^0`$. Prove that if the matrices $`A_j=A_j^0`$ are constructed like this, then the centralizer $`𝒵^{}`$ of their $`(p+2)`$-tuple is trivial. Let $`X𝒵^{}`$. Denote by $`F`$ a matrix from the algebra generated by the matrices $`A_j`$ whose restriction $`F^{}`$ to $`L_1`$ is non-degenerate. Such a matrix exists – in Case A) this follows from the irreducibility of the algebra generated by the matrices $`A_j^s`$, in Case B) this follows from Lemma 32.
The commutation relation $`[X,F]|_{H_{s+1,1}}=0`$ implies $`X|_{H_{s+1,1}}=0`$ (because $`F|_{H_{\mu ,\nu }}=0`$ if $`\mu >\nu `$ and if $`\mu =\nu >1`$, and one has $`[X,F]|_{H_{s+1,1}}=(XF)|_{H_{s+1,1}}=X|_{H_{s+1,1}}F^{}=0`$ and $`F^{}`$ is non-degenerate).
The commutation relation $`[X,F]|_{H_{s,1}}=0`$ implies $`X|_{H_{s,1}}=0`$ (because $`X|_{H_{s+1,1}}=0`$ and one has $`[X,F]|_{H_{s,1}}=(XF)|_{H_{s,1}}=X|_{H_{s,1}}F^{}=0`$). Similarly $`X|_{H_{k,1}}=0`$ for $`k>1`$.
For $`k=1`$ the commutation relations $`[X,A_j]|_{H_{1,1}}=0`$ are equivalent to $`[X|_{H_{1,1}},A_j^s]=0`$, hence, $`X|_{H_{1,1}}=\alpha I`$ by triviality of the centralizer of the $`(p+2)`$-tuple of matrices $`A_j^s`$, see Lemmas 31 and 32. Assume that $`\alpha =0`$.
$`8^0`$. After $`7^0`$ the commutation relations $`[X,A_j]|_{H_{k,2}}=0`$ ($`k=1,\mathrm{},s+1`$) become equivalent to $`(A_jX)|_{H_{k,2}}=0`$. They imply $`X|_{H_{k,2}}=0`$, otherwise the columns of the $`(p+2)`$-tuple of matrices $`A_j`$ would be linearly dependent – a contradiction with condition d). In the same way one deduces that $`X|_{H_{k,l}}=0`$ for $`k=1,\mathrm{},s+1`$; $`l2`$. Hence, $`X=0`$. Without the assumption $`\alpha =0`$, see $`7^0`$, this means that $`X=\alpha I`$, i.e. $`𝒵^{}`$ is trivial.
Conditions 1) – 8) of the lemma follow immediately from the construction of the matrices $`A_j`$. The lemma is proved.
### 5.5 Proof of Lemma 32
$`1^0`$. Block-decompose the matrices from $`gl(n_s,𝐂)`$, all diagonal blocks being of size $`\chi =2`$,3,4 or 6 (recall that the Jordan normal forms $`J_j^{n_s}`$ correspond to one of the special cases from Subsection 6.1). Denote the blocks by $`U_{\mu ,\nu }`$. Construct a block-diagonal $`(p+1)`$-tuple of matrices $`A_j^0`$ with diagonal blocks defining irreducible non-equivalent representations. If they are constructed like in examples (ex0) – (ex7) from Subsection 6.2, see Lemma 39 as well, then the matrix $`B`$ can have distinct non-zero eigenvalues (recall that one has the right to multiply the diagonal blocks by non-zero constants – this preserves the conjugacy classes of $`A_j`$ and changes the eigenvalues of $`B`$).
$`2^0`$. Define $`A_{p+2}`$. Denote by $`F^{}`$ the right upper block $`m_0\times (n_sm_0)`$ (recall that $`m_0<dn_s/2`$). If $`m_0`$ is divisible by the size $`\chi `$, then set $`F=F^{}`$ and $`n^{}=n_sm_0`$. If not, then define $`F^{\prime \prime }`$ as obtained from $`F^{}`$ by deleting its first (i.e. left) $`\chi \psi `$ columns where $`\psi `$ is the rest of the division of $`m_0`$ by $`\chi `$; set $`n^{}=n_sm_0\chi +\psi `$.
Notice that the block $`F^{\prime \prime }`$ is a union of $`[m_0/\chi ]`$ rows each of $`[n^{}/\chi ]`$ entire blocks $`U_{\mu ,\nu }`$ and (when $`\psi >0`$) of a row of $`[n^{}/\chi ]`$ non-entire blocks $`U_{\mu ,\nu }`$ (of which only the first $`\psi `$ rows belong to $`F^{\prime \prime }`$). Denote by $`F`$ the block which is the union of all blocks $`U_{\mu ,\nu }`$ which entirely or partially belong to $`F^{\prime \prime }`$. Denote by $`{}_{}{}^{t}F`$ the transposed to $`F`$.
Require the restriction of $`A_{p+2}`$ to $`F^{\prime \prime }`$ to be of maximal rank and its restriction to every block $`U_{\mu ,\nu }`$ which entirely or partially belongs to $`F^{\prime \prime }`$ to be non-zero. Hence, $`J(A_{p+2})`$ consists of $`m_0`$ blocks of size 2 and of $`n2m_0`$ ones of size 1.
$`3^0`$. Define the matrices $`A_j`$, $`jp+1`$. Their restriction to every diagonal block $`U_{\mu ,\mu }`$ equals the one of $`A_j^0`$ to it, their restriction to every block $`U_{\mu ,\nu }F`$ is of the form $`A_j^0|_{U_{\mu ,\mu }}D_{j;\mu ,\nu }D_{j;\mu ,\nu }A_j^0|_{U_{\nu ,\nu }}`$, $`D_{j;\mu ,\nu }gl(\chi ,𝐂)`$, their other blocks are 0.
The representations defined by $`A_j^0|_{U_{\mu ,\mu }}`$ and $`A_j^0|_{U_{\nu ,\nu }}`$ being non-equivalent for $`\mu \nu `$, the map
$$(D_1,\mathrm{},D_{p+1})\underset{j=1}{\overset{p+1}{}}(A_j^0|_{U_{\mu ,\mu }}D_jD_jA_j^0|_{U_{\nu ,\nu }})$$
is surjective onto $`gl(\chi ,𝐂)`$. Hence, one can choose the blocks $`D_{j;\mu ,\nu }`$ such that the sum of the matrices $`A_j`$ to be 0.
Notice that for all $`j`$ the matrix $`A_j`$ is conjugate to $`A_j^0`$.
$`4^0`$. There remains to be proved that the centralizer $`𝒵`$ of the $`(p+1)`$-tuple of matrices $`A_j`$ is trivial. Denote by $`U`$ a block $`U_{\mu _0,\nu _0}`$ from $`{}_{}{}^{t}F`$. For $`X𝒵`$ the commutation relation $`[A_{p+2},X]=0`$ restricted to $`U_{\mu _0,\nu _0}`$ yields $`X_U=0`$.
One has $`A_j|_{U_{\mu _0,\mu _0}}X|_{U_{\mu _0,i}}X|_{U_{\mu _0,i}}A_j|_{U_{i,i}}=0`$, hence, if $`i\mu _0`$, then $`X|_{U_{\mu _0,i}}=0`$, if $`i=\mu _0`$, then $`X|_{U_{\mu _0,i}}=\gamma _iI`$. In the same way one deduces that $`X|_{U_{i,\nu _0}}=0`$, if $`i\nu _0`$, and $`X|_{U_{i,\nu _0}}=\gamma _iI`$ if $`i=\nu _0`$.
Let $`U^{}=U_{\mu _1,\nu _1}F`$. Set $`X_U^{}=X^{}`$, $`A_j|_U^{}=A_j^{}`$ and recall that $`X_{U_{i,i}}=\gamma _iI`$. The commutation relations restricted to $`U^{}`$ yield
$$(\gamma _{\mu _1,\mu _1}\gamma _{\nu _1,\nu _1})A_j^{}+A_j|_{U_{\mu _1,\mu _1}}X^{}X^{}A_j|_{U_{\nu _1,\nu _1}}=0$$
Sum them up from 1 to $`p+1`$. One has $`_{j=1}^{p+1}A_j|_{U_{i,i}}=0`$ for all $`i`$. Hence, the sum is $`(\gamma _{\mu _1,\mu _1}\gamma _{\nu _1,\nu _1})A_{p+2}|_U^{}`$ which is 0 only if $`\gamma _{\mu _1,\mu _1}=\gamma _{\nu _1,\nu _1}`$. But then one must have $`A_j|_{U_{\mu _1,\mu _1}}X^{}X^{}A_j|_{U_{\nu _1,\nu _1}}=0`$ for all $`j`$. The non-equivalence of the representations defined by the blocks $`A_j|_{U_{i,i}}`$ for the different values of $`i`$ implies that $`X^{}=0`$. Hence, the centralizer is trivial.
The lemma is proved.
## 6 On the existence of irreducible representations defined by nilpotent matrices
### 6.1 The basic result
In this section we consider the question of existence of irreducible $`(p+1)`$-tuples of nilpotent matrices from given conjugacy classes $`c_j`$, with given quantities $`r_j=r(c_j)`$ and with zero sum. This question was considered in \[Ko2\]. We have to repeat some of the reasoning from \[Ko2\] for two reasons:
– there is an inexactitude in the formulation of the basic result from \[Ko2\], so we explain here what is correct and what is wrong;
– we want to prove something more than the existence of such $`(p+1)`$-tuples.
Define as special the following cases. In each case the Jordan normal form defined by the class $`c_j`$ has Jordan blocks of one and the same size $`l_j`$. The special cases are
$$\begin{array}{ccccc}a)\hfill & p=3,\hfill & n=2g,\hfill & g>1,\hfill & l_1=l_2=l_3=l_4=2;\hfill \\ b)\hfill & p=2,\hfill & n=3g,\hfill & g>1,\hfill & l_1=l_2=l_3=3;\hfill \\ c)\hfill & p=2,\hfill & n=4g,\hfill & g>1,\hfill & l_1=l_2=4,l_3=2;\hfill \\ d)\hfill & p=2,\hfill & n=6g,\hfill & g>1,\hfill & l_1=6,l_2=3,l_3=2\hfill \end{array}$$
Define also some almost special cases. They are obtained from one of the special ones (it is understood from the notation from which) by replacing in one of the Jordan normal forms a couple of blocks of size $`l_j`$ by two blocks of sizes $`l_j+1`$ and $`l_j1`$. For these cases we give the sizes of the blocks:
$$\begin{array}{ccccc}a1)\hfill & J_1:3,1,2,\mathrm{},2\hfill & J_2,J_3,J_4:2,\mathrm{},2\hfill & & n=2g,g>1\hfill \\ b1)\hfill & J_1:4,2,3,\mathrm{},3\hfill & J_2:3,\mathrm{},3\hfill & J_3:3,\mathrm{},3\hfill & n=3g,g>1\hfill \\ c1)\hfill & J_1:4,\mathrm{},4\hfill & J_2:4,\mathrm{},4\hfill & J_3:3,1,2,\mathrm{},2\hfill & n=4g,g>1\hfill \\ c2)\hfill & J_1:5,3,4,\mathrm{},4\hfill & J_2:4,\mathrm{},4\hfill & J_3:2,\mathrm{},2\hfill & n=4g,g>1\hfill \\ d1)\hfill & J_1:6,\mathrm{},6\hfill & J_2:3,\mathrm{},3\hfill & J_3:3,1,2,\mathrm{},2\hfill & n=6g,g>1\hfill \\ d2)\hfill & J_1:6,\mathrm{},6\hfill & J_2:4,2,3,\mathrm{},3\hfill & J_3:2,\mathrm{},2\hfill & n=6g,g>1\hfill \\ d3)\hfill & J_1:7,5,6,\mathrm{},6\hfill & J_2:3,\mathrm{},3\hfill & J_3:2,\mathrm{},2\hfill & n=6g,g>1\hfill \end{array}$$
Definition. A representation defined by a $`(p+1)`$-tuple of matrices $`A_j`$ with zero sum is called nice if it is either irreducible or is reducible, with a trivial centralizer and the matrices $`A_j`$ admit a simultaneous conjugation to a block upper-triangular form in which the restrictions of the $`(p+1)`$-tuple to all diagonal blocks (they are all of sizes $`>1`$) define non-equivalent irreducible representations. (Hence, the algebra defined by the matrices $`A_j`$ contains a non-degenerate matrix.)
We prove in this section the following :
###### Theorem 34
1) Let the $`(p+1)`$-tuple of nilpotent conjugacy classes $`c_jgl(n,𝐂)`$ be given with $`r_1+\mathrm{}+r_{p+1}2n`$ and not corresponding to any of the special cases. Then there exists a $`(p+1)`$-tuple of matrices $`A_jc_j`$ defining a nice representation and there exists a $`(p+1)`$-tuple of distinct non-zero complex numbers $`\alpha _j`$ such that the matrix $`B:=\alpha _1A_1+\mathrm{}+\alpha _{p+1}A_{p+1}`$ has a simple non-zero eigenvalue.
2) In the conditions of 1), if the $`(p+1)`$-tuple of nilpotent conjugacy classes $`c_jgl(n,𝐂)`$ does not correspond to any of the almost special cases a1), b1), c2), or d3), then there exists an irreducible $`(p+1)`$-tuple of matrices $`A_jc_j`$ satisfying (2). In all 7 almost special cases one can obtain a matrix $`B`$ with distinct non-zero eigenvalues.
3) If the conjugacy classes $`C_j`$ are unipotent, satisfying $`r_1+\mathrm{}+r_{p+1}2n`$ and not corresponding to any of the special cases, then there exists a $`(p+1)`$-tuple of matrices $`M_jC_j`$ defining a nice representation and satisfying (1). If they do not correspond to any of the almost special cases a1), b1), c2), or d3), then there exists an irreducible $`(p+1)`$-tuple of matrices $`M_jC_j`$ satisfying (1).
Remarks: 1) Condition $`(\omega _n)`$ is necessary for the existence of nice representations – it must hold for the restrictions of the matrices to each of the diagonal blocks and the rank of a nilpotent matrix is $``$ the sum of the ranks of its restrictions to these blocks.
2) In \[Ko2\] the existence of irreducible representations in cases a1), b1), c2) and d3) is claimed which is not true. The proof of the basic result from \[Ko1\] (which uses the result from \[Ko2\]) needs only the existence of nice representations (not necessarily irreducible) and is performed with the corrected result in exactly the same way as it is done in \[Ko1\]. The rest of the results from \[Ko2\] are correct.
The theorem is proved in Subsection 6.3. We don’t prove part 3) of it which follows from parts 1) and 2) when one considers as matrices $`M_j`$ the monodromy operators of a fuchsian system with matrices-residua $`A_j`$ satisfying the conditions of part 1) or 2) (remember Proposition 11). Condition $`(\omega _n)`$ is necessary for the existence of irreducible $`(p+1)`$-tuples of nilpotent matrices satisfying (2) or of unipotent matrices $`M_j`$ satisfying (1), see \[Ko2\].
We precede the proof of the lemma by the construction of examples of irreducible triples or quadruples of matrices $`A_j`$ in some particular cases, see the next subsection, and by deducing Corollary 35 from the theorem. We refer to the examples as to (ex1), (ex2) etc. The irreducibility of the triples or quadruples from each example is checked by proving that the matrix algebra generated by the matrices is $`gl(n,𝐂)`$. To this end one first finds a matrix $`S`$ from the algebra with a single non-zero entry (by multiplying in a suitable order the matrices) and then again by suitable multiplications of the matrix $`S`$ by the matrices $`A_j`$ one obtains all other elements of the canonical basis of $`gl(n,𝐂)`$, see examples of applications of this technique in \[Ko2\].
###### Corollary 35
Let $`d>1`$ and let the $`(p+1)`$-tuple of nilpotent conjugacy classes $`c_jgl(n,𝐂)`$ be given with $`r_1+\mathrm{}+r_{p+1}2n`$ and not corresponding to any of the special cases. Then there exists an irreducible $`(p+1)`$-tuple of matrices $`A_jc_j`$ satisfying (2) and a $`(p+1)`$-tuple of distinct non-zero complex numbers $`\alpha _j`$ such that the matrix $`B:=\alpha _1A_1+\mathrm{}+\alpha _{p+1}A_{p+1}`$ has at least $`d`$ simple non-zero eigenvalues.
Proof:
$`1^0`$. Denote by $`c_j^1gl(n/d,𝐂)`$ the conjugacy classes obtained from $`c_j`$ by reducing the size of the matrices and the number of Jordan blocks of a given size $`d`$ times. Denote by $`c_j^l`$ the conjugacy classes obtained from $`c_j^1`$ by increasing the size of the matrices and the number of Jordan blocks of a given size $`l`$ times. In particular, $`c_j^d=c_j`$.
$`2^0`$. Suppose first that the conjugacy classes $`c_j^1`$ don’t correspond to an almost special case (they don’t correspond to a special case, even with $`g=1`$, because in this case $`c_j`$ would also correspond to such a case). We show that there exist irreducible $`(p+1)`$-tuples of matrices $`A_j^lc_j^l`$. For $`l=1`$ this follows from the above theorem. Suppose that it is true for $`ll_0`$. Then there exists an irreducible triple of matrices $`A_j^{l_0}c_j^{l_0}`$ whose sum is zero.
$`3^0`$. Consider the matrices $`\left(\begin{array}{cc}A_j^{l_0}& 0\\ 0& A_j^1\end{array}\right)`$. The irreducible representations $`L^{l_0}`$ and $`L^1`$ defined by the triples of matrices $`A_j^{l_0}`$ and $`A_j^1`$ satisfy the condition dim Ext$`{}_{}{}^{1}(L^{l_0},L^1)2`$. Indeed, one has dim Ext$`{}_{}{}^{1}(L^{l_0},L^1)=l_0(_{j=1}^{p+1}d(c_j^1)2(n/d)^2)2`$ because $`_{j=1}^{p+1}d(c_j^1)2n^2+2`$ (see \[Ko2\], Lemma 3). One subtracts $`(n/d)^2`$ twice – once to factor out conjugations with block upper-triangular matrices and once because the sum of the matrices is 0.
###### Lemma 36
1) If for the non-equivalent irreducible representations $`L^{}`$, $`L^{\prime \prime }`$ defined by the matrices $`A_j^{}c_j^{}`$, $`A_j^{\prime \prime }c_j^{\prime \prime }`$ (each satisfying (2)) one has dim Ext$`{}_{}{}^{1}(L^{},L^{\prime \prime })1`$, then there exists a representation defined by matrices $`A_j^{\prime \prime \prime }`$ satisfying (2), with $`A_j^{\prime \prime \prime }c_j^{}c_j^{\prime \prime }`$, which is their semi-direct sum. If dim Ext$`{}_{}{}^{1}(L^{},L^{\prime \prime })2`$, then there exists an irreducible such representation. The conjugacy classes $`c_j^{}`$, $`c_j^{\prime \prime }`$ are arbitrary (not necessarily nilpotent).
2) Let the conjugacy classes $`c_j^{}`$, $`c_j^{\prime \prime }`$ be nilpotent. Let the matrices $`B^{(i)}=_{j=1}^{p+1}\alpha _jA_j^{(i)}`$, $`i=1,2`$ have respectively $`m^{}`$ and $`m^{\prime \prime }`$ simple non-zero eigenvalues. If the matrices $`A_j^{(i)}`$ are like in 1) and if dim Ext$`{}_{}{}^{1}(L^{},L^{\prime \prime })2`$, then there exists an irreducible representation defined by matrices $`A_j`$ satisfying (2), with $`A_jc_j^{}c_j^{\prime \prime }`$ and such that the matrix $`B`$ (defined like $`B^{}`$, $`B^{\prime \prime }`$) has at least $`m^{}+m^{\prime \prime }`$ simple non-zero eigenvalues.
The lemma is proved after the corollary. It implies the existence of irreducible triples of matrices $`A_j^{l_0+1}c_j^{l_0+1}`$ with at least $`l_0+1`$ simple non-zero eigenvalues.
$`4^0`$. If the conjugacy classes $`c_j^1`$ correspond to an almost special case, then for $`l>1`$ the classes $`c_j^l`$ correspond to a $`(p+1)`$-tuple of conjugacy classes which contains in its closure a neighbouring case (their definition is given in 8. of Subsubsection 6.3.1). For such cases we prove that there exist irreducible $`(p+1)`$-tuples with $`B`$ having distinct non-zero eigenvalues, see Subsubsection 6.3.7. Hence, one can choose an irreducible $`(p+1)`$-tuple of matrices $`A_jc_j^l`$ which is close to a neighbouring case and $`B`$ will still be with distinct non-zero eigenvalues.
The corollary is proved.
Proof of Lemma 36:
$`1^0`$. As dim Ext$`{}_{}{}^{1}(L^{},L^{\prime \prime })1`$, there exists a semi-direct sum of $`L^{}`$ and $`L^{\prime \prime }`$ which is not reduced to a direct one, i.e. there exists a $`(p+1)`$-tuple of matrices $`A_j^{\prime \prime \prime }=\left(\begin{array}{cc}A_j^{}& G_j\\ 0& A_j^{\prime \prime }\end{array}\right)`$ which by simultaneous conjugation can’t be transformed into a block-diagonal one.
$`2^0`$. The representations $`L^{}`$, $`L^{\prime \prime }`$ are not equivalent which means that the centralizer $`𝒵`$ of the $`(p+1)`$-tuple of matrices $`A_j^{\prime \prime \prime }`$ is trivial (if $`\left(\begin{array}{cc}U& M\\ N& P\end{array}\right)𝒵`$, then $`NA_j^{}A_j^{\prime \prime }N=0`$; as $`L^{}`$ and $`L^{\prime \prime }`$ are irreducible and non-equivalent, one has $`N=0`$; then $`[A_j^{},U]=0`$ and $`[A_j^{\prime \prime },P]=0`$, i.e. $`U=\alpha I`$, $`P=\beta I`$ (Schur’s lemma); then $`A_j^{}MMA_j^{\prime \prime }=(\beta \alpha )G_j`$ which means that $`M=0`$, $`\alpha =\beta `$, otherwise the sum of $`L^{}`$ and $`L^{\prime \prime }`$ must be direct).
$`3^0`$. The condition dim Ext$`{}_{}{}^{1}(L^{},L^{\prime \prime })2`$ implies the existence of infinitesimal conjugations of $`A_j^{\prime \prime \prime }`$ by matrices $`I+\epsilon X_j`$, $`X_j=\left(\begin{array}{cc}Y_j& 0\\ Z_j& T_j\end{array}\right)`$ which are not tantamount to a simultaneous infinitesimal conjugation of the $`(p+1)`$-tuple of matrices $`A_j^{\prime \prime \prime }`$. Indeed, one has
$$\stackrel{~}{A}_j:=(I+\epsilon X_j)^1A_j^{\prime \prime \prime }(I+\epsilon X_j)=A_j^{\prime \prime \prime }+\epsilon \left(\begin{array}{cc}[A_j^{},Y_j]+G_jZ_j& 0\\ A_j^{\prime \prime }Z_jZ_jA_j^{}& [A_j^{\prime \prime },T_j]Z_jG_j\end{array}\right)+o(\epsilon )$$
Set $`\mathrm{\Phi }=\{(P_1,\mathrm{},P_{p+1})|P_j=A_j^{\prime \prime }Z_jZ_jA_j^{},_{j=1}^{p+1}P_j=0\}`$, $`\mathrm{\Psi }=\{(Q_1,\mathrm{},Q_3)|Q_j=A_j^{\prime \prime }ZZA_j^{}\}`$ (as $`_{j=1}^{p+1}A_j^{(i)}=0`$, $`i=1,2`$, one has $`_{j=1}^{p+1}Q_j=0`$). The condition dim Ext$`{}_{}{}^{1}(L^{},L^{\prime \prime })2`$ implies that dim$`\mathrm{\Phi }`$dim$`\mathrm{\Psi }2`$.
$`4^0`$. Denote by $`\mathrm{\Xi }`$ the codimension 1 subspace of $`\mathrm{\Phi }`$ satisfying the condition tr$`(_{j=1}^{p+1}G_jZ_j)=0`$. Hence, if the $`(p+1)`$-tuple of blocks $`Z_j`$ of the matrices $`X_j`$ belongs to $`\mathrm{\Xi }/\mathrm{\Psi }`$, then one can find blocks $`Y_j`$ and $`T_j`$ such that $`_{j=1}^{p+1}\stackrel{~}{A}_j=o(\epsilon )`$ (i.e. one can solve the equations $`_{j=1}^{p+1}[A_j^{},Y_j]+G_jZ_j=0`$ and $`_{j=1}^{p+1}([A_j^{\prime \prime },T_j]Z_jG_j)=0`$ w.r.t. $`Y_j`$ and $`T_j`$).
$`5^0`$. Hence, one can find a conjugation of $`A_j^{\prime \prime \prime }`$ by matrices $`I+\epsilon X_j+\epsilon ^2S_j(\epsilon )`$ analytic in $`\epsilon `$, with $`X_j`$ as above and such that the sum of the conjugated matrices (denoted by $`A_j`$) to be 0 (identically in $`\epsilon `$). The existence of $`S_j`$ is proved by complete analogy with the basic technical tool.
As $`(X_1,\mathrm{},X_{p+1})(\mathrm{\Xi }/\mathrm{\Psi })`$, the $`(p+1)`$-tuple is irreducible. (Indeed, the matrix algebra $`𝒜`$ generated by the matrices $`\stackrel{~}{A}_j`$ contains matrices of the form $`Y^{}=Y+O(\epsilon )`$ for every matrix $`Y`$ blocked as $`A_j^{\prime \prime \prime }`$; this follows from the main result of \[Ko4\]. In particular, for $`Y=H=\left(\begin{array}{cc}I& 0\\ 0& 0\end{array}\right)`$. Conjugate $`Y^{}`$ (with $`Y=H`$) by a matrix $`I+O(\epsilon )`$ to make its left lower block 0 (identically in $`\epsilon `$). This conjugation can’t annihilate the left lower blocks of all matrices from $`𝒜`$ due to $`(X_1,\mathrm{},X_{p+1})(\mathrm{\Xi }/\mathrm{\Psi })`$. If $`S𝒜`$ has a non-zero such block, then $`SHHSH`$ has the same left lower block and all its entries are $`O(\epsilon )`$. Multiplying $`S`$ by matrices of the form $`Y^{}`$ and adding the matrices $`Y^{}`$, one obtains a basis of $`gl(n,𝐂)`$.)
This proves 1).
$`6^0`$. Prove 2) One can assume that the matrices $`B^{}`$, $`B^{\prime \prime }`$ have no non-zero eigenvalue in common. This can be achieved by multiplying one of the $`(p+1)`$-tuples of matrices $`A_j^{(i)}`$ by $`h𝐂^{}`$ (we use here the fact that the matrices are nilpotent – such a multiplication does not change the conjugacy classes). Hence, the matrix $`B^{\prime \prime \prime }`$ has at least $`m^{}+m^{\prime \prime }`$ simple non-zero eigenvalues. For $`\epsilon 0`$ small enough the eigenvalues of the matrix $`B=\alpha _1A_1+\mathrm{}+\alpha _{p+1}A_{p+1}`$ will be close to the ones of $`B^{\prime \prime \prime }`$, hence, it will have at least $`m^{}+m^{\prime \prime }`$ non-zero simple eigenvalues.
The lemma is proved.
### 6.2 Some examples
We explain two methods for constructing irreducible triples or quadruples of nilpotent matrices. The first one places the non-zero entries of the matrices $`A_j`$ in positions $`(k,k+1)`$, $`k=1,\mathrm{},n1`$, and $`(n,1)`$. In all examples except (ex0) one has $`p=2`$.
Example (ex0): Let $`n=2`$, $`p=3`$. Set $`A_1=A_2=\left(\begin{array}{cc}0& 1\\ 0& 0\end{array}\right)`$, $`A_3=A_4=\left(\begin{array}{cc}0& 0\\ 1& 0\end{array}\right)`$. The quadruple is irreducible and for almost all values of $`\alpha _j`$ the matrix $`B`$ has two different non-zero eigenvalues.
Example (ex1): Let $`n4`$. Let
$$A_1=\left(\begin{array}{cccccccc}0& 1& 0& 0& 0& \mathrm{}& 0& 0\\ 0& 0& 1& 0& 0& \mathrm{}& 0& 0\\ 0& 0& 0& 1& 0& \mathrm{}& 0& 0\\ 0& 0& 0& 0& 1& \mathrm{}& 0& 0\\ 0& 0& 0& 0& 0& \mathrm{}& 0& 0\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& 0& 0& 0& 0& \mathrm{}& 0& 1\\ 0& 0& 0& 0& 0& \mathrm{}& 0& 0\end{array}\right),A_2=\left(\begin{array}{cccccccc}0& 1& 0& 0& 0& \mathrm{}& 0& 0\\ 0& 0& 0& 0& 0& \mathrm{}& 0& 0\\ 0& 0& 0& 1& 0& \mathrm{}& 0& 0\\ 0& 0& 0& 0& 1& \mathrm{}& 0& 0\\ 0& 0& 0& 0& 0& \mathrm{}& 0& 0\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& 0& 0& 0& 0& \mathrm{}& 0& 1\\ 1& 0& 0& 0& 0& \mathrm{}& 0& 0\end{array}\right),$$
$`A_3=A_1A_2`$. One checks directly that the matrices are nilpotent, that $`r_1=r_2=n1`$, $`r_3=2`$ and $`(A_3)^2=0`$.
Example (ex2): Let $`n=3`$. Let $`A_1=\left(\begin{array}{ccc}0& 1& 0\\ 0& 0& 1\\ 0& 0& 0\end{array}\right)`$, $`A_2=\left(\begin{array}{ccc}0& 1& 0\\ 0& 0& 0\\ 1& 0& 0\end{array}\right)`$, $`A_3=\left(\begin{array}{ccc}0& 0& 0\\ 0& 0& 1\\ 1& 0& 0\end{array}\right)`$. Hence, each matrix is nilpotent, of rank 2.
###### Lemma 37
The characteristic polynomial of a matrix having non-zero entries in positions $`(k,k+1)`$, $`k=1,\mathrm{},n1`$, and $`(n,1)`$ and zeros elsewhere is of the form $`\lambda ^n+a`$, $`a0`$. Hence, the eigenvalues of such a matrix are all non-zero and distinct.
The lemma is to be checked directly. It implies that the eigenvalues of the matrices $`B`$ defined after examples (ex1) and (ex2) are non-zero and distinct.
Another method is the non-zero entries to be in positions $`(k,k+1)`$, $`k=1,\mathrm{},n1`$, and $`(n1,1)`$, $`(n,2)`$. The idea to give such examples – the first of the matrices is initially in Jordan normal form and then one conjugates it with $`I+E_{n,1}`$. As our matrices will be sparce, we’ll list only the entries of these positions, in this order. E.g., we write $`A_1:110|11`$ instead of $`A_1=\left(\begin{array}{cccc}0& 1& 0& 0\\ 0& 0& 1& 0\\ 1& 0& 0& 0\\ 0& 1& 0& 0\end{array}\right)`$ (the vertical line separates the last two entries just for convenience). The reader is advised to draw in each example the matrices oneself.
Example (ex3): Let $`n=6`$. Let
$$\begin{array}{ccccccccc}\hfill A_1:& \hfill 1& \hfill 1& \hfill 1& \hfill 1& \hfill 1& \hfill |& \hfill 1& \hfill 1\\ \hfill A_2:& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill |& \hfill 0& \hfill 0\\ \hfill A_3:& \hfill 0& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill |& \hfill 1& \hfill 1\end{array}$$
Hence, $`J(A_1)`$ consists of a single block of size 6, $`J(A_2)`$ consists of two blocks of size 3 and $`J(A_3)`$ consists of three blocks of size 2. One can give examples when $`n`$ is even, $`J(A_2)`$ consists of two blocks of size 3 and of $`(n6)/2`$ blocks of size 2; $`J(A_3)`$ consists of $`n/2`$ blocks of size 2; $`J(A_1)`$ consists of a single block of size $`n`$. To this end one adds to the right of the fourth from the left column of numbers in the above example a pack of $`n6`$ units in the row of $`A_1`$, a pack of $`(n6)/2`$ groups of the form $`0,1`$ in the row of $`A_2`$ and a pack of $`(n6)/2`$ groups of the form $`1,0`$ in the row of $`A_3`$.
Example (ex4): Let $`n=9`$. Let
$$\begin{array}{cccccccccccc}\hfill A_1:& \hfill 1& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill 1& \hfill 1& \hfill |& \hfill 1& \hfill 1\\ \hfill A_2:& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill |& \hfill 0& \hfill 0\\ \hfill A_3:& \hfill 0& \hfill 0& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill |& \hfill 1& \hfill 1\end{array}$$
One checks directly that $`J(A_1)`$ consists of two Jordan blocks, of sizes 4 and 5, $`J(A_2)`$ consists of three Jordan blocks of size 3 and $`J(A_3)`$ consists of three Jordan blocks of size 2 and of one of size 3.
Example (ex5): Let $`n=10`$. Let
$$\begin{array}{ccccccccccccc}\hfill A_1:& \hfill 1& \hfill 1& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill 1& \hfill 1& \hfill |& \hfill 1& \hfill 1\\ \hfill A_2:& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill |& \hfill 0& \hfill 0\\ \hfill A_3:& \hfill 0& \hfill 0& \hfill 1& \hfill 0& \hfill 1& \hfill 0& \hfill 1& \hfill 0& \hfill 0& \hfill |& \hfill 1& \hfill 1\end{array}$$
In this example $`J(A_1)`$ consists of two Jordan blocks of size 5, $`J(A_2)`$ consists of two Jordan blocks of size 3 and of one of size 4, $`J(A_3)`$ consists of five Jordan blocks of size 2.
Example (ex6): Let $`n=12`$. Let
$$\begin{array}{ccccccccccccccc}\hfill A_1:& \hfill 1& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill 1& \hfill |& \hfill 1& \hfill 1\\ \hfill A_2:& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill |& \hfill 0& \hfill 0\\ \hfill A_3:& \hfill 0& \hfill 0& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill 0& \hfill 1& \hfill 1& \hfill 0& \hfill 0& \hfill |& \hfill 1& \hfill 1\end{array}$$
Hence, $`J(A_1)`$ consists of three Jordan blocks of size 4, $`J(A_2)`$ consists of four Jordan blocks of size 3 and $`J(A_3)`$ consists of two Jordan blocks of size 3 and of three of size 2.
Example (ex7): Let $`n=5`$. Let
$$\begin{array}{cccccccc}\hfill A_1:& \hfill 1& \hfill 1& \hfill 1& \hfill 1& \hfill |& \hfill 1& \hfill 1\\ \hfill A_2:& \hfill 1& \hfill 1& \hfill 0& \hfill 1& \hfill |& \hfill 0& \hfill 0\\ \hfill A_3:& \hfill 0& \hfill 0& \hfill 1& \hfill 0& \hfill |& \hfill 1& \hfill 1\end{array}$$
$`J(A_1)`$ consists of a single block of size 5, $`J(A_2)`$ and $`J(A_3)`$ consist each of a block of size 3 and of a block of size 2. One can give examples when $`n>5`$ is odd, each of $`J(A_2)`$ and $`J(A_3)`$ consists of one block of size 3 and of $`(n3)/2`$ blocks of size 2; $`J(A_1)`$ consists of a single block of size $`n`$. To this end one adds to the left of the vertical line in the above example a pack of $`n5`$ units in the row of $`A_1`$, a pack of $`(n5)/2`$ groups of the form $`0,1`$ in the row of $`A_2`$ and a pack of $`(n5)/2`$ groups of the form $`1,0`$ in the row of $`A_3`$.
###### Lemma 38
The characteristic polynomial of a matrix having non-zero entries only in positions $`(k,k+1)`$, $`k=1,\mathrm{},n1`$, and $`(n1,1)`$, $`(n,2)`$ is of the form $`\lambda ^n+b\lambda `$. Hence, when $`b0`$ its roots are all distinct and one of them equals 0.
The lemma is to be checked directly.
###### Lemma 39
For the triples of matrices $`A_j`$ from each of the examples (ex3) – (ex7) there exist conjugations of the matrices $`A_j`$ with matrices $`I+O(\epsilon )`$ analytic in $`\epsilon (𝐂,0)`$ such that for $`\epsilon 0`$ the eigenvalues of the matrix $`B`$ are distinct and non-zero.
Proof:
$`1^0`$. In examples (ex4) – (ex6) an infinitesimal conjugation of $`A_1`$ with $`I+\epsilon (E_{2,1}E_{n,n1})`$, of $`A_2`$ with $`I+\epsilon E_{2,1}`$ and of $`A_3`$ with $`I\epsilon E_{n,n1}`$ creates a new triple (of matrices $`A_j^1`$) satisfying (2) in first approximation w.r.t. $`\epsilon `$. Hence, there exist true conjugations differing from the above ones by terms $`O(\epsilon ^2)`$ after which for the deformed matrices $`A_j^2`$ (2) holds. One has $`A_j^1A_j^2=o(\epsilon )`$.
$`2^0`$. Choose $`\alpha _j`$ such that the entries of $`B^1|_{\epsilon =0}`$ in positions $`(k,k+1)`$, $`k=1,\mathrm{},n1`$, and $`(n1,1)`$, $`(n,2)`$ to be $`0`$. The entries in the following positions (and only they) of the matrix $`B^1`$ (hence, $`B^2`$ as well) are non-zero and $`O(\epsilon )`$: $`(1,1)`$, $`(2,2)`$, $`(n1,n1)`$, $`(n,n)`$ and $`(n,1)`$. Hence, det$`B^1=v\epsilon +o(\epsilon )`$, $`v0`$ and for $`\epsilon 0`$ small enough all eigenvalues of $`B^1`$ and $`B^2`$ are non-zero and distinct, the last of them is $`O(\epsilon )`$.
$`3^0`$. In examples (ex7) and (ex3) the triple of matrices $`A_j^1`$ is created by an infinitesimal conjugation of $`A_1`$ and $`A_2`$ with $`I\epsilon E_{n,n1}`$ and of $`A_3`$ with $`I+\epsilon E_{2,1}`$. The entries in the following positions (and only they) of the matrix $`B^1`$ (hence, $`B^2`$ as well) are non-zero and $`O(\epsilon )`$: $`(n1,n1)`$, $`(n,n)`$ and $`(n,1)`$. For the rest the reasoning is the same.
The lemma is proved.
###### Lemma 40
1) Denote by $`\mathrm{\Phi }_1`$, $`\mathrm{\Phi }_2`$ two irreducible representations defined by any of examples (ex1) – (ex7) (with $`n=6`$ in (ex3) and $`n=5`$ in (ex7)). Then Ext$`{}_{}{}^{1}(\mathrm{\Phi }_1,\mathrm{\Phi }_2)2`$.
2) The same is true if each of $`\mathrm{\Phi }_1`$, $`\mathrm{\Phi }_2`$ corresponds to one of the examples (ex3) or (ex7) without the restrictions respectively $`n=6`$ and $`n=5`$.
The proof is left for the reader.
### 6.3 Proof of Theorem 34
#### 6.3.1 Simplification and plan of the proof
1. The first observation to be made is that if there exists a $`(p+1)`$-tuple of matrices $`A_jc_j`$ satisfying 1) or 2) of the theorem, then there exist such $`(p+1)`$-tuples for every $`(p+1)`$-tuple of conjugacy classes $`c_j^{}`$ where for each $`j`$ either $`c_j=c_j^{}`$ or $`c_j`$ is subordinate to $`c_j^{}`$. To this end one has to apply the basic technical tool.
2. Call operation $`(s,l)`$, $`sl1`$ the changing of a given nilpotent conjugacy class $`c`$ containing two Jordan blocks of sizes $`s`$ and $`l`$ to $`c^{}`$ in which these blocks are replaced by two blocks of sizes $`s+1`$ and $`l1`$. The class $`c`$ lies in the closure of $`c^{}`$. If the nilpotent conjugacy class $`c`$ lies in the closure of the nilpotent conjugacy class $`c^{\prime \prime }`$, then there exists a sequence of conjugacy classes $`c_0=c`$, $`c_1`$, $`\mathrm{}`$ ,$`c_\mu =c^{\prime \prime }`$ such that $`c_i`$ is obtained from $`c_{i1}`$ as a result of some operation $`(s,l)`$. This is proved in \[Ko1\].
3. Let $`Ac`$, $`A^{}c^{}`$, $`c`$, $`c^{}`$ being nilpotent conjugacy classes. Then $`c`$ lies in the closure of $`c^{}`$ if and only if for all $`i`$ one has rk$`(A)^i`$rk$`(A^{})^i`$ (proved in \[Ko1\]).
4. For given $`r𝐍`$ denote by $`\mathrm{\Omega }_0(r)`$ the nilpotent orbit of rank $`r`$ of least dimension. It is unique and has Jordan blocks either of one and the same size or of two consecutive sizes. This follows from 2. If the size is one and the same, then there exists a single orbit $`\mathrm{\Omega }_1(r)`$ containing $`\mathrm{\Omega }_0(r)`$ in its closure and contained in the closure of any orbit with the same value of $`r`$ and different from $`\mathrm{\Omega }_0(r)`$ and $`\mathrm{\Omega }_1(r)`$. It is obtained from $`\mathrm{\Omega }_0(r)`$ by an operation $`(k,k)`$, $`k`$ being the size of the blocks of $`\mathrm{\Omega }_0(r)`$.
5. Let $`r_1+\mathrm{}+r_{p+1}>2n`$. Then it is possible to change some of the conjugacy classes to subordinate ones of smaller rank to get the condition $`r_1+\mathrm{}+r_{p+1}=2n`$ and the new $`(p+1)`$-tuple of ranks not to correspond to any of the ones of special or almost special cases.
6. Call merging of two nilpotent conjugacy classes $`c`$ and $`c^{}`$ the following procedure defined for $`r(c)+r(c^{})n1`$ and when $`c=\mathrm{\Omega }_0(r(c))`$, $`c^{}=\mathrm{\Omega }_0(r(c^{}))`$ (hence, at least one of the two Jordan normal forms has only blocks of size $`2`$). If $`Ac`$, $`A^{}c^{}`$ are Jordan matrices with decreasing order of the sizes $`b_i`$, $`b_i^{}`$ of the blocks, rk$`A`$rk$`A^{}`$ (hence, $`b_i^{}2`$), then construct the nilpotent matrix $`A^{\prime \prime }`$ as follows: insert on the first superdiagonal (it comprises the positions $`(k,k+1)`$) between the packs of $`b_i1`$ and $`b_{i+1}1`$ units from $`A`$ a unit from $`A^{}`$ as long as this is possible. These packs are units from Jordan blocks of sizes $`b_i`$ and $`b_{i+1}`$. When inserting a unit, it takes the place of the 0 and the units of $`A`$ do not change their positions.
Given the matrix $`A^{\prime \prime }`$ it is self-evident how to represent it in the form $`A+A^{}`$ with $`Ac`$, $`A^{}c^{}`$. If there exists an irreducible or nice representation in which one of the matrices is from $`c^{\prime \prime }`$, then there exists such a representation with one more matrix, two of the matrices being from $`c`$ and $`c^{}`$, the other conjugacy classes remaining the same.
7. We prove the theorem only in the case $`r_1+\mathrm{}+r_{p+1}=2n`$ making use of 5. Making use of 4., when $`p4`$, one might restrict oneself to the case when $`c_j=\mathrm{\Omega }_0(r_j)`$ for all $`j`$.
It is also possible to restrict oneself to the cases $`p=2`$ and $`p=3`$ due to the possibility to merge Jordan normal forms. Indeed, if $`r_1+\mathrm{}+r_{p+1}=2n`$, if $`c_j=\mathrm{\Omega }_0(r(c_j))`$ for all $`j`$ and if $`p4`$, then a merging is always possible. Moreover, it is possible to be done with avoiding to come to the special case a) or to the almost special case a1). Indeed, when passing from $`p=4`$ to $`p=3`$ by a merging, the sum of three of the quantities $`r_j`$ is $`n`$ and one can choose a couple to be merged such that one of the four quantities $`r_j`$ which remain after the merging to be $`<n/2`$.
When passing from $`p=3`$ to $`p=2`$ by merging (the passage is not defined for the cases a) and a1)), one can avoid to come to any of the other special or almost special cases. Indeed, merging results either in a Jordan normal form with a single Jordan block of size $`m>1`$ or with greatest difference $`h`$ between the sizes of two of the Jordan blocks $`2`$; one has $`m3`$ and the cases $`m=3`$ and $`h=2`$ are possible only if one of the Jordan normal forms to be merged has only one block of size $`2`$, the rest of size 1. In this case one can merge other two of the Jordan normal forms to avoid the special and almost special cases.
8. Finally, after having restricted oneself to the cases $`p=2`$ and $`p=3`$, one can again use 1., 2. and 4. to replace the given conjugacy classes by the triple or quadruple of conjugacy classes $`\mathrm{\Omega }_0(r_j)`$. This is not always possible because one could come to a special or almost special case.
Let this be not a special or an almost special case. Use the notation ”$`A_1=2`$” or ”$`A_1=(2,3)`$” in the sense ”the Jordan normal form of $`A_1`$ has only Jordan blocks of size 2” (”of sizes 2 and 3”). The following cases of triples $`\mathrm{\Omega }_0(r_1)`$, $`\mathrm{\Omega }_0(r_2)`$, $`\mathrm{\Omega }_0(r_3)`$ are possible (see \[Ko2\]):
$$\begin{array}{ccccc}(A)\hfill & A_1=(1,2)& & (B)\hfill & A_1=2\mathrm{or}A_1=(2,3);A_2=(2,3)\\ (C)\hfill & A_1=(2,3);A_2=3;A_3=(3,4)& & (D)\hfill & A_1=(2,3);A_2=3;A_3=4\\ (E)\hfill & A_1=(2,3);A_2=3;A_3=(4,5)& & (F)\hfill & A_1=(2,3);A_2=3;A_3=5\\ (G)\hfill & A_1=(2,3);A_2=3;A_3=(5,6)& & (H)\hfill & A_1=2\mathrm{or}(2,3);A_2=(3,4);A_3=4\\ (I)\hfill & A_1=2\mathrm{or}(2,3);A_2=(3,4);A_3=(4,5)& & (J)\hfill & A_1=2\mathrm{or}(2,3);A_2=(3,4);A_3=5\\ (K)\hfill & A_1=2\mathrm{or}(2,3);A_2=(3,4);A_3=(5,6)& & & \end{array}$$
Case (C) is considered in Subsubsection 6.3.2, case (F) is considered in Subsubsection 6.3.4. The other cases are considered in Subsubsection 6.3.3.
If the triple or quadruple of classes $`\mathrm{\Omega }_0(r_j)`$ is a special case, then we use 2. and 4. and prove part 1) of the theorem for all almost special cases stemming from it – this will imply that part 1) holds for the initial triple of nilpotent orbits, see Subsubsection 6.3.6. For the almost special cases c1), d1) and d2) we prove that 2) holds as well, see Subsubsection 6.3.5.
Finally, we consider all cases obtained from one of the almost special ones a1), b1), c2) and d3) by an operation $`(s,l)`$ on one of the three orbits (call such cases neighbouring), see Subsubsection 6.3.7. We prove that part 2) of the theorem holds for all these cases. Hence, part 2) will hold for all cases when $`p=2`$ or 3, when $`r_1+\mathrm{}+r_{p+1}=2n`$ and when all special and almost special cases are avoided. Hence, it will hold when $`r_1+\mathrm{}+r_{p+1}2n`$ and when all special and almost special cases are avoided.
#### 6.3.2 Proof of the theorem for $`p=2`$, in case (C)
We construct triples of matrices satisfying the requirements of the lemma. The construction is done by induction on $`n`$. The induction base are the examples from Subsection 6.2.
The size $`n`$ must be $`>3`$. Decrease the size of the matrices by 3 and delete a block of size 3 from each Jordan normal form. Denote the new Jordan normal forms by $`J_j^{}`$. This can lead only to case (C) again or to case (D) (with $`n`$ replaced by $`n3`$). By inductive assumption, there exists an irreducible triple of nilpotent matrices $`A_j^{}`$ of size $`n3`$ with $`J(A_j^{})=J_j^{}`$ and such that the matrix $`B^{}=\alpha _1A_1^{}+\alpha _2A_2^{}+\alpha _3A_3^{}`$ have a simple non-zero eigenvalue.
Denote by $`A_j^{\prime \prime }`$ the triple of matrices $`A_j`$ from example (ex2). One can assume that the matrices $`B^{}`$ and $`B^{\prime \prime }=\alpha _1A_1^{\prime \prime }+\alpha _2A_2^{\prime \prime }+\alpha _3A_3^{\prime \prime }`$ have no eigenvalue in common (to achieve this one can, if necessary, multiply one of the triples by $`h𝐂^{}`$). By Lemma 36 (the reader should check that it is applicable), there exist irreducible triples of nilpotent matrices $`A_j`$ with $`J(A_j)=J_j`$, the matrix $`B`$ having at least two simple non-zero eigenvalues (hence, at least one). This proves the induction step in case (C).
#### 6.3.3 Proof of the theorem for $`p=2`$, in cases (A), (B), (D), (E) and (G) – (K)
$`1^0`$. The cases (B), (D), (E), (G) and (H) – (K) are considered by analogy with (C) and we define only the matrices $`A_j^{\prime \prime }`$. As we make use of examples (ex3) – (ex7), one should keep in mind Lemma 39. Lemma 36 is applicable in all cases because there holds Lemma 40. (One can represent the triple of Jordan normal forms as a direct sum of triples of Jordan normal forms corresponding to one of the examples (ex1) – (ex7) and then use Lemma 40.)
In case (B) $`J(A_3)`$ must contain a block of size $`\kappa 5`$. The matrices $`A_j^{\prime \prime }`$ of size $`\kappa `$ are defined by examples (ex7) and (ex3) depending on whether $`n`$ is even or odd. If $`J(A_3)`$ consists of a single block, then these examples define directly the matrices $`A_j`$. If not, then the matrices $`A_j^{}`$ correspond again to case (B).
In case (D) the matrices $`A_j^{\prime \prime }`$ are of size 12; they are the triple from example (ex6) (if $`n=12`$, then case (D) is proved directly by example (ex6)). The matrices $`A_j^{}`$ are from case (D) again.
In case (E) the matrices $`A_j^{\prime \prime }`$ are of size 9; they are defined in example (ex4). The matrices $`A_j^{}`$ are from case (E) or from case (D) or from case (F).
In case (G) there are at least three blocks of size 2 in $`J(A_1)`$. The matrices $`A_j^{\prime \prime }`$ are defined by example (ex3). The matrices $`A_j^{}`$ are either from case (G) or from case (F).
In cases (H) and (I) the matrices $`A_j^{\prime \prime }`$ are of size 4; they are defined by example (ex1) with $`n=4`$. One has to prove that there are at least two Jordan blocks of size 2 in $`J(A_1)`$, see \[Ko2\]. In case (H) the matrices $`A_j^{}`$ are from case (H) or from case (D), in case (I) they are from case (I), (J), (H), (E), (F) or (D).
In case (J) the matrices $`A_j^{\prime \prime }`$ are of size 10; they are defined by example (ex5) ($`J(A_2)`$ contains at least five blocks of size 3, otherwise $`r_1+r_2+r_3>2n`$). The matrices $`A_j^{}`$ are either from case (J) or from case (F).
In case (K) there are at least three blocks of size 2 in $`J(A_1)`$ and at least two blocks of size 3 in $`J(A_2)`$ for the same reason, see \[Ko2\], and the matrices $`A_j^{\prime \prime }`$ are of size 6, defined by example (ex3). The matrices $`A_j^{}`$ are from case (K), (J), (G) or (F).
In all these cases the triple of Jordan normal forms of the matrices $`A_j^{}`$ corresponds to neither of the special or almost special cases.
$`2^0`$. Consider case (A) (we follow the same ideas as in \[Ko2\]). Decrease by 1 the sizes of the matrices and decrease by 1 the sizes of two Jordan blocks respectively of $`J(A_2)`$ and $`J(A_3)`$. Delete a Jordan block of size 1 from $`J(A_1)`$. One can choose the diminished blocks of $`J(A_2)`$, $`J(A_3)`$ such that the triple of Jordan normal forms of size $`n1`$ obtained like this not to correspond to any of the special or almost special cases. (It suffices to leave in each of the two Jordan normal forms of size $`n1`$ a couple of blocks of different size.) This defines the Jordan normal forms of the matrices $`A_j^{}`$. The sum of their ranks equals $`2(n1)`$.
Set $`A_j^1=\left(\begin{array}{cc}A_j^{}& G_j\\ 0& A_j^{\prime \prime }\end{array}\right)`$. The matrices $`A_j^{\prime \prime }`$ are of size 1 and equal 0. One constructs the blocks $`G_j`$ such that the Jordan normal form of $`A_2^1`$ to be the necessary one (i.e. $`\mathrm{\Omega }_0(r_2)`$). One sets $`G_3=0`$, $`G_2=G_1`$. (The condition $`G_3=0`$ can be achieved by conjugating the triple.) Hence, the matrices $`A_j^1`$ do not define a semi-direct sum of $`L^{}`$ and $`L^{\prime \prime }`$ because $`J(A_2^1)`$ is not a direct sum of $`J(A_2^{})`$ and $`J(A_2^{\prime \prime })`$.
Thus the matrix $`B^1`$ has a simple non-zero eigenvalue. It is an eigenvalue of $`B^{}`$.
After this one deforms the triple of matrices $`A_j^1`$ into an irreducible one, with the necessary Jordan normal forms, see \[Ko2\]. One can also use the same reasoning as the one from case (F), see the next subsubsection. Hence, for the deformed triple the analog of the matrix $`B^1`$ still has a simple non-zero eigenvalue.
#### 6.3.4 Proof of the theorem for $`p=2`$, in case (F)
There exists an irreducible triple of nilpotent $`9\times 9`$ matrices $`A_1^{}`$, $`A_2^{}`$, $`A_3^{}`$, $`A_1^{}+A_2^{}+A_3^{}=0`$ where $`J(A_1^{})`$ ($`J(A_2^{})`$; $`J(A_3^{})`$) contains 1 block $`3\times 3`$ and 3 blocks $`2\times 2`$ (3 blocks $`3\times 3`$; 1 block $`4\times 4`$ and 1 block $`5\times 5`$), see case (E). Hence, the matrix $`B^{}`$ has a non-zero simple eigenvalue.
There exists a triple of nilpotent $`15\times 15`$-matrices $`A_1^0`$, $`A_2^0`$, $`A_3^0`$, $`A_1^0+A_2^0+A_3^0=0`$ with $`J(A_j^0)=𝒥_j`$ where $`𝒥_1`$ ($`𝒥_2`$; $`𝒥_3`$) consists of 1 block $`3\times 3`$ and 6 blocks $`2\times 2`$ (of 5 blocks $`3\times 3`$; of 3 blocks $`5\times 5`$) and the matrices $`A_1^0`$, $`A_2^0`$ look like this (recall that $`A_j^{}`$ are $`9\times 9`$):
$$\left(\begin{array}{ccccccc}A_1^{}& \eta _1& \eta _2& \eta _3& \eta _4& \eta _5& \eta _6\\ 0& 0& 1& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 1& 0& 0\\ 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0\end{array}\right),\left(\begin{array}{ccccccc}A_2^{}& \phi _1& \phi _2& \phi _3& \phi _4& \phi _5& \phi _6\\ 0& 0& 1& 0& 0& 0& 0\\ 0& 0& 0& 1& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 1& 0\\ 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0\end{array}\right)$$
The matrix $`B^0`$ has the same non-zero eigenvalues as $`B^{}`$. Hence, it has a simple non-zero eigenvalue. One can deform the triple of matrices $`A_j^0`$ into an irreducible triple of matrices $`A_j`$ with the necessary Jordan normal forms, see \[Ko2\]. Hence, the matrix $`B`$ for small values of the deformation parameter still has a simple non-zero eigenvalue.
We explain the details of this deformation to show why this method is not applicable to some of the almost special cases (as was claimed in \[Ko2\]).
One looks for matrices $`A_j`$ of the form $`A_1=A_1^0+\epsilon L`$ where only the last row of $`L`$ is non-zero, with $`L_{15,14}=1`$; we assume that $`L_{15,j}=0`$ for $`j=10,11,12,13,15`$ and that for $`\epsilon 0`$ the sizes of the Jordan blocks of $`J(A_1)`$ equal 3,2,2,2,2,2,2; one sets $`A_j=(I+\epsilon X_j(\epsilon ))^1A_j^{}(I+\epsilon X_j(\epsilon )`$, $`j=2,3`$, where $`X_j`$ are analytic in $`\epsilon (𝐂,0)`$ (their existence is justified by the basic technical tool).
The matrix algebra $`𝒜^0`$ generated by the matrices $`A_j`$ contains matrices of the form $`\left(\begin{array}{cc}P& Q\\ O(\epsilon )& O(\epsilon )\end{array}\right)`$ for all $`P`$, $`Q`$ where $`Pgl(9,𝐂)`$. The centralizer of $`𝒜^0`$ is trivial, see \[Ko2\].
Fix a matrix $`F𝒜^0`$ with $`P=I`$, $`Q=0`$. By taking powers of $`F`$, one can assume that the 6 right columns of $`F`$ are identically 0.
Conjugate the matrices $`A_j`$ by a matrix $`\left(\begin{array}{cc}I& 0\\ O(\epsilon )& I\end{array}\right)`$ so that after the conjugation the matrix $`F`$ become equal to $`\left(\begin{array}{cc}I& 0\\ 0& 0\end{array}\right)`$. Hence, $`𝒜^0`$ contains all matrices of the form $`Y=\left(\begin{array}{cc}P& Q\\ 0& 0\end{array}\right)`$ (if $`S𝒜^0`$, then $`Y=FS𝒜^0`$). Two cases are possible after the conjugation:
1) the three matrices become block upper-triangular, with diagonal blocks of sizes 9 and 6;
2) there exists an entry in the left lower block $`6\times 9`$ which is $`0`$ for $`\epsilon 0`$.
Eliminate case 1). If this were true, then the restrictions of $`A_j`$ to the right lower block $`6\times 6`$ would be nilpotent, with sizes of the blocks equal to 2,2,2; 3,3; 5,1. They must define a nilpotent algebra $`𝒜^1`$.
Indeed, it is impossible to have an irreducible triple with such sizes of the blocks because one would have $`r_1+r_2+r_3=11<12`$. It is impossible to conjugate the algebra to a block upper-triangular form with at least one diagonal block $`\stackrel{~}{P}`$ irreducible, of size $`m>1`$. Indeed, the restrictions of the matrices to such a block would be nilpotent, with sizes of the blocks not greater respectively than 2, 3 and 5. One checks directly that for $`m=2,3,4`$ and 5 it is impossible to have $`r_1^{}+r_2^{}+r_3^{}2m`$, $`r_j^{}`$ being the ranks of the restrictions of the matrices to the block $`\stackrel{~}{P}`$. (It is this part of the reasoning which is not applicable to the proof of the almost special cases a1), b1), c2) and d3); e.g., in case b1) there exist block upper-triangular triples $`6\times 6`$ with irreducible diagonal blocks $`3\times 3`$.)
On the other hand if $`𝒜^1`$ is nilpotent, then it can be conjugated to an upper-triangular form. The restriction to the right lower block $`6\times 6`$ of the matrix $`(A_1+A_2/2)|_{\epsilon =0}`$ has non-zero entries in positions $`(k,k+1)`$, $`k=10,\mathrm{},14`$. Hence, the conjugation can be carried out by a matrix $`I+O(\epsilon )`$. Such a conjugation cannot annihilate the entry $`A_{1;15,14}`$. Hence, the algebra $`𝒜^1`$ is not nilpotent.
Consider case 2). Let the algebra $`𝒜^0`$ contain a matrix $`S`$ with $`S_{10,j}|_{\epsilon =0}0`$ for some $`j9`$ (if one has $`S_{i,j}|_{\epsilon =0}0`$ for $`i>10`$ and for some $`j9`$, then one can multiply $`S`$ by $`(A_1+A_2/2)^{i10}`$ to have $`S_{10,j}|_{\epsilon =0}0`$).
Then one can assume that only the left lower block $`6\times 9`$ of $`S`$ is non-zero (one can consider instead of $`S`$ the matrix $`SFFSF`$). Multiply the matrix $`S`$ by matrices $`Y`$ defined above. Hence, $`𝒜^0`$ contains matrices of the form $`\left(\begin{array}{cc}P& Q\\ O(\epsilon )& O(\epsilon )\end{array}\right)`$ with $`Pgl(10,𝐂)`$, for all $`P`$ and $`Q`$ and one can repeat the reasoning which led us to cases 1) and 2), but this time the size of the block $`P`$ has increased by 1. Continuing like this, we see that $`𝒜^0=gl(15,𝐂)`$, i.e. the triple $`A_1`$, $`A_2`$, $`A_3`$ is irreducible.
#### 6.3.5 Proof of the theorem in the almost special cases c1), d1) and d2)
There exist irreducible triples of nilpotent matrices $`A_j`$ satisfying (2) with sizes of the Jordan blocks like in cases c1), d1) or d2) but with $`g=1`$. These triples can be obtained by deforming respectively the ones from examples (ex1) with $`n=4`$ for case c1) and (ex3) for cases d1) and d2).
Consider the direct sum of such a triple and of a triple from example (ex1) with $`n=4`$ for case c1) and of one from example (ex3) for cases d1) and d2). Lemma 36 is applicable to such a direct sum which provides the existence of irreducible triples from cases c1), d1) and d2) for $`g=2`$. In the same way one constructs such triples for all $`g>1`$ – by deforming the direct sum of a triple for $`g1`$ and of one from example (ex1) with $`n=4`$ for case c1) or (ex3) for cases d1) and d2) and by using Lemma 36.
The irreducible representations thus obtained can be considered as deformations of certain direct or semi-direct sums of representations whose diagonal blocks are of sizes 4 or 6 and whose matrices $`B`$ have distinct non-zero eigenvalues, see Lemma 39. This property persists under small deformations.
#### 6.3.6 Proof of the theorem in the almost special cases a1), b1), c2) and d3)
We consider only case b1) in detail. The other cases are treated by analogy and we explain the differences at the end of the subsubsection. Recall that in all these cases we prove the existence of nice representations.
Construct the matrices $`A_j=\left(\begin{array}{ccccc}A_j^1& 0& \mathrm{}& 0& H_j^1\\ 0& A_j^2& \mathrm{}& 0& H_j^2\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& 0& \mathrm{}& A_j^{g1}& H_j^{g1}\\ 0& 0& \mathrm{}& 0& A_j^g\end{array}\right)`$ where $`A_j^k`$ are $`3\times 3`$, nilpotent, of rank 2 and the representations defined by the triples $`A_j^k`$ are irreducible for all $`k`$. Moreover, they are presumed to be non-equivalent (this can be achieved by multiplying them by constants $`g_k𝐂^{}`$) and the matrices $`B^k`$ to have non-zero distinct eigenvalues; the eigenvalues of the matrix $`B`$ can be presumed non-zero and distinct as well; see example (ex2).
Assume that the matrix $`A_1`$ is in upper-triangular Jordan normal form. Then we set $`H_1^k=\left(\begin{array}{ccc}0& 0& 0\\ 0& 0& 0\\ 0& 0& 1\end{array}\right)`$ for all $`k`$. Hence, the Jordan normal form of $`A_1`$ consists of $`g2`$ blocks of size 3, of a block of size 4 and of a block of size 2 (to be checked directly).
The blocks $`H_j^k`$ for $`j=2,3`$ are defined such that $`H_j^k=A_j^kD_j^kD_j^kA_j^d`$, $`D_j^kgl(3,𝐂)`$ and $`H_1^k+H_2^k+H_3^k=0`$. Such a choice of $`H_j^k`$ is possible because the representations defined by the triples $`A_j^k`$ for different values of $`k`$ are irreducible non-equivalent and the mapping
$$(D_2^k,D_3^k)A_2^kD_2^kD_2^kA_2^g+A_3^kD_3^kD_3^kA_3^g$$
is surjective onto $`gl(3,𝐂)`$. Notice that the blocks $`H_2^k`$, $`H_3^k`$ result from conjugation of $`A_j`$ with $`I+D_j`$ where $`D_j`$ has non-zero entries only in the last 3 columns and first $`3(g1)`$ rows (its restriction to the rows with indices $`3k2`$, $`3k1`$, $`3k`$ and to the last three columns equals $`D_j^k`$). Hence, $`J(A_2)`$ and $`J(A_3)`$ consist each of $`g`$ blocks of size 3.
Prove that the centralizer $`𝒵`$ of the triple of matrices $`A_j`$ is trivial. Block-decompose a matrix from $`gl(n,𝐂)`$ into blocks $`3\times 3`$ (denoted by $`X^{\mu ,\nu }`$). For $`Y𝒵`$ set $`Y^{\mu ,\nu }=Y|_{X^{\mu ,\nu }}`$. One has (first for $`\nu <\mu =g`$ and then for $`\mu \nu `$, $`\nu g1`$) $`Y^{\mu ,\nu }A_j^\nu A_j^\mu Y^{\mu ,\nu }=0`$. Hence, if $`\mu \nu <g`$, then $`Y^{\mu ,\nu }=0`$, if $`\mu =\nu `$, then $`Y^{\mu ,\nu }=\alpha _\mu I`$ (we use the non-equivalence of the representations and Schur’s lemma).
For $`\nu =g`$ one has $`(\alpha _k\alpha _g)H_1^k+A_1^kY^{k,g}Y^{k,g}A_1^g=0`$. Hence, $`\alpha _k=\alpha _g`$ because $`H_1^k`$ is not of the form $`A_1^kY^{k,g}Y^{k,g}A_1^g`$. But then $`A_j^kY^{k,g}Y^{k,g}A_j^g=0`$ for $`j=2,3`$ which implies $`Y^{k,g}=0`$ for $`k<g`$. Hence, the centralizer is trivial.
In all other almost special cases one similarly constructs triples or quadruples of matrices $`A_j`$ satisfying the conclusions from 1) of the theorem.
In all cases the matrix $`A_j`$ whose Jordan form has to be changed (from equal sizes of the Jordan blocks to one obtained by replacing a couple $`l_j`$, $`l_j`$ of sizes by $`l_j+1`$, $`l_j1`$) has equal blocks $`H_j^k`$ which have a unit in the right lower corner and zeros elsewhere; the restriction of the matrix $`A_j`$ to the diagonal blocks is in Jordan normal form. The diagonal blocks are of sizes 2, 4 or 6. We leave the details for the reader.
#### 6.3.7 Proof of the theorem in the neighbouring cases
There are two types of neighbouring cases. Recall that an almost special case is obtained from a special one by an operation $`(s,l)`$ performed on one of the three or four Jordan normal forms and a neighbouring case is obtained by performing another such operation. In the first type the second operation is performed on one of the other Jordan normal forms, in the second type it is performed on the same one. We consider only cases neighbouring to a1), b1), c2) and d3), in the cases neighbouring to c1), d1) and d2) there exist irreducible representations, see Subsubsection 6.3.5.
Remark: In all neighbouring cases the triples or quadruples can be considered as deformations of triples or quadruples from almost special cases, in which the matrix $`B`$ has distinct non-zero eigenvalues. Hence, this is so in all neighbouring cases as well.
Neighbouring cases of the first type.
Construct an irreducible triple or quadruple in a neighbouring case of the first type. Let $`A_j^{}`$ be the matrices from the triple or quadruple of the corresponding almost special case as constructed in the previous subsubsection.
Consider the only case of first type neighbouring to b1). (It is the only one up to permutation of the three Jordan normal forms. The rest of neighbouring cases of the first type are considered by analogy.) Suppose that it is the orbit of $`A_2^{}`$ to be changed and that $`A_2^{}`$ is in upper-triangular Jordan normal form. We assume that the triple $`A_1^{}`$, $`A_2^{}`$, $`A_3^{}`$ is obtained from the triple of matrices $`A_j`$ constructed in the previous subsubsection by conjugation with $`(I+D_2)^1`$. Set $`A_2=A_2^{}+\epsilon A_2^0`$ where the matrix $`A_2^0`$ has non-zero entries only in the last three rows and first $`n3`$ columns. The restriction of $`A_2^0`$ to the block $`X^{g,k}`$ defined in the previous subsubsection equals $`H_1^k`$ (defined also there). Hence, for $`\epsilon 0`$ the Jordan normal form of $`A_2`$ is the required one.
After this look for $`A_j`$ ($`j=1,3`$) in the form $`A_j=(I+\epsilon X_j(\epsilon ))^1A_j^{}(I+\epsilon X_j(\epsilon )`$ where $`X_j`$ are analytic in $`\epsilon (𝐂,0)`$ (their existence is justified by the basic technical tool).
Show that the algebra $`𝒜^{}`$ generated by the matrices $`A_j`$ is $`gl(n,𝐂)`$ from where part 2) of the theorem follows. The algebra $`𝒜`$ generated by the matrices $`A_j^{}`$ is the one of all matrices $`W`$ having arbitrary entries in the diagonal blocks $`X^{i,i}`$ and in the blocks $`X^{g,i}`$. This follows from the basic result in \[Ko4\]. Hence, $`𝒜^{}`$ contains matrices of the form $`S+\epsilon T`$ for all $`S𝒜`$.
The algebra $`𝒜^{}`$ contains the matrix $`(A_2)^3/\epsilon =_{\kappa =1}^{g1}E_{3g2,3\kappa }`$. By multiplying and postmultiplying it by matrices from $`𝒜^{}`$, one can obtain matrices of the form $`V+\epsilon Z`$ for $`V`$ having any restriction to the block $`X^{i,j}`$, for all $`i`$ and for $`jg1`$.
The matrices $`W`$ and $`V`$ contain a basis of $`gl(n,𝐂)`$ for $`\epsilon 0`$ small enough. Hence, the triple $`A_1`$, $`A_2`$, $`A_3`$ is irreducible.
Neighbouring cases of the second type.
$`1^0`$. The cases of second type neighbouring to a given almost special case can be characterized by the sizes of the blocks of the Jordan normal form which changes w.r.t. the corresponding special case. There are four possibilities:
$$\begin{array}{ccccc}1)\hfill & l_j+2,l_j2,l_j,\mathrm{},l_j\hfill & & 2)\hfill & l_j+2,l_j1,l_j1,l_j,\mathrm{},l_j\hfill \\ 3)\hfill & l_j+1,l_j+1,l_j2,l_j,\mathrm{},l_j\hfill & & 4)\hfill & l_j+1,l_j+1,l_j1,l_j1,l_j,\mathrm{},l_j\hfill \end{array}$$
Possibilities 1), 2), 3) and 4) appear for the first time respectively for $`g=2`$, $`g=3`$, $`g=3`$ and $`g=4`$. We explain the construction for these minimal values of $`g`$, for all others the existence is proved by induction on $`g`$, when considering direct sums of triples or quadruples constructed for $`g1`$ and triples or quadruples defined by examples (ex0), (ex1) with $`n=4`$ or (ex3). We deform such direct sums into irreducible representations by means of Lemma 36.
$`2^0`$. Possibility 1).
Let $`g=2`$. Explain in details the case neighbouring to b1). Denote by $`A_j^{}=\left(\begin{array}{cc}A_j^1& H_j^1\\ 0& A_j^2\end{array}\right)`$ matrices defining triples from case b1) with $`g=2`$ and $`H_j^1`$ defined like in the previous subsubsection. The centralizer of the triple is trivial and the representations defined by the matrices $`A_j^1`$ and $`A_j^2`$ are non-equivalent. $`A_1^1`$ and $`A_1^2`$ are upper-triangular Jordan blocks of size 3. Hence, the matrix algebra $`𝒜`$ generated by the matrices $`A_j^{}`$ contains all block upper-triangular matrices with blocks $`3\times 3`$, see \[Ko4\]. In particular, it contains the matrix $`S=\left(\begin{array}{cc}0& 0\\ 0& I\end{array}\right)`$.
Set $`A_1=A_1^{}+\epsilon Y`$, $`Y=E_{4,1}`$. Hence, $`A_1`$ has for $`\epsilon 0`$ Jordan blocks of sizes 5 and 1. Set $`A_j=(I+\epsilon X_j(\epsilon ))^1A_j^{}(I+\epsilon X_j(\epsilon ))`$, $`X_j`$ being analytic in $`\epsilon (𝐂,0)`$ and such that $`_jA_j=0`$ (the existence of $`X_j`$ follows from the basic technical tool).
The algebra $`𝒜^{}`$ contains for all $`F𝒜`$ a matrix $`F+O(\epsilon )`$. Hence, $`𝒜^{}`$ contains a matrix $`S^{}=S+O(\epsilon )`$. Hence, it contains the matrix $`Q=S^{}(A_1)^3`$ which is of the form $`\epsilon \left(\begin{array}{cc}& \\ G& \end{array}\right)+o(\epsilon )`$ with $`G=E_{4,3}0`$. By multiplying and postmultiplying $`Q/\epsilon `$ by matrices from $`𝒜^{}`$, one can obtain matrices of the form of $`Q/\epsilon `$ with any block $`G`$. These matrices together with the matrices $`F+O(\epsilon )`$ form a basis of $`gl(6,𝐂)`$. Hence, the triple of matrices $`A_j`$ is irreducible.
In all other neighbouring cases with possibility 1) the left lower block of the matrix $`Y`$ has a single unit in its left upper corner and zeros elsewhere.
$`3^0`$. Possibility 2).
Consider the case neighbouring to b1) (the ones neighbouring to a1), c2) and d3) are considered by analogy). Consider a block upper-triangular triple of matrices $`A_j^1`$ with a trivial centralizer like in case b1) with $`g=2`$; the diagonal blocks are $`3\times 3`$, they define non-equivalent representations. Consider its direct sum with an irreducible triple of matrices $`A_j^2`$ defined by example (ex2). There exists a semi-direct sum of such triples (one can use arguments like the ones from the proof of Lemma 36; the matrices are block upper-triangular, with diagonal blocks $`3\times 3`$). After this one deforms the triple into a nearby irreducible one like in the previous example – one sets
$$A_1=\left(\begin{array}{ccccccccc}0& 1& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 1& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 1& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 1& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 0\\ \epsilon & 0& 0& 0& 0& 0& 0& 1& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 0& 0\end{array}\right)$$
and $`A_j=(I+\epsilon X_j(\epsilon ))^1A_j^{}(I+\epsilon X_j(\epsilon ))`$ for $`j=2,3`$. The matrix algebra $`𝒜`$ contains all block upper-triangular matrices from $`gl(9,𝐂)`$ with blocks $`3\times 3`$, see \[Ko4\]. Hence, $`𝒜^{}`$ contains a matrix of the form $`F=E_{7,7}+O(\epsilon )`$.
One has $`(A_1)^3=E_{1,4}+\epsilon E_{7,3}`$. The matrix $`R=F(A_1)^3/\epsilon `$ belongs to $`𝒜^{}`$. One has $`R_{7,3}0`$ for $`\epsilon =0`$. Like in the previously considered case one concludes that $`𝒜^{}=gl(9,𝐂)`$ and that the triple $`A_1`$, $`A_2`$, $`A_3`$ is irreducible.
$`4^0`$. Possibility 3).
In the case neighbouring to b1) (the ones neighbouring to a1), c2) and d3) are considered by analogy) one has $`g3`$. For $`g=3`$ one sets $`A_j=A_j^{}+\epsilon A_j^0`$ where the triple of matrices $`A_j^{}`$ is block upper-triangular, with blocks $`3\times 3`$, the diagonal blocks defining non-equivalent irreducible representations. Set
$$A_1^{}+\epsilon A_1^0=\left(\begin{array}{ccccccccc}0& 1& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 1& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 1& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 1& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 1\\ \epsilon & 0& 0& \epsilon & 0& 0& 0& 1& 0\\ 0& 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 0& 0\end{array}\right)$$
For $`j=2,3`$ one sets $`A_j^{}=\left(\begin{array}{ccc}A_j^1& 0& A_j^1D_j^1D_j^1A_j^3\\ 0& A_j^2& A_j^2D_j^2D_j^2A_j^3\\ 0& 0& A_j^3\end{array}\right)`$ (i.e. $`A_j^{}`$ are the matrices $`A_j`$ from the previous subsubsection). The blocks $`A_j^i`$, $`i=1,2,3`$ are nilpotent rank 2 matrices. Set $`A_j=(I+\epsilon X_j(\epsilon ))^1A_j^{}(I+\epsilon X_j(\epsilon ))`$ for $`j=2,3`$. We let the reader check oneself that the matrices $`A_j`$ have the necessary Jordan normal forms.
One shows next that the algebra $`𝒜`$ contains all matrices with arbitrary blocks in positions (1,1), (1,3), (2,2), (2,3), (3,3) (using \[Ko4\]). Hence, the algebra $`𝒜^{}`$ contains a matrix $`P=E_{7,7}+O(\epsilon )`$ and the matrix $`P(A_1)^3/\epsilon `$ has (for $`\epsilon =0`$) non-zero entries in positions (7,3) and (7,6). Like in the previous case one concludes that $`𝒜^{}=gl(9,𝐂)`$, i.e. the triple $`A_1`$, $`A_2`$, $`A_3`$ is irreducible.
$`5^0`$. Possibility 4).
One must have $`g4`$. Let $`g=4`$. Consider the case neighbouring to b1). One constructs a triple of matrices $`A_j^{}`$ with
$$A_1^{}=\left(\begin{array}{cccc}A_1^0& \phi _1& \phi _2& \phi _3\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\end{array}\right),A_2^{}=\left(\begin{array}{cccc}A_2^0& \eta _1& \eta _2& \eta _3\\ 0& 0& 1& 0\\ 0& 0& 0& 1\\ 0& 0& 0& 0\end{array}\right)$$
and $`A_3^{}=A_1^{}A_2^{}`$, $`A_j^0`$ are $`9\times 9`$. The blocks of $`J(A_j^0)`$ are of sizes 4,4,1; 3,3,3; 3,3,3, the triple of nilpotent matrices $`A_j^0`$ is irreducible. Its existence follows from the case neighbouring to b1) from possibility 3). One chooses the vector-columns $`\phi _j`$ and $`\eta _j`$ such that $`J(A_2^{})`$ and $`J(A_3^{})`$ to have each four blocks of size 3 and $`J(A_1^{})`$ to have blocks of sizes 4,4,2,1,1.
Assume that $`A_1^{}`$ is in Jordan normal form and that $`\phi _1`$ has a unit in its last position and zeros elsewhere. Set $`A_1=A_1^{}+\epsilon L`$ where only the last row of $`L`$ is non-zero, with $`L_{12,11}0`$ and $`L_{12,10}=L_{12,12}=0`$. One chooses $`L`$ such that for $`\epsilon 0`$ the sizes of the blocks of $`J(A_1)`$ to be 4,4,2,2. Set for $`j=2,3`$ $`A_j=(I+\epsilon X_j(\epsilon ))^1A_j^{}(I+\epsilon X_j(\epsilon ))`$. After this the irreducibility of the triple $`A_1`$, $`A_2`$, $`A_3`$ is proved by analogy with case (F).
In the case neighbouring to c2) one sets
$$A_1^{}=\left(\begin{array}{ccccc}A_1^0& \phi _1& \phi _2& \phi _3& \phi _4\\ 0& 0& 0& 0& 0\\ 0& 0& 0& 1& 0\\ 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0\end{array}\right),A_2^{}=\left(\begin{array}{ccccc}A_2^0& \eta _1& \eta _2& \eta _3& \eta _4\\ 0& 0& 1& 0& 0\\ 0& 0& 0& 1& 0\\ 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0\end{array}\right)$$
where $`A_j^0gl(12,𝐂)`$. The sizes of the blocks of $`J(A_j^{})`$ are 5,5,3,3; 4,4,4,4; 2,2,2,2,2,2,2,1,1. The sizes of the blocks of $`J(A_j^0)`$ are 5,5,2; 4,4,4; 2,2,2,2,2,2. The existence of such an irreducible triple of nilpotent matrices $`A_j^0`$ follows from the case neighbouring to b1) from possibility 3). Set $`A_3=A_3^{}+\epsilon L`$, $`A_j=(I+\epsilon X_j(\epsilon ))^1A_j^{}(I+\epsilon X_j(\epsilon ))`$, $`j=1,2`$. One has $`L_{16,15}0`$ and $`L_{16,13}=L_{16,14}=L_{16,16}=0`$. For the rest of the reasoning is like in the previous case.
In the case neighbouring to d3) one sets
$$A_1^{}=\left(\begin{array}{cc}A_1^0& T_1\\ 0& A_1^{}\end{array}\right),A_2^{}=\left(\begin{array}{cccc}A_2^0& T_2& & \\ 0& A_2^{}& & \end{array}\right)$$
(18)
where $`A_j^0gl(18,𝐂)`$ and
$$A_1^{}=\left(\begin{array}{cccccc}0& 0& 0& 0& 0& 0\\ 0& 0& 1& 0& 0& 0\\ 0& 0& 0& 1& 0& 0\\ 0& 0& 0& 0& 1& 0\\ 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0\end{array}\right),A_2^{}=\left(\begin{array}{cccccc}0& 1& 0& 0& 0& 0\\ 0& 0& 1& 0& 0& 0\\ 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 1& 0\\ 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0\end{array}\right);$$
$`J(A_2^{})`$ has blocks of size 3; $`J(A_3^{})`$ has eleven blocks of size 2 and two of size 1; $`J(A_1^{})`$ has blocks of sizes 7,7,5,5. The sizes of the blocks of $`J(A_j^0)`$ are respectively 7,7,4; six times 3; nine times 2.
Set $`A_3=A_3^{}+\epsilon L`$, $`A_j=(I+\epsilon X_j(\epsilon ))^1A_j^{}(I+\epsilon X_j(\epsilon ))`$, $`j=1,2`$. One has $`L_{24,23}0`$ and $`L_{24,j_0}=0`$ for $`1j_024`$, $`j_023`$. Irreducible triples of such nilpotent matrices $`A_j^0`$ exist by the case neighbouring to d3) from possibility 3). The rest of the reasoning is like in the previous two cases.
In the case neighbouring to a1) one represents $`A_j^{}`$ in the form (18) with
$$A_1^{}=A_4^{}=0,A_2^{}=A_3^{}=\left(\begin{array}{cc}0& 1\\ 0& 0\end{array}\right)$$
with $`A_j^0gl(6,𝐂)`$. The sizes of the blocks of $`J(A_j^{})`$ are 3,3,1,1; 2,2,2,2; 2,2,2,2; 2,2,2,1,1. The ones of $`J(A_j^0)`$ are 3,3; 2,2,2; 2,2,2; 2,2,2. The existence of an irreducible quadruple of nilpotent matrices $`A_j^0`$ follows from the case neighbouring to a1) from possibility 3). Set $`A_4=A_4^{}+\epsilon L`$, $`A_j=(I+\epsilon X_j(\epsilon ))^1A_j^{}(I+\epsilon X_j(\epsilon ))`$, $`j=1,2,3`$. The matrix $`L`$ has a single non-zero entry in position $`(8,7)`$. The rest of the reasoning is like in the previous three cases.
The theorem is proved.
|
warning/0005/nlin0005028.html
|
ar5iv
|
text
|
# Parametric localized modes in quadratic nonlinear photonic structures
## I Introduction
The physics and applications of photonic band gap materials (or photonic crystals) have been an active topic of research for more than a decade . The next step in application of photonic crystals is to create tunable band gap materials where the gap could be controlled by an external parameter. One of the recent suggestions is based on the inverse opal structure — a microscopic lattice of spheres of air surrounded by silicon — with a layer of liquid crystal material that can make the transmission properties programmable by applying an electric field. Continuous tuning can also be realized by varying the liquid crystal temperature .
Another important idea towards creating dynamically tunable band gap materials for switches and transistors operating entirely with light is to employ their nonlinear properties, thus creating nonlinear photonic crystals. The concept of nonlinear photonic crystals, defined as having a spatially periodic nonlinearity, was introduced by Scalora et. al. in a numerical study of ultra-fast optical switching and limiting in cubic nonlinear Kerr materials. Bistability and localized modes in photonic superlattices with embedded layers possessing nonresonant cubic (or Kerr-type) nonlinearities have also been discussed in the literature . Recent advances in the so-called cascaded nonlinearities demonstrate an effective way to lower the switching power by employing parametric interaction and frequency conversion in noncentrosymmetric quadratic nonlinear optical materials. Parametric interactions are also known to support solitary waves, spatial quadratic solitons , that exist in homogeneous media where spatial localization is induced by two-wave parametric mixing processes between the fundamental wave and its second harmonic. Quadratic nonlinear photonic crystal was introduced as a concept by Berger in a study of multi-wavelength frequency conversion.
Taking into account the similarities between the localized defect modes in linear inhomogeneous media and nonlinear localized modes in homogeneous media , we wonder if parametric interactions can support localized modes in a variety of photonic band gap structures with quadratic (or $`\chi ^{(2)}`$) nonlinearities. The first step in this direction has been recently presented in Ref. , where we have analyzed second-harmonic generation (SHG) at a thin, effectively quadratic nonlinear layer separating two (generally different) homogeneous linear media, and predicted multistability of SHG for both plane waves and localized modes, also describing two-color localized photonic modes that can be excited at the interface.
The main purpose of this paper is twofold. First, we generalize the results of Ref. to the case of a thin quadratic nonlinear layer embedded into an arbitrarily stratified periodic linear medium. In particular, we consider a nonlinear defect layer with second-order nonlinear response in a perfectly periodic dielectric structure — a one-dimensional analog of a photonic crystal with a nonlinear impurity. Secondly, we develop a general formalism for analyzing localized modes in multilayer structures – nonlinear superlattices — and describe two-color localized gap modes supported by a periodic lattice of thin layers with quadratic nonlinearity.
The structure of the paper is the following. In Sec. II we present our model that is described by a system of two coupled nonlinear equations for the envelopes of the fundamental and second-harmonic waves. In the stationary case, the solution can be presented as a superposition of the forward and backward traveling waves, and this allows to derive an effective system of discrete coupled-mode equations for the wave amplitudes at the layers, the so-called discrete $`\chi ^{(2)}`$ equations (Sec. III A). Solutions of these equations for localized nonlinear modes are briefly discussed in Sec. III B. At last, in Sec. IV we consider a generalization of the results of Ref. to the case of a single quadratic nonlinear layer embedded into a periodic linear medium, and describe a number of new features of the localized modes that appear due to a bandgap structure.
## II Model
First, we discuss the physical motivation for our model. Let us consider an interface between two semi-infinite bulk optical media with inversion symmetry. The interface layer breaks the symmetry and therefore it should possess a nonvanishing surface quadratic response that can be enhanced by a proper coating creating a nonlinear layer with quadratic nonlinearity . Such a layer corresponds to an effective nonlinear defect that can support photonic modes localized at the interface. There exists a strong experimental evidence of SHG in such type of localized photonic modes. For example, recent experimental results reported SHG in periodic photonic band-gap structures with embedded nonlinear defect layers. An enhancement of the parametric interaction in the vicinity of the defects was observed, suggesting that SHG occurs in localized modes, being suppressed for propagating modes. The surface nature of SHG was confirmed by comparison of experimental data and results of direct numerical simulations .
To introduce an analytically solvable model for SHG in localized waves, we follow Ref. and consider a fundamental frequency (FF) wave propagating along the $`Z`$-direction in a linear slab waveguide, as shown in Fig. 1. We assume that the interfaces (or defect layers) possess a quadratic nonlinear response, so that a FF wave can parametrically couple to its second harmonic (SH) wave. The coupled-mode equations for the complex envelope functions $`E_j(x,Z)`$ (we use $`j`$=1,2 for FF and SH, respectively) can be written in the form,
$$\begin{array}{c}i\frac{E_1}{Z}+D_1\frac{^2E_1}{x^2}+\epsilon _1(x)E_1+\mathrm{\Gamma }_1(x)E_1^{}E_2=0,\hfill \\ i\frac{E_2}{Z}+D_2\frac{^2E_2}{x^2}+\epsilon _2(x)E_2+\mathrm{\Gamma }_2(x)E_1^2=0,\hfill \end{array}$$
(1)
where $`D_j`$ are the diffraction coefficients ($`D_j>0`$). In the approximation of infinitely thin interface layers (valid when the width of each layer is much smaller than the characteristic transverse scale of the FF and SH wave envelopes), we take $`\epsilon _j(x)=\epsilon _{0j}(x)+_n\kappa _j\delta (xx_n)`$ and $`\mathrm{\Gamma }_j(x)=_n\gamma _j\delta (xx_n)`$, where $`x_n`$ denotes the position of the $`n`$-th nonlinear interface, $`\gamma _j`$ are the nonlinearity coefficients, $`\epsilon _{0j}(x)`$ and $`\kappa _j`$ account for the phase velocity differences in bulk and interface materials.
In order to reduce the number of physical parameters, we normalize Eqs. (1) as follows: $`E_1(Z)=u(z)e^{i\overline{\epsilon }_{01}Z}/\sqrt{\gamma _1\gamma _2}`$, $`E_2(Z)=v(z)e^{2i\overline{\epsilon }_{01}Z}/\gamma _1`$, $`\sigma =D_2/D_1`$, $`\nu _j(x)=(\epsilon _{0j}(x)j\overline{\epsilon }_{01})/D_1`$, $`\overline{\epsilon }_{0j}`$ is the average value of $`\epsilon _{0j}(x)`$, and $`\beta _j=\kappa _j/D_1`$, where $`z=Z/D_1`$ is measured in units of $`D_1`$. Then, the dimensionless equations become
$$\begin{array}{c}i\frac{u}{z}+\frac{^2u}{x^2}+\nu _1\left(x\right)u\hfill \\ +\underset{n}{}\delta \left(xx_n\right)\left(\beta _1u+u^{}v\right)=0,\hfill \\ i\frac{v}{z}+\sigma \frac{^2v}{x^2}+\nu _2\left(x\right)v\hfill \\ +\underset{n}{}\delta \left(xx_n\right)\left(\beta _2v+u^2\right)=0.\hfill \end{array}$$
(2)
At this point, it is important to note that our equations \[Eqs. (1) or (2)\] describe the beam evolution in the framework of the so-called parabolic approximation, valid for the rays propagating mainly along the $`Z`$ direction. In other words, the characteristic length of the beam distortion due to diffraction and refraction along the $`Z`$ axis should be much larger than the beam width in the transverse direction $`x`$. This leads to the requirement of a shallow grating, $`|\epsilon _{0j}(x)\overline{\epsilon }_{0j}|\overline{\epsilon }_{0j}`$. On the other hand, to make the parametric interaction effective, the mismatch between the phase velocities of the FF and SH waves should be small, $`2\overline{\epsilon }_{01}\overline{\epsilon }_{02}`$. Then, the ratio of the diffraction coefficients is approximately $`\sigma =1/2`$, and we use this value in the numerical simulations presented below.
For spatially localized solutions, the system (2) conserves the Hamiltonian
$`H`$ $`=`$ $`{\displaystyle _{\mathrm{}}^+\mathrm{}}\left\{\right|{\displaystyle \frac{u}{x}}|^2+{\displaystyle \frac{\sigma }{2}}|{\displaystyle \frac{v}{x}}|^2\nu _1(x)|u|^2{\displaystyle \frac{\nu _2(x)}{2}}|v|^2`$
$`{\displaystyle \underset{n}{}}\delta (xx_n)\left[\beta _1\right|u|^2+{\displaystyle \frac{\beta _2}{2}}|v|^2+\mathrm{Re}\left(u^2v^{}\right)]\}dx`$
and the total power $`P=_{\mathrm{}}^+\mathrm{}\left(|u|^2+|v|^2\right)𝑑x`$.
## III Periodic layered structures
### A General formalism
To develop a general formalism for describing stationary, spatially localized modes, we consider an infinite system of uniformly spaced nonlinear interfaces located periodically at the positions $`x_n=nh`$, which separate identical linear layers, $`\nu _j\left(x+nh\right)\nu _j\left(x\right)`$. Example of such a structure is shown in Fig. 2. This type of a one-dimensional (1D) nonlinear photonic crystal (NPC) can be used to prohibit light propagation along the transverse $`x`$ axis under certain conditions, resulting in field localization, and it resembles the operation of the so-called photonic crystal fiber in the linear regime . The fundamental properties of NPC can be understood by studying nonlinear localized modes. Such modes appear in the frequency gap of the layered linear structure, and they can be called discrete gap solitons or, in a broader context, intrinsic localized modes .
We search for solutions of Eqs. (2) in the form:
$`u(x,z)=c_1\left(x\right)e^{i\lambda _1z},v(x,z)=c_2\left(x\right)e^{i\lambda _2z},`$
where $`c_j\left(x\right)`$ describe the transverse profiles of FF and SH waves, respectively, and the real propagation constants $`\lambda _j`$ satisfy the phase-matching condition $`\lambda _2=2\lambda _1`$. Then, as the FF and SH fields do not interact in a linear bulk medium, each of them can be presented as a composition of two linear eigenmodes, corresponding to pairs of counter-propagating waves with opposite wavevector components along the $`x`$ axis. Regardless the internal structure of a linear layer between nonlinear interfaces, the wave amplitudes at its boundaries can be related by a transfer matrix:
$$\left(\begin{array}{c}a_j^{(n+1,)}\hfill \\ b_j^{(n+1,)}\hfill \end{array}\right)=T^{(j)}\left(\lambda _j\right)\left(\begin{array}{c}a_j^{(n,+)}\hfill \\ b_j^{(n,+)}\hfill \end{array}\right).$$
(3)
Here indices $`\pm `$ stand for the waves on the right and on the left of a given nonlinear layer, respectively, and the amplitudes of counter-propagating waves are denoted by $`a_j`$ and $`b_j`$, as shown in Fig. 3.
In order to calculate the dependence of the matrix elements on the propagation constants, we should solve the corresponding linear problem. For practical applications, NPC can be produced by embedding nonlinear layers in an otherwise linear Bragg grating structure, see, e.g., Ref. . Thus, we assume that each linear layer consists of several sub-layers with a constant refractive index, see Fig. 2, i.e. $`\nu _j(x)=\nu _{j,m}`$ for $`x_{n,m}xx_{n,m+1}`$, where $`m`$ is used to number the sub-layer inside a linear slice ($`1mM`$), and we have, by definition, $`x_{n,1}=x_{n1,M+1}=x_n=nh`$. Then, the field in each of the sub-layers can be written as:
$$c_j\left(x\right)=a_j^{(n,m)}e^{\mu _{j,m}\left(xx_{n,m}\right)}+b_j^{(n,m)}e^{\mu _{j,m}\left(xx_{n,m}\right)}.$$
(4)
By definition, the amplitudes from Eq. (3) are $`a_j^{(n,+)}=a_j^{(n,1)}`$, $`a_j^{(n+1,)}=a_j^{(n,M+1)}`$, with similar relations holding for $`b_j^{(n,\pm )}`$.
Transverse wavenumbers can be calculated using the dispersion relations, $`\lambda _1=\mu _{1,m}^2+\nu _{1,m}`$ and $`\lambda _2=\sigma \mu _{2,m}^2+\nu _{2,m}`$, together with the phase matching condition $`\lambda _2=2\lambda _1`$. From these expressions we conclude that, in general, for given $`\lambda _j`$, the waves can be either localized (i.e., $`\mu _{j,m}`$ is real) or propagating (i.e., $`\mu _{j,m}`$ is imaginary). Then, we notice that the waves in a multi-layered linear medium can be determined as linear eigenmodes localized at the sub-layer boundaries. First, from Eq. (4) it follows that the variation of the field amplitude through the $`m`$-th sub-layer of the width $`h_mx_{n,m+1}x_{n,m}`$ is characterized by the following transfer matrix:
$$T_\mathrm{p}^{(j,m)}=\left(\begin{array}{cc}e^{\mu _{j,m}h_m},\hfill & 0\hfill \\ 0,\hfill & e^{\mu _{j,m}h_m}\hfill \end{array}\right).$$
(5)
Second, the variation of the wave amplitudes at the boundary can be calculated by applying the field continuity conditions following from the model (2). Specifically, we equate the amplitudes $`c_j`$ and their derivatives $`dc_j/dx`$ on both sides of the interface, and find the transfer matrix accounting for the field localization at a boundary separating the $`m`$-th and $`(m+1)`$-th sub-layers,
$$T_\mathrm{r}^{(j,m)}=\frac{1}{2}\left(\begin{array}{cc}1+t_\mathrm{r}^{j,m},\hfill & 1t_\mathrm{r}^{j,m}\hfill \\ 1t_\mathrm{r}^{j,m},\hfill & 1+t_\mathrm{r}^{j,m}\hfill \end{array}\right),$$
(6)
where $`t_\mathrm{r}^{j,m}=\mu _{j,m}/\mu _{j,m+1}`$. To simplify the further analysis, we include all linear properties in the transfer matrix. Then, $`\mu _{j,M+1}\mu _{j,1}`$ due to periodicity of the underlying grating, and the linear $`\delta `$-response is characterized by the following matrix:
$$T_\delta ^{(j,m)}=\frac{1}{2}\left(\begin{array}{cc}2+t_\delta ^{j,m},\hfill & t_\delta ^{j,m}\hfill \\ t_\delta ^{j,m},\hfill & 2t_\delta ^{j,m}\hfill \end{array}\right),$$
(7)
where $`t_\delta ^{j,m}=\beta _j/\mu _{j,m}`$, and $`m`$ is the index of the linear layer with the delta-interface (in our case, $`M+1`$). The total transfer matrix can be then found as a product,
$$T^{(j)}=T_\delta ^{(j,M+1)}T_\mathrm{r}^{(j,M)}T_\mathrm{p}^{(j,M)}\mathrm{}T_\mathrm{r}^{(j,1)}T_\mathrm{p}^{(j,1)}.$$
(8)
After calculating the matrix elements, we express the field in terms of the mode amplitudes defined at the nonlinear interfaces, $`c_j^{(n)}=c_j\left(x_n\right)`$, combining the relation (3) with the continuity conditions at the layers, $`c_j^{(n)}=a_j^{(n,1)}+b_j^{(n,1)}=a_j^{(n,M+1)}+b_j^{(n,M+1)}`$. Integrating Eqs. (2) over small intervals including the nonlinear interfaces at the positions $`x_n`$ (note that we exclude linear responses as they are already accounted for in the transfer matrices), we derive the so-called discrete $`\chi ^{(2)}`$ equations:
$$\begin{array}{c}\eta _1U_n+\left(U_{n1}+U_{n+1}\right)+\chi _1U_n^{}V_n=0,\hfill \\ \eta _2V_n+\left(V_{n1}+V_{n+1}\right)+\chi _2U_n^2=0,\hfill \end{array}$$
(9)
where $`U_n=c_1^{(n)}\sqrt{|\xi _1\xi _2|}`$ and $`V_n=c_2^{(n)}|\xi _1|`$ are the normalized FF and SH amplitudes at the $`n`$-th nonlinear layer. Parameters $`\chi _j`$ and $`\eta _j`$ are defined by the matrix elements,
$$\begin{array}{c}\xi _1=T_s^{(1)}/\left(2\mu _{1,1}\right),\xi _2=T_s^{(2)}/\left(2\sigma \mu _{2,1}\right)\hfill \\ \eta _j=\mathrm{Tr}(T^{(j)}),\chi _j=\mathrm{sign}\left(\xi _j\right),\hfill \end{array}$$
(10)
where
$$\begin{array}{c}T_s^{(j)}=T_{[1,1]}^{(j)}+T_{[2,1]}^{(j)}T_{[1,2]}^{(j)}T_{[2,2]}^{(j)},\hfill \\ \mathrm{Tr}(T^{(j)})T_{[1,1]}^{(j)}+T_{[2,2]}^{(j)}.\hfill \end{array}$$
(11)
Hereafter we use the notation $`T_{[n,m]}^{(j)}`$ to denote the matrix element in the row $`n`$ and column $`m`$ of the matrix $`T^{(j)}`$. It is easy to verify that the parameters $`\eta _j`$ and normalization coefficients $`\xi _j`$ are real for any (real) propagation constants $`\lambda _j`$. A proof of this fact, together with discussion of some other properties of the constructed transfer matrices, is given in Appendix A.
We note that some particular cases of the system (9) have been earlier discussed (but, in fact, never derived in a consistent manner) in the analysis of the nonlinear interface dynamics under the condition of Fermi resonance , arrays of weakly interacting quadratic waveguides , and beam propagation in nonlinear lattices .
### B Discrete gap solitons
Now we use Eqs. (9) to find the stationary localized modes of nonlinear superlattices, or discrete gap solitons. A similar problem was earlier analyzed in Ref. in the so-called continuum limit, when the modes become wide and are effectively supported by many interfaces, with the excitation profiles approaching those of quadratic solitons . On the other hand, it has been demonstrated that discrete states in a closed system of few interfaces can have different topologies and possess quite peculiar properties . For the case at hand, when the number of interfaces is infinite (e.g. much larger than the characteristic mode width), the different types of highly localized waves have been identified , and their profiles were described by approximate analytical solutions. However, until now the transitional case of moderately localized modes has not been addressed. Thus, we develop a more complete analytical description of discrete gap solitons, which can predict their properties in all the parameter regions.
In order to find approximate solutions for highly localized modes, we use the variational method. First, we have to choose the trial functions. We use the fact that in the high localization limit the tails are almost linear, so that the amplitudes decay nearly exponentially. Then, we introduce two sets of trial functions to account for different topologies : odd modes, when the center of symmetry is located at a layer,
$$U_n^{(\mathrm{o})}=U_0s_1^{|n|}e^{\rho _1|n|},V_n^{(\mathrm{o})}=V_0s_2^{|n|}e^{\rho _2|n|},$$
(12)
and even modes, when the center of symmetry is located between two neighboring layers,
$$\begin{array}{c}U_n^{(\mathrm{e})}=\{\begin{array}{c}U_0s_1^{|n|}e^{\rho _1|n|},n0,\hfill \\ U_0ts_1^{|n+1|}e^{\rho _1|n+1|},n<0,\hfill \end{array}\hfill \\ V_n^{(\mathrm{e})}=\{\begin{array}{c}V_0s_2^{|n|}e^{\rho _2|n|},n0,\hfill \\ V_0s_2^{|n+1|}e^{\rho _2|n+1|},n<0.\hfill \end{array}\hfill \end{array}$$
(13)
Here the parameters $`s_j=\pm 1`$ are introduced to describe unstaggered and staggered profiles, and $`t=\pm 1`$ to produce either untwisted or twisted modes (for the signs ”$`+`$” or ”$``$”, respectively).
After selecting the mode topology and fixing the values of $`s_j`$ and $`t`$, the unknown values $`U_0`$, $`V_0`$, $`\rho _j`$ ($`\rho _j>0`$) are determined by minimizing the Lagrangian corresponding to Eqs. (9). Details of these calculations will be presented elsewhere , here we give a brief summary of the main results, and discuss their physical consequences.
As the system (9) possesses the symmetries (i) $`\chi _1\chi _1`$, $`s_1s_1`$, $`\eta _1\eta _1`$, and (ii) $`\chi _2\chi _2`$ and $`V_nV_n`$, we consider, without a lack of generality, the case $`\chi _j=1`$. The analysis shows that localized solutions exist only if $`s_1\eta _1<2`$. The latter condition means that the FF component is unstaggered for $`\eta _1<2`$, and staggered, otherwise ($`\eta _1>2`$). Similarly, we consider only the case $`|\eta _2|>2`$, as for other values the localized solutions are unstable due to resonant interaction with linear waves .
Analyzing the linear problem, it is straightforward to see that the SH mode can be staggered only if $`\eta _2>2`$. Then, we distinguish between two limits: (i) a strongly localized FF mode ($`\eta _12`$), when the SH consists of staggered linear tails , and (ii) the cascading limit ($`\eta _22`$), when the SH profile is unstaggered and can be found as $`V_nU_n^2/\eta _2`$, resulting in an effectively cubic nonlinearity for the FF wave (see also ). In the intermediate case, a transition of the SH profile between staggered and unstaggered topologies should be observed. Indeed, our variational calculations predict that the SH is staggered for $`2<\eta _2<\eta _{22}`$, and unstaggered for $`\eta _2>\eta _{22}`$. Here, the critical parameter value depends on $`\eta _1`$: for odd modes, it is found from the quadratic equation, $`\sqrt{\eta _{22}}+1/\sqrt{\eta _{22}}=|\eta _1|`$, and a similar relation holds for even modes $`\sqrt{\eta _{22}+1}+1/\sqrt{\eta _{22}+1}=|\eta _1|`$. The topology does not change sharply as the parameter $`\eta _2`$ crosses the critical value $`\eta _{22}\left(\eta _1\right)`$, but numerical calculations confirm that such a transition occurs in a region close to the separation line.
We performed numerical analysis and found that the approximate variational solutions provide close matching for the highly localized profiles, i.e. for relatively large values of $`|\eta _j|`$. Examples of odd and even modes are presented in Figs. 4 and 5, respectively. We see that the profiles of staggered modes, supported by only a few interfaces, are described very accurately. However, for wider modes the deviations between the exact numerical and variational solutions are more pronounced, see Figs. 4(c) and 5(c). The limitation of the variational solution is due to the specific choice of the trial functions (12),(13), which are not suitable for description of “moderately” localized waves with smoother profiles.
It can be demonstrated that in the continuum limit ($`\eta _j2^{}`$) untwisted modes acquire the profiles of quadratic solitons , that can be well approximated with the $`\mathrm{sech}`$-type functions . A special quasi-continuous approach, which allows to determine the mode profiles as a soliton bound state can also be developed. The resulting approximate solutions provide very good estimates, and they are very useful for understanding the mode scaling properties, i.e. a change from broad solitons to narrow highly localized states. On the other hand, it should be noted that the twisted modes do not exist close to the continuum limit, because their profiles are intrinsically discrete due to a sharp amplitude change between the layers at the mode center. A comprehensive description of these results goes beyond the scope of the present paper, and will be presented elsewhere .
## IV A single $`\chi ^{(2)}`$ nonlinear layer embedded in a periodic structure
Let us now consider a special situation, when there exists only a single nonlinear layer located at $`x_0=0`$ embedded into a linear grating. A similar problem has been considered in our recent paper for the case when the linear media on both sides of the nonlinear interface are uniform. Here, we generalize those results to the case of non-uniform linear media, considering a nonlinear layer with quadratic nonlinearity embedded into a periodic structure. Therefore, we modify the model (2) as follows: (i) nonlinear coupling terms are taken into account only for $`n=0`$, and (ii) a linear response of the central layer is assumed to be different from that of other layers, i.e. we change $`\beta _j\alpha _j+\beta _j`$ at $`n=0`$. In connection with the previous problem (see Secs. II and III), the case with $`\alpha _j=0`$ corresponds to the limit of a highly localized mode, when the mode width is much smaller than the distance in between the nonlinear layers (see Fig. 2).
Following the general approach outlined in the previous section, we first analyze the linear properties. We also use the similar notations, but for a single nonlinear interface we omit the index $`n=0`$. From the theory of linear structures, it follows that a link between the linear-wave amplitudes can be characterized by the reflection coefficients, $`r_j^+=b_j^+/a_j^+`$ and $`r_j^{}=a_j^{}/b_j^{}`$. We do not assume that the linear structure is symmetric \[in general $`\nu _j(x)\nu _j(x)`$\] and denote with $`+`$ and $``$ the wave characteristics at the right and left boundaries of the nonlinear layer. If the linear structure is periodic, the coefficients $`r_j^\pm `$ can be found by solving the following eigenvalue problems:
$$T^{(j,\pm )}\left(\begin{array}{c}1\hfill \\ r_j^\pm \hfill \end{array}\right)=\tau ^{(j,\pm )}\left(\begin{array}{c}1\hfill \\ r_j^\pm \hfill \end{array}\right).$$
(14)
Here $`T^{(j,\pm )}`$ is the transfer matrix of one linear segment in a periodic lattice, starting on the right ($`+`$) or the left ($``$) side of a nonlinear interface, and it can be calculated using Eqs. (5)-(8). Then, we notice that the determinants are: $`T_\mathrm{p}^{(j,m,\pm )}=1`$, $`T_\delta ^{(j,m,\pm )}=1`$, $`T_\mathrm{r}^{(j,m,\pm )}=\mu _{j,m}^\pm /\mu _{j,m+1}^\pm `$, and thus $`T^{(j,\pm )}1`$. Using this relation, we solve Eq. (14) and find the values of the reflection coefficients:
$$r_j^\pm =T_{[2,1]}^{(j,\pm )}\left(\tau ^{(j,\pm )}T_{[2,2]}^{(j,\pm )}\right)^1,$$
(15)
where
$`\tau ^{(j,\pm )}=(\eta _j^\pm /2)\left[1\sqrt{1\left(2/\eta _j^\pm \right)^2}\right].`$
It can be proved that $`\eta _j^\pm `$ are real (see Appendix A). Then, we see that the modes decay at infinity, i.e. $`|\tau ^{(j,\pm )}|<1`$, only if $`|\eta _j^\pm |>2`$. This condition, which has already been mentioned in Sec. III, defines a band gap, when the wave propagation is prohibited for a specific range of the propagation constant $`\lambda _j`$ (see an example in Fig. 6).
By applying the continuity conditions, $`c_j=a_j^\pm +b_j^\pm `$, we obtain the expressions for the amplitudes at the nonlinear layer: $`|c_1|^2=\stackrel{~}{\alpha }_1\stackrel{~}{\alpha }_2`$, $`c_2=\stackrel{~}{\alpha }_1`$, where $`\stackrel{~}{\alpha }_1=\alpha _1\zeta _1^+\zeta _1^{}`$, $`\stackrel{~}{\alpha }_2=\alpha _2\sigma \zeta _2^+\sigma \zeta _2^{}`$, and $`\zeta _j^\pm =\mu _j^\pm \left(1r_j^\pm \right)/\left(1+r_j^\pm \right)`$ are the effective transverse wave numbers at the right ($`+`$) and left ($``$) boundaries of the nonlinear layer; their values are determined by the dispersion relation of the periodic linear gratings. As has been demonstrated in the previous section, the wave numbers $`\mu _j^\pm `$, the linear transfer matrices, and, according to Eq. (15), the reflection coefficients $`r_j^\pm `$ depend on the propagation constants $`\lambda _j`$, which, in turn, are related by the phase-matching condition $`\lambda _2=2\lambda _1`$. Thus, for fixed physical characteristics, the localized modes constitute a one-parameter family, and we choose $`\lambda _1`$ as a free parameter. We note that the coefficients $`\zeta _1^\pm `$ are real in the band gap (see Appendix B).
In order to illustrate the features of nonlinear modes, we consider a structure similar to that used in experiments on the SH generation in localized modes . In the experimental setup, a periodic linear grating was built of two materials with different refractive indices, characterized by the parameters $`\nu _{j,\pm }`$, with corresponding finite widths of the layers, $`h_\pm `$. The nonlinear interface was created by cutting the grating in two parts and coating the interface to enhance the effective quadratic nonlinearity. Characteristics of the defect layer were controlled by adjusting the gap. An example of the band gap for such a photonic structure is presented in Fig. 6. Note that because the linear structures on either sides of the nonlinear layer are chosen to be the same \[up to a constant shift, in our notation $`\nu _{j,}(x)=\nu _{j,+}(x+h_+)`$\], the corresponding gaps coincide.
Many of the properties of the localized states can be understood by analyzing the power diagram $`P\left(\lambda _1\right)`$. Characteristic examples of such a dependence are shown in Figs. 7(a,b,c). Similar to the case of homogeneous linear media , there always exists a branch in a parameter region unbounded from above for $`\lambda _1`$ larger than some critical value. Quite remarkably, the corresponding mode properties are very similar to those of quadratic solitons . In particular, for large values of the propagation constant $`\lambda _1`$ the power dependence $`P\left(\lambda _1\right)`$ always has a positive slope, while for smaller $`\lambda _1`$ the slope can be negative, resulting in bistability. Such a case is demonstrated in Fig. 7(a). We found that stability of the corresponding modes can be determined by the Vakhitov-Kolokolov criterion, i.e. the localized modes are stable provided $`dP/d\lambda _1>0`$, and unstable, otherwise.
On the other hand, the spectrum of a linear periodic structure consists of several bands. Moreover, even inside a band gap the modes cannot exist if the condition $`\stackrel{~}{\alpha }_1\stackrel{~}{\alpha }_2>0`$ is not satisfied. Thus, in a sharp contrast with two-color parametric solitons in homogeneous media, other branches can appear for smaller values of $`\lambda _1`$, and we found that the localized mode properties can be very different compared to the modes corresponding to the right branch. For example, for Figs. 7(b,c), the left branches at higher intensities correspond to smaller $`\lambda _1`$ and wider profiles. This happens because the SH amplitude is negative, $`c_2<0`$, which results in effectively self-defocusing nonlinear response. It is interesting to note that similar types of power dependencies occur for a layer possessing a self-defocusing Kerr-type nonlinearity , and the corresponding localized modes are stable. We have performed a linear stability analysis and have found that, similar to the case of the Kerr-type nonlinearity, the Vakhitov-Kolokolov type criterion cannot be applied to determine stability of such modes existing in an effectively defocusing medium. However, our calculations demonstrate that parametric resonances can lead to oscillatory instability, as illustrated in Figs. 7(b,c). A comprehensive discussion of the mode stability will be presented elsewhere.
The two-color modes can be generated by launching a localized FF wave at the interface, as shown in Fig. 8(a). We have also studied the evolution of perturbed unstable modes, which can evolve toward a stable state. Example in Fig. 8(b) demonstrates the development of an oscillatory instability with subsequent switching to a stable branch.
## V Conclusions
We have developed a general formalism for analyzing spatially localized nonlinear modes (discrete gap solitons) in periodic photonic structures with embedded quadratic (or $`\chi ^{(2)}`$) nonlinear interfaces — nonlinear quadratic superlattices. Our approach can be applied to different types of periodic linear media with isolated or periodic nonlinear interfaces, where nonlinearity can support parametric interaction and generation of the second-harmonic field. In the case of a nonlinear superlattice, i.e. periodically spaced layers with quadratic nonlinear response, we have derived an effective discrete model. For such a periodic structure, we have found discrete gap solitons of different topologies in the form of fundamental and second-harmonic fields coupled parametrically at the nonlinear interfaces.
For a single nonlinear layer embedded in a linear periodic medium, we have described a novel type of nonlinear localized defect mode — a two-color photonic mode. Some of the properties of these two-color localized modes, such as stability, generation, and switching, have been shown to be remarkably similar with those of quadratic parametric solitons in homogeneous media. However, we have revealed a number of specific properties of such modes, and also demonstrated their generation from a localized beam of the fundamental frequency, as well as switching from an unstable to a stable state.
We believe our results are important, on one hand, for the theory of nonlinear photonic crystals where nonlinearities appear due to phase-matched harmonic generation, and, on the other hand, for creating tunable band gap materials where gaps could be opened or closed depending on the input intensity.
## Acknowledgments
The authors are indebted to F. Lederer, N. N. Rosanov, and R. Vilaseca for useful comments. Costas Soukoulis acknowledges the hospitality of the Optical Sciences Centre. The work has been partially supported by the Planning and Performance Fund of the Institute of Advanced Studies, by the Australia-Denmark collaborative project of the Department of Industry, Science, and Tourism, by the Australian Photonics Cooperative Research Centre, and by the Danish Technical Research Council under Talent Grant No. 9800400.
## Appendix A: Transfer matrix properties
In order to demonstrate some features of linear modes existing in periodic structures, we consider the properties of the corresponding transfer matrices. First, we see that a matrix describing changes of the wave amplitudes at the boundary between linear layers, $`T_\mathrm{r}`$, depends only on the wave numbers on either side of the interface, as follows from Eq. (6). It is also easy to check that the following relation holds:
$$\begin{array}{c}T_\mathrm{r}^{(j,m)}T_\mathrm{r}(\mu _{j,m+1},\mu _{j,m})=\hfill \\ T_\mathrm{r}(\mu _{j,m+1},\stackrel{~}{\mu })T_\mathrm{r}(\stackrel{~}{\mu },\mu _{j,m}),\hfill \end{array}$$
(16)
where $`\stackrel{~}{\mu }`$ is arbitrary, and we assume that it is real. Let us introduce a new matrix $`\stackrel{~}{T}^{(j)}=T_\mathrm{r}(\stackrel{~}{\mu },\mu _{j,M+1})T^{(j)}T_\mathrm{r}(\mu _{j,1},\stackrel{~}{\mu })`$, and use Eq. (16) to present it in a special form:
$`\stackrel{~}{T}^{(j)}=\stackrel{~}{T}_\delta ^{(j,M+1)}\stackrel{~}{T}_\mathrm{p}^{(j,M)}\mathrm{}\stackrel{~}{T}_\mathrm{p}^{(j,1)}.`$
Here $`\stackrel{~}{T}_\mathrm{p}^{(j,n)}T_\mathrm{r}(\stackrel{~}{\mu },\mu _{j,n})T_\mathrm{p}^{(j,n)}T_\mathrm{r}(\mu _{j,n},\stackrel{~}{\mu })`$ and $`\stackrel{~}{T}_\delta ^{(j,n)}T_\mathrm{r}(\stackrel{~}{\mu },\mu _{j,n})T_\delta ^{(j,n)}T_\mathrm{r}(\mu _{j,n},\stackrel{~}{\mu })`$. It can be verified by direct substitution that these matrices are real (we use the fact that, for stationary modes, $`\mu _{j,n}`$ are real or purely imaginary), and therefore $`\stackrel{~}{T}^{(j)}`$ is real as well. Finally, we find that $`\mathrm{Tr}(T^{(j)})\mathrm{Tr}(\stackrel{~}{T}^{(j)})`$ and $`T_s^{(j)}/\mu _{j,1}\stackrel{~}{T}_s^{(j)}/\stackrel{~}{\mu }`$, which proves that coefficients $`\xi _j`$ and $`\eta _j`$ in Eq. (9) are real.
## Appendix B: Reflection coefficients for modes in a band gap
Although it is possible to extend the technique presented in Appendix A to prove the properties of the reflection coefficients in the case of an infinite linear grating (introduced in Sec. IV), here we employ a different approach. We note that for stationary waves in a band gap there should be no energy flow along the $`x`$ axis. The corresponding restrictions follows from Eq. (2):
$$\mathrm{Im}\left(u\frac{u^{}}{x}\right)=0,\mathrm{Im}\left(v\frac{v^{}}{x}\right)=0,$$
(17)
These conditions are satisfied when the amplitudes of the counter-propagating waves coincide, i.e. $`|r_j^\pm |=1`$ if $`\mu _j^\pm `$ is imaginary. On the other hand, for the layers with real $`\mu _j^\pm `$ the linear modes should be in phase, i.e. $`\mathrm{Im}(r_j^\pm )0`$. Then, it immediately follows that in a band gap the coefficients $`\zeta _j^\pm `$ are real.
|
warning/0005/hep-ex0005022.html
|
ar5iv
|
text
|
# Cornering Higgs Bosons at LEP aafootnote aTalk given at the XXXV^"th" Rencontres de Moriond QCD and Hadronic interactions.
## 1 The Standard Model
In the standard model, the spontaneous breaking of the electroweak symmetry is achieved at the expense of the introduction of multiplets of complex scalar fields in self-interaction. In particular, in its minimal version, a single scalar doublet is required (with the two fold advantage of simplicity and of preserving $`\rho _o=1`$). As it develops a vacuum expectation value, gauge bosons acquire their masses absorbing three of the four initial degrees of freedom and a massive scalar physical state remains, the Higgs boson h. Its mass is a free parameter of the theory as a function of which its production and decays are predicted unambiguously.
At LEP, the minimal standard model Higgs boson dominant production process is the Higgs-strahlung ($`\mathrm{e}^+\mathrm{e}^{}\mathrm{HZ}`$) and its prominent decay is into a pair of b quarks ($``$ 81% for a 105 $`\mathrm{GeV}/c^2`$ Higgs boson). Four search topologies arise from the decays of the accompanying Z boson. 1) Four jets, when the Z decays into a pair of quarks; 2) two acoplanar jets and missing energy, when the Z decays to a pair of neutrinos; 3) the leptonic channel, when the Z decays to a pair of electrons or muons; and 4) the $`\tau \tau \text{q}\overline{\text{q}}`$ and $`\text{b}\overline{\text{b}}\tau \tau `$ channels, where either the Z or the Higgs boson decay to a pair of taus. In all these channels but the latter, the Higgs boson is searched for in its $`\text{b}\overline{\text{b}}`$ decay mode. The b jet tagging is therefore an essential component of all the analyses.
All the data collected have been analysed by all four experiments $`^\mathrm{?}`$, nevertheless the most significant contribution for the limit are the $``$40$`\mathrm{pb}^1`$/experiment collected at 202 GeV. The distribution of the reconstructed mass of the Higgs boson for all channels and all experiments together is illustrated in Fig. 1. Only the most signal-like events are chosen with the further requirement that the contribution from all experiments be roughly the same.
The observed LEP combined $`CL_b`$ (which estimates the probability that the data be compatible with the background) $`^\mathrm{?}`$ as a function of the Higgs boson mass hypothesis is illustrated in Fig. 2a. The expected values of $`CL_b`$ in the presence of signal are also shown in Fig. 2a, yielding a 5$`\sigma `$ discovery sensitivity of 106.3 $`\mathrm{GeV}/c^2`$(corresponding to $`CL_b5.7\times 10^7`$).
In Fig. 2b the $`CL_s`$ (which estimates the probability that the data be compatible with background and signal) $`^\mathrm{?}`$, computed without taking into account the correlations among systematic errors, is illustrated as a function of the Higgs boson mass hypothesis. Hypotheses yielding a $`CL_s<5`$% are excluded at the 95% confidence level. The sensitivity of the combination is 109.1 $`\mathrm{GeV}/c^2`$. According to Fig. 2b, Higgs boson mass hypotheses below 107.9 $`\mathrm{GeV}/c^2`$ are excluded at the 95% CL. However the effect of correlations of the systematic errors result in a downward shift of 150 MeV/c<sup>2</sup> of the limit, leading to combined lower limit on the mass of the Minimal Standard Model Higgs boson of 107.7 $`\mathrm{GeV}/c^2`$.
## 2 Extending the Higgs Sector of the Standard Model
The simplest extension of the minimal standard model Higgs sector is obtained with the introduction of an additional doublet of complex scalar fields (still preserving $`\rho _o=1`$). In two-higgs-doublet models (2HDM) FCNC can become important, but can be avoided if all fermions of a given charge do not couple to more than one Higgs doublet. The two most popular models obeying this requirement are the model I, where one Higgs doublet couples only to fermions and the other to bosons, and the model II where one Higgs doublet couples only to up-type fermions and the other to down-type fermions. In both cases, the electroweak symmetry breaking leaves five massive scalar physical states among which two are neutral CP-even (h and H), one is neutral CP-odd (A) and two are charged (H<sup>±</sup>). These models are the main motivation for the search for Higgs bosons decaying mostly to photons (model I) and for charged Higgs bosons.
### 2.1 2HDM of type I, fermiophobia
In 2HDM of type I when the CP-even Higgs bosons do not mix, the lightest CP even Higgs boson does not couple to fermions. In a Higgs boson mass range well below 2$`m_{\mathrm{W}^\pm }`$ it predominantly decays to a pair of photons. The topologies arising from such models are similar to those of the minimal standard model Higgs boson searches where a pair of photons is substituted for the pair of b-quark jets. Such topologies were searched for by all four experiments. No evidence for a resonant production of a pair of photons was found as illustrated in Fig. 3a. The negative result of these searches can be interpreted in a quasi-model-independent fashion as an upper limit on the production rate of a Higgs boson decaying to a pair of photons normalised to the standard model Higgs boson production, as a function of the Higgs boson mass hypothesis. The ALEPH exclusion $`^\mathrm{?}`$ is illustrated in Fig. 3b. The intercept of the model independent exclusion and the branching fraction of the fermiophobic Higgs boson into a pair of photons, which starts dropping severely above 100 $`\mathrm{GeV}/c^2`$ due to the growing contribution of $`HWW^{}`$, gives the lower mass limit of the fermiophobic Higgs boson. A fermiophobic Higgs boson lighter than 101.0 $`\mathrm{GeV}/c^2`$ is excluded by ALEPH with a sensitivity of 100.2 $`\mathrm{GeV}/c^2`$. The other experiments $`^{\mathrm{?},\mathrm{?},\mathrm{?}}`$ find similar results —DELPHI, L3 and OPAL exclude fermiophobic Higgs boson masses below 98.0 $`\mathrm{GeV}/c^2`$, 98.8 $`\mathrm{GeV}/c^2`$ and 97.8 $`\mathrm{GeV}/c^2`$ respectively.
### 2.2 2HDM, Charged Higgs bosons
Charged Higgs bosons are produced at LEP via $`\gamma `$ and Z bosons exchange in the s-channel. The dominant decay modes are $`\text{H}^\pm \tau \nu _\tau ,\text{cs}`$. The relative contribution of these two modes depend on the model parameter $`\mathrm{tan}\beta `$ (the ratio of the vev’s of the two Higgs doublets). Analyses must therefore cover all possible final states arising from charged Higgs bosons production, i.e., $`\tau ^{}\overline{\nu }_\tau \tau ^+\nu _\tau `$, $`\tau ^{}\overline{\nu }_\tau c\overline{s}`$ and $`c\overline{s}s\overline{c}`$. Dedicated searches for each of these topologies were performed by all LEP experiments $`^\mathrm{?}`$. No evidence for a signal was found, as illustrated in Fig. 4a for the fully hadronic and semi-leptonic channels. (The charged Higgs boson mass cannot be reconstructed in the fully leptonic channel.) These results can be interpreted in terms of 95% CL charged Higgs boson mass lower limits as a function of the leptonic branching fraction. As shown in Fig. 4b a lower charged Higgs boson mass limit on of 78.6 $`\mathrm{GeV}/c^2`$ with a sensitivity of 78.0 $`\mathrm{GeV}/c^2`$ can be set independently of the $`\tau \nu _\tau `$ branching fraction.
## 3 Minimal Supersymmetric extension of the Standard Model (MSSM)
The presence of Higgs bosons in the standard model is quite unnatural, due to the quadratically divergent contributions of the radiative corrections to their masses. Supersymmetry elegantly solves this problem and provides a framework for unified theories and quantum gravity.
In its minimal version, the Higgs sector of the supersymmetric standard model is a type II 2HDM. Supersymmetry constrains Higgs bosons masses. At tree level $`m_{\mathrm{H}^\pm }m_{\mathrm{W}^\pm }`$ and $`m_\mathrm{h}m_\mathrm{Z}|\mathrm{cos}2\beta |`$. In the MSSM, two free parameters, commonly chosen either as $`m_\mathrm{A}`$ and $`\mathrm{tan}\beta `$ or $`m_\mathrm{h}`$ and $`\mathrm{tan}\beta `$, are required at tree level to describe the Higgs sector. However radiative corrections due to the large Yukawa coupling to the top quark can increase the theoretical upper bound on the lightest CP-even Higgs boson up to about 130 $`\mathrm{GeV}/c^2`$. All the parameters relevant to the radiative corrections to the Higgs boson masses are chosen to maximise $`m_\mathrm{h}`$ ($`m_\mathrm{h}`$-max scenario).
The Higgs-strahlung production of h is reduced, compared to the standard model, by a factor $`\mathrm{sin}^2(\beta \alpha )`$. The results of searches for the standard model Higgs boson can be turned into an 95% CL exclusion domain in the ($`m_\mathrm{h}`$,$`\mathrm{sin}^2(\beta \alpha )`$) plane as illustrated in Fig. 5.
For small values of $`\mathrm{sin}^2(\beta \alpha )`$ the Higgs-strahlung contribution vanishes but the associated $`\mathrm{e}^+\mathrm{e}^{}\mathrm{hA}`$ production process, which cross section is proportional to $`\mathrm{cos}^2(\beta \alpha )`$, becomes dominant. In this regime, the h and A Higgs bosons are roughly mass degenerate and both decay predominantly to a pair of b-quarks. This process is thus searched for in two additional topologies $`^\mathrm{?}`$. 1) Four jets likely to originate from b quarks; and 2) Two b-like jets and a pair of taus. No evidence for a signal was found as illustrated in Fig. 6a. The negative result of these searches combined with the search for the standard model Higgs boson can be interpreted in terms of an exclusion domain in the ($`m_\mathrm{h}`$,$`\mathrm{tan}\beta `$) plane as shown in Fig. 6b. Neutral Higgs boson masses in excess of 88.3 $`\mathrm{GeV}/c^2`$ are excluded at the 95% CL with a sensitivity of 90.5 $`\mathrm{GeV}/c^2`$. The lower mass limit on the CP-odd Higgs boson is set to 88.4 $`\mathrm{GeV}/c^2`$ with a sensitivity of 91.1 $`\mathrm{GeV}/c^2`$. These limits have proven to be robust while scanning over all parameters pertaining to radiative corrections to the Higgs boson masses $`^{\mathrm{?},\mathrm{?},\mathrm{?}}`$. As can be seen in Fig. 6b values of $`\mathrm{tan}\beta `$ comprised within \[0.7,1.8\] are excluded at the 95% CL, for m$`_{\text{top}}`$=174.3 $`\mathrm{GeV}/c^2`$ and \[0.8,1.5\] for m$`_{\text{top}}`$=179.4 $`\mathrm{GeV}/c^2`$.
## 4 Further Exotism
### 4.1 Invisible decays of Higgs bosons
In a MSSM conserving R-parity, the domains in which Higgs bosons can decay to neutralinos (supersymmetric partners of the neutral Higgs and gauge bosons, here the lightest neutralino is assumed to be the lightest supersymmetric particle) is tightly constrained by direct searches for charginos (supersymmetric partners of charged Higgs and gauge bosons). Nevertheless the latter exclusion assumes universality of gaugino masses at a large unification scale. If this assumption is relaxed the indirect chargino constraints do not necessarily hold and the Higgs boson can thus decay to a pair of neutralinos. Neutralinos are weakly interacting neutral particles, like neutrinos. They therefore escape detection and the Higgs boson decays invisibly. A variety of other theories, from models involving Majorons to large extra dimensions, predict such decays.
The two main topologies under consideration correspond to the Z decays to a pair of quarks and to a pair of leptons (either an electron or a muon). 1) Two acoplanar jets; and 2) two leptons, both with missing energy. Such topologies were searched for by all four LEP collaborations. No evidence for a signal was found, as shown in Fig. 7a for ALEPH data only $`^\mathrm{?}`$. This result can then be interpreted in a quasi-model-independent fashion as a limit on the production rate of an invisible Higgs boson normalised to the standard model cross section: $`\xi ^2=Br(\mathrm{h}\mathrm{invisible})\times (\sigma _{hZ}/\sigma _{HZ}^{\mathrm{MSM}})`$ as a function of $`m_\mathrm{h}`$ as illustrated in Fig. 7b. Higgs bosons decaying invisibly and produced at standard model rate ($`\xi ^2=1`$) with masses below 106.4 $`\mathrm{GeV}/c^2`$ are excluded at 95% CL with a sensitivity of 103.9 $`\mathrm{GeV}/c^2`$. The other experiments find similar results —93.8 $`\mathrm{GeV}/c^2`$, 100.5 $`\mathrm{GeV}/c^2`$ and 94.4 $`\mathrm{GeV}/c^2`$ are the limits (for $`\xi ^2=1`$) set by DELPHI (with data up to 189 GeV), L3 and OPAL respectively— (unless their high energy data set had not yet been entirely analysed) $`^{\mathrm{?},\mathrm{?},\mathrm{?},\mathrm{?}}`$.
### 4.2 Anomalous Higgs couplings
The effect of theories which supersede the standard model at some large scale $`\mathrm{\Lambda }`$ can also be parametrised at low energy by an effective Lagrangian in a model independent manner. The simplest corrections to the standard model Lagrangian distorting the couplings of
Higgs bosons originate from terms of the type $`_{\mathrm{eff}}=_n(f_n/\mathrm{\Lambda }^2)𝒪_n`$ where the $`𝒪_n`$ operator involves vector boson and/or Higgs boson fields with couplings $`f_n`$. Such terms can give rise to anomalous couplings of the type $`g_{\text{H}\gamma \gamma }`$, $`g_{\text{HZ}\gamma }`$ and $`g_{\text{HZZ}}`$ which affect the expected phenomenology of a standard Higgs sector. For instance the Higgs boson can be produced along with a photon and decay itself to a pair of photons. The DELPHI collaboration performed a searched for topologies with three photons in the framework of anomalous couplings of Higgs bosons $`^\mathrm{?}`$. No evidence for a signal was found and limits on the generic $`F`$ coupling —here all the underlying couplings $`f_n`$ relevant for the Higgs anomalous couplings are assumed to be equal to $`F`$— can be set as a function of $`m_\mathrm{h}`$ as shown in Fig. 8 (region A).
In this general framework, the Higgs boson can also predominantly couple to photons but still be produced via the Higgs-strahlung process and thus give rise to the topologies searched for in the framework of fermiophobicity. The ALEPH searches presented in Section 2.1 were reinterpreted here in terms of an exclusion of the coupling $`F`$ as a function of the Higgs boson mass hypothesis, as illustrated in Fig. 8 (region C).
These searches require non-zero anomalous couplings and can therefore not explore the small $`F`$ region. However the searches for the standard model Higgs boson can also be reinterpreted in this more general framework as shown in Fig. 8 (region B) where the ALEPH standard model Higgs boson searches are used. Finally a combination of all these searches allow a mass lower limit to be set at $`106.7`$ $`\mathrm{GeV}/c^2`$ on the mass of a Higgs boson allowing it to couple anomalously to photons with the assumption that all relevant couplings $`f_n`$ are equal.
## 5 Summary and prospects
All LEP experiments have made meaningful efforts to corner all possible forms in which Higgs bosons might appear. All the topologies and the relevant theoretical frameworks which were investigated with the data collected in 1999 have been presented. The resulting mass lower limits on the mass of Higgs bosons in each of these models and all the topologies are sketched in Table 1.
Topologies not covered in this review are those searched for in flavour-independent analyses performed by OPAL $`^\mathrm{?}`$ (interpreted in general 2HDM models) and ALEPH $`^\mathrm{?}`$ (to exclude pathological MSSM parameter sets in their scan beyond benchmark configurations).
For its presumably final year of running LEP aims at collecting data at energies up to 207 GeV (optimistically up to 209 GeV). The foreseen sensitivity of the standard model Higgs boson searches are expected to be between 112 and 115 $`\mathrm{GeV}/c^2`$. The 3$`\sigma `$ evidence sensitivity for a standard model Higgs boson lies 300 MeV/c<sup>2</sup> below the 95% CL exclusion and the 5$`\sigma `$ discovery is 2 $`\mathrm{GeV}/c^2`$ below this limit. A sizeable window is thus still open at LEP, in a region where, if it exists, the standard model Higgs boson is most expected $`^\mathrm{?}`$.
## Acknowledgements
It is a pleasure to thank the organisers and the secretariat of the XXXV<sup>th</sup> Rencontres de Moriond for their wonderful hospitality. I would like to thank all my colleagues from the four LEP collaborations and especially Vanina Ruhlman Kleider, Sofia Andringa Dias, André Tilquin, Jean-Baptiste de Vivie, Patrick Janot, Satoru Yamashita, Peter Igo-Kemenes and the LEP Higgs working group for their help in preparing this talk. I also wish to express my gratitude to Gaëlle Boix, Jean-Ba and Patrick for their careful reading of these proceedings.
## References
|
warning/0005/astro-ph0005480.html
|
ar5iv
|
text
|
# Irradiation and mass transfer in low-mass compact binaries
## 1 Introduction
It is only a few years since the realization that irradiation of the donor star of a semi-detached binary by accretion luminosity generated in the vicinity of a compact accretor can have far-reaching consequences for the long-term evolution of such binary systems. Podsiadlowski (1991), treating irradiation in spherical symmetry, first addressed this problem in the context of the evolution of low-mass X-ray binaries (LMXBs). Later, Gontikakis and Hameury (1993), Hameury et al. (1993), and Ritter (1994) pointed out that spherically symmetric irradiation is probably not an adequate approximation for the case in point. Preliminary calculations by Ritter (1994) showed that the reaction of low-mass stars to anisotropic irradiation is qualitatively (and quantitatively) different from that to isotropic irradiation. Subsequently, King et al. (1995, 1996, 1997), hereafter KFKR95, KFKR96 and KFKR97, Ritter, Zhang and Kolb (1995, 1996), Ritter, Zhang and Hameury (1996), Hameury and Ritter (1997), hereafter HR97, and McCormick and Frank (1998), hereafter MF98, have dealt with the case of anisotropic irradiation in various ways and detail. In most of the above-quoted papers reference was made to a paper by the present authors in which a detailed and systematic treatment of the basic properties of anisotropic irradiation of low-mass stars was to be found. The preparation of this material has been much delayed. The main purpose of the present paper is to close this gap in the recent literature, and to provide the base for further research.
For this purpose we shall discuss in section 2 in some detail the basic concepts of computing the mass transfer rate in a semi-detached binary with and without irradiation, and present observational evidence and theoretical arguments in support of the notion of anisotropic irradiation. In section 3 we shall present a simple analytic calculation which shows that the reaction of a low-mass star to anisotropic irradiation is qualitatively and quantitatively different from what one obtains in the isotropic case. We show also that the simple analytic results are fully supported by corresponding calculations of full 1D stellar models (hereafter simply referred to as full stellar models). In section 4 we study the stability of mass transfer in binaries in which the donor star is exposed to irradiation generated by the accreting companion. In section 5 we shall present a more detailed semi-analytic model than the one in Sect. 3 for the reaction of a low-mass star to anisotropic irradiation in the limit of small fluxes. The implications of the instability against irradiation-induced runaway mass transfer for the long-term evolution of CVs and LMXBs is discussed in section 6. In section 7 we shall present and discuss results of numerical computations of the secular evolution of CVs subject to the irradiation instability. A summary of our main results and our main conclusions are given in the final section 8.
## 2 Computing the mass transfer rate
In the context of this paper, compact binaries are either cataclysmic variables (CVs) or LMXBs, i.e. systems consisting of a low-mass star (the secondary) with a mass $`M_s1\text{M}_{}`$ which fills its critical Roche lobe and transfers matter to a compact companion (the primary), either a white dwarf (in CVs), or a neutron star or a black hole (in LMXBs). The secular evolution of such systems is a consequence of mass loss from the secondary which, in turn, is driven either by the secondary’s nuclear evolution or by loss of orbital angular momentum and possibly other mechanisms such as irradiation on which this paper focuses. In the standard picture of the secular evolution see (e.g. King 1988; Kolb and Ritter 1992, hereafter KR92; Ritter 1996) the nature of the compact star is of no importance, i.e. the star is considered as a point mass (of mass $`M_c`$). The nature of the compact component is, however, of importance for accretion phenomena occurring in such systems, e.g. dwarf nova and classical nova outbursts in CVs and X-ray bursts in LMXBs, and when illumination of the secondary by radiation generated through accretion is considered, as we shall do in the following. Thus, calculating the standard secular evolution of such a binary reduces to calculating the evolution of a low-mass star under mass loss, where the mass loss rate derives from the boundary conditions imposed by the fact that the star is in a binary. In the simplest case one obtains the mass loss rate $`\dot{M}_s`$ by requiring that the radius of the secondary $`R_s`$ is always exactly equal to its critical Roche radius $`R_R`$. By decomposing the temporal change of $`R_s`$ and $`R_R`$ as (see e.g. Ritter 1988, 1996)
$$\frac{d\mathrm{ln}R_s}{dt}=\zeta _S\frac{\dot{M}_s}{M_s}+\left(\frac{\mathrm{ln}R_s}{t}\right)_{\mathrm{th}}+\left(\frac{\mathrm{ln}R_s}{t}\right)_{\mathrm{nuc}}$$
(1)
and
$`{\displaystyle \frac{d\mathrm{ln}R_R}{dt}}`$ $`=`$ $`\zeta _R{\displaystyle \frac{\dot{M}_s}{M_s}}+\left({\displaystyle \frac{\mathrm{ln}R_R}{t}}\right)_{\dot{M}_s=0}`$ (2)
$`=`$ $`\zeta _R{\displaystyle \frac{\dot{M}_s}{M_s}}+2\left({\displaystyle \frac{\mathrm{ln}J}{t}}\right)_{\dot{M}_s=0}`$
one obtains
$`\dot{M}_s={\displaystyle \frac{M_s}{\zeta _S\zeta _R}}\times `$
$`\left[2\left({\displaystyle \frac{\mathrm{ln}J}{t}}\right)_{\dot{M}_s=0}\left({\displaystyle \frac{\mathrm{ln}R_s}{t}}\right)_{\mathrm{th}}\left({\displaystyle \frac{\mathrm{ln}R_s}{t}}\right)_{\mathrm{nuc}}\right].`$ (3)
Here
$$\zeta _S=\left(\frac{\mathrm{ln}R_s}{\mathrm{ln}M_s}\right)_S$$
(4)
is the adiabatic mass radius exponent of the secondary, and
$$\zeta _R=\left(\frac{\mathrm{ln}R_R}{\mathrm{ln}M_s}\right)_{}$$
(5)
the mass radius exponent of the critical Roche radius, where the subscript * indicates that for evaluating this quantity one has to specify how mass and angular momentum are redistributed in the system. $`J`$ is the orbital angular momentum, $`(\mathrm{ln}R_s/t)_{\mathrm{th}}`$ the rate of change of $`R_s`$ due to thermal relaxation and $`(\mathrm{ln}R_s/t)_{\mathrm{nuc}}`$ the one due to nuclear evolution. The virtue of (3) is that it shows immediately how secular evolution works: If the binary is dynamically stable against mass transfer; i.e. if $`\zeta _S\zeta _R>0`$, then mass transfer must be driven by some mechanism. This can either be the secondary’s expansion due to nuclear evolution or the shrinkage of the orbit due to loss of orbital angular momentum. In the standard evolution of low-mass binaries, mass transfer is usually not driven by thermal relaxation. However, as we shall see below, this is not necessarily true if irradiation of the secondary is taken into account.
Attempts to bring the observed population of millisecond pulsars in line with the death rate of their presumed progenitors, i.e. the LMXBs, have resulted in the speculation that the secular evolution of LMXBs might be drastically different from, and their lifetime much shorter than that of CVs (e.g. Kulkarni and Narayan 1988). A possible reason for this is seen in the fact that the donor in a LMXB is exposed to a high flux of hard X-ray radiation emitted by the accreting compact star. In fact, Podsiadlowski (1991) has shown that irradiating a low-mass main-sequence star ($`M0.8\text{M}_{}`$) spherically symmetrically with a flux $`F_{\mathrm{irr}}10^{11}10^{12}`$erg cm<sup>-2</sup>s<sup>-1</sup> results in an expansion of the star on a thermal time scale and in its gradual transformation into a fully radiative star. It is this expansion which can drive mass transfer on a thermal time scale and this was thought to shorten the lifetime of a LMXB. On the formal level the effect of irradiation is taken into account by replacing (1) by
$`{\displaystyle \frac{d\mathrm{ln}R_s}{dt}}=\zeta _S{\displaystyle \frac{\dot{M}_s}{M_s}}`$
$`+\left({\displaystyle \frac{\mathrm{ln}R_s}{t}}\right)_{\mathrm{ml}}+\left({\displaystyle \frac{\mathrm{ln}R_s}{t}}\right)_{\mathrm{nuc}}+\left({\displaystyle \frac{\mathrm{ln}R_s}{t}}\right)_{\mathrm{irr}}.`$ (6)
Here, $`(\mathrm{ln}R_s/dt)_{\mathrm{ml}}`$ is the thermal relaxation term arising from mass loss alone and $`(\mathrm{ln}R_s/t)_{\mathrm{irr}}`$ the thermal relaxation term caused by irradiation (at constant mass). The latter term arises because irradiating the donor star means that its surface boundary conditions are changed and that it tries to adjust to them on a thermal time scale. With (6) instead of (1) we have now for the mass transfer rate
$`\dot{M}_s`$ $`=`$ $`{\displaystyle \frac{M_s}{\zeta _S\zeta _R}}[2{\displaystyle \frac{\mathrm{ln}J}{t}}\left({\displaystyle \frac{\mathrm{ln}R_s}{t}}\right)_{\mathrm{ml}}`$ (7)
$``$ $`\left({\displaystyle \frac{\mathrm{ln}R_s}{t}}\right)_{\mathrm{nuc}}\left({\displaystyle \frac{\mathrm{ln}R_s}{t}}\right)_{\mathrm{irr}}].`$
From (7) it is clearly seen that if $`(\mathrm{ln}R_s/t)_{\mathrm{irr}}>0`$, irradiation amplifies mass transfer. However, there are limits to what irradiation can do because the time scale and the amplitude of the effect are limited. The secondary’s expansion due to irradiation saturates at the latest when it has become fully radiative, where it is larger by $`\delta \mathrm{ln}R_s`$ than a star of the same mass in thermal equilibrium without irradiation. As we shall show below the time scale $`\tau `$ on which the irradiated star initially expands is of the order of or less than the thermal time scale of the convective envelope $`\tau _{\mathrm{ce}}`$. As a result, we can expect a maximum contribution from irradiation to $`\dot{M}_s`$ at the onset of irradiation (i.e. of mass transfer) which is of the order
$$(\delta \dot{M}_s)_{\mathrm{irr}}=\frac{M_s}{\zeta _S\zeta _R}\left(\frac{\mathrm{ln}R_s}{t}\right)_{\mathrm{irr}}\frac{M_s}{\zeta _S\zeta _R}\frac{\delta \mathrm{ln}R_s}{\tau }.$$
(8)
However, with ongoing irradiation not only does $`(\mathrm{ln}R_s/t)_{\mathrm{irr}}`$ decrease roughly exponentially on the time scale $`\tau `$, but other effects also tend to decrease $`\dot{M}_s`$. With the star’s transformation from a mainly convective to a fully radiative structure, $`\zeta _S`$ increases from $`1/3\zeta _S<0`$ to a large positive value. Thus $`\zeta _S\zeta _R`$ increases, with the result that $`\dot{M}_s`$ decreases. In addition, even the driving angular momentum loss is affected: for both braking mechanisms discussed in the literature, i.e. magnetic braking (e.g. Verbunt and Zwaan 1981; Mestel and Spruit 1987) and gravitational radiation (e.g. Kraft et al. 1962), $`(\mathrm{ln}|\dot{J}|/\mathrm{ln}R_s)<0`$. Thus $`|\dot{J}|`$ decreases if the secondary inflates. Worse, magnetic braking which is thought to be coupled to the presence of a convective envelope might cease altogether once the star has become fully radiative. In this case the system might no longer be able to sustain the mass transfer necessary to keep the secondary in its fully radiative state. As a consequence, it is therefore possible that such systems evolve through a limit cycle in which short phases (of duration $`\tau `$) with irradiation-enhanced mass transfer alternate with long detached phases during which the oversized secondary shrinks as it approaches thermal equilibrium, again on the thermal time scale, whereas the system contracts on the much longer time scale of angular momentum loss.
A number of theoretical studies along these lines, under the assumption of spherically symmetric irradiation, have been performed (e.g. by Podsiadlowski 1991; Harpaz and Rappaport 1991; Frank et al. 1992; D’Antona and Ergma 1993; Hameury et al. 1993; Vilhu et al. 1994), all more or less confirming the behaviour outlined above, including cyclic evolution (Hameury et al. 1993; Vilhu et al. 1994). It should be stressed once more that the validity of these studies, in which the effect of irradiation is maximized, rests on the validity of the assumption that spherically symmetric irradiation is an adequate approximation. In fact, Gontikakis and Hameury (1993) and Hameury et al. (1993) have examined whether the spherically symmetric approximation is adequate and find that it is not. Worse, when taking into account the anisotropy of the irradiation they find that the long-term evolution differs significantly from the spherically symmetric case. Now, anisotropic irradiation is a rather difficult 3-dimensional problem involving hydrostatic disequilibrium to some extent, and with it circulations which can transport heat from the hot to the cool side of the star. Because of this one might dismiss simple theoretical arguments such as those given by Gontikakis and Hameury (1993) and Hameury et al. (1993). There is, however, a much more compelling argument supporting the case of anisotropic irradiation. This derives from the observations of a number of close but detached binary systems in which a low-mass companion, exposed to an intense radiation field emerging from its hot (degenerate) companion, shows a bright illuminated and an essentially undisturbed cool hemisphere. A compilation of the systems in question is given in Table 1, where we list the object’s name, its association with a known planetary nebula, its orbital period (in days), the effective temperature of the irradiating white dwarf, the amplitude (in magnitudes) of the “reflection effect”, and relevant references. These systems all show a pronounced “reflection effect”in their light curves which, in turn, is explained by the anisotropic temperature distribution on the irradiated companion. Among them are 7 binary central stars of planetary nebulae. The systems listed in Table 1 demonstrate that a cool star, exposed to strong irradiation from a hot companion, can live without problems with a hot and a cool hemisphere. Moreover equal effective temperature over the whole surface is not established, at least not over atime scale $`10^4`$yr associated with the age of central stars of planetary nebulae, despite the fact that the irradiated star need not even rotate nearly synchronously. Obviously, heat transport from the hot side to the cool side by means of circulations is sufficiently ineffective that the large difference in effective temperatures can be maintained over long times. Thus the case of anisotropic irradiation has to be taken seriously and deserves a more detailed study. This is the objective of this paper.
## 3 Reaction of a low-mass star upon reducing its effective surface
We start our examination of anisotropic irradiation by recalling one of the main results obtained in the studies of spherically symmetric irradiation (e.g. Podsiadlowski 1991; D’Antona and Ergma 1993), namely that the main effect of irradiating a low-mass main-sequence star is that the star cannot lose energy as effectively (or at all) through the irradiated parts of its surface. If irradiation is spherically symmetric, the star has no choice but to store the blocked energy in gravitational and internal energy with the well-known result that it swells. In the case of anisotropic irradiation, such as one-sided irradiation from an accreting companion, the situation is qualitatively different: in addition to storing the blocked luminosity in internal and gravitational energy, the star can also divert its energy flow to and lose energy from the unirradiated parts of its surface. This is easily possible in the adiabatic convection zone. Because in this zone the energy flow is almost fully decoupled from the mechanical and thermal structure (the flow being proportional to $`(_a)^{3/2}`$ with $`_a1`$ rather than to $``$ (where $`=\mathrm{ln}T/\mathrm{ln}P`$ is the actual temperature gradient and $`_a=(\mathrm{ln}T/\mathrm{ln}P)_a`$ the adiabatic temperature gradient, and $`T`$ and $`P`$ are respectively the temperature and pressure) the mechanical and thermal structure of the star can still be considered to be spherically symmetric despite the fact that the energy flow might be highly anisotropic. The only thing which changes is the surface boundary condition which replaces the Stefan-Boltzmann law in the unirradiated case.
In the following we shall make a simple model for studying the situation described above. In this model we assume that over a fraction $`s_{\mathrm{eff}}`$ of the stellar surface the energy outflow is totally blocked (because of irradiation, or e.g. star spots, see Spruit and Ritter 1983) and that the remaining fraction of the surface $`(1s_{\mathrm{eff}})`$ radiates with an effective temperature $`T_{\mathrm{eff}}`$. The surface luminosity of the star $`L`$ can therefore be written as
$$L(s_{\mathrm{eff}})=4\pi R^2(1s_{\mathrm{eff}})\sigma T_{\mathrm{eff}}^4,0s_{\mathrm{eff}}<1,$$
(9)
where $`R`$ is the stellar radius. It is clear that in a more realistic model $`s_{\mathrm{eff}}`$ itself must depend on the irradiating flux $`F_{\mathrm{irr}}`$. In section 4 we shall discuss results of numerical calculations and a model with which we can evaluate $`s_{\mathrm{eff}}(F_{\mathrm{irr}})`$. For the moment we note only that $`s_{\mathrm{eff}}0`$ as $`F_{\mathrm{irr}}0`$ and that in the limit of high $`F_{\mathrm{irr}}`$, $`s_{\mathrm{eff}}`$ approaches the surface fraction of the star which sees the irradiation source.
The purpose of this simple modelling is, on the one hand, to provide estimates for the magnitude of and the time scale associated with the thermal relaxation process enforced by anisotropic irradiation, and, on the other hand, to show that the effects of anisotropic irradiation are not only quantitatively but also qualitatively different from those obtained in the spherically symmetric case.
The internal structure of a low-mass star with a deep outer convection zone can be described by the simple analytical model by Kippenhahn and Weigert (1994) for stars on or near the Hayashi line. This model is particularly applicable in our case, since we are only interested in the differential behaviour. Assuming a power law approximation for the frequency independent, i.e. grey photospheric opacity $`\kappa _{\mathrm{ph}}`$ on the unirradiated surface of the form
$$\kappa _{\mathrm{ph}}=\text{const.}P^aT^b,$$
(10)
the Eddington approximation yields the photospheric solution
$$\mathrm{log}T_{\mathrm{eff}}=\frac{a+1}{b}\mathrm{log}P_{\mathrm{ph}}+\frac{1}{b}\mathrm{log}M\frac{2}{b}\mathrm{log}R+\text{const.},$$
(11)
where $`P_{\mathrm{ph}}=P(\tau =2/3)`$ is the photospheric pressure at an optical depth $`\tau =2/3`$. The interior solution can be approximated by a polytrope with index $`n=3/2`$ and yields
$$\mathrm{log}T=\frac{2}{5}\mathrm{log}P+\frac{2}{5}\left[\frac{3}{2}\mathrm{log}R\frac{1}{2}\mathrm{log}M+\text{const.}\right].$$
(12)
Taking (12) at the photospheric point $`(P=P_{\mathrm{ph}},T=T_{\mathrm{eff}})`$, i.e. equating (11) and (12), together with (9) and (10) yields the luminosity $`L`$ lost by the star with radius $`R`$ over the surface area $`4\pi R^2(1s_{\mathrm{eff}})`$:
$$L(s_{\mathrm{eff}})=L_0(1s_{\mathrm{eff}})\left(\frac{R}{R_0}\right)^{(22a+4b+6)/(5a+2b+5)}.$$
(13)
Here $`L_0`$ and $`R_0`$ are respectively the luminosity and the radius of the unirradiated star in thermal equilibrium. In general, the irradiated star will not be in thermal equilibrium, i.e. $`L(R,s_{\mathrm{eff}})`$ does not equal the nuclear luminosity $`L_{\mathrm{nuc}}`$. For the disturbed star, the latter can be estimated by using homology relations. If we write the rate of nuclear energy generation $`\epsilon _{\mathrm{nuc}}`$ in the form appropriate for hydrogen burning via the pp-chain, i.e.
$$\epsilon _{\mathrm{nuc}}=\text{const.}\varrho T^\nu ,$$
(14)
where $`\varrho `$ is the density, we obtain (see e.g. Kippenhahn and Weigert 1994)
$$L_{\mathrm{nuc}}(R)=L_0\left(\frac{R}{R_0}\right)^{(\nu +3)}.$$
(15)
From (13) and (15) we obtain the gravitational luminosity
$$L_g(R,s_{\mathrm{eff}})=L(R,s_{\mathrm{eff}})L_{\mathrm{nuc}}(R).$$
(16)
$`L_g(R,s_{\mathrm{eff}})=0`$ defines the thermal equilibrium values of the irradiated star. These can be expressed in terms of the thermal equilibrium values of the unirradiated star as follows:
$`R_e(s_{\mathrm{eff}})=R_0(1s_{\mathrm{eff}})^r`$
with
$`r={\displaystyle \frac{5a+2b+5}{(\nu +3)(5a+2b+5)+22a+4b+6}},`$
$`L_e(s_{\mathrm{eff}})=L_0(1s_{\mathrm{eff}})^{\mathrm{}}`$
with
$`\mathrm{}=(\nu +3)r,`$
and
$`T_{\mathrm{eff}}(s_{\mathrm{eff}})=T_{\mathrm{eff},0}(1s_{\mathrm{eff}})^t`$
with
$`t={\displaystyle \frac{\mathrm{}2r1}{4}}.`$
Since the dominant opacity source in the photosphere of cool stars is $`H^{}`$ bound-free absorption, appropriate values for $`a`$ and $`b`$ in (10) are $`a0.5`$ and $`b45`$. Nuclear energy generation in low-mass stars occurs mainly via the pp-I chain and the appropriate value of $`\nu `$ in (14) is $`\nu 35`$. As a result, we find that
$$r0.1$$
$$\mathrm{}0.75,$$
and
$$t0.003.$$
This means that the effective temperature on the unirradiated part of the surface hardly changes, reflecting a well-known property of stars on or near the Hayashi line. Furthermore, since $`r0.1`$, the response of the stellar radius to anisotropic irradiation $`(s_{\mathrm{eff}}<1)`$ is much weaker than if isotropic irradiation is assumed. Specifically, if $`s_{\mathrm{eff}}0.5`$, as is the case for one-sided irradiation by an accreting companion, the equilibrium radius with irradiation is larger than the one without irradiation by at most $`7`$%. On the other hand, the total luminosity $`L_e`$ is significantly reduced: because of the rather large value of $`\nu +368`$ already a slight expansion of the star leads to a marked reduction of its nuclear luminosity.
Clearly, our model is not applicable if $`s_{\mathrm{eff}}1`$. This is because on the formal level $`L0`$ if $`s_{\mathrm{eff}}1`$ (see Eq. 9) and the values of $`R`$ and $`T_{\mathrm{eff}}`$ in thermal equilibrium with $`r<0`$ and $`t<0`$ (Eqs. 3 and 3) diverge. There is also a physical reason why this model does not apply in this case: $`s_{\mathrm{eff}}=1`$ corresponds to strong, spherically symmetric irradiation where the star in thermal equilibrium is fully radiative (e.g. Podsiadlowski 1991), whereas our model applies only to the extent that the star in question retains a deep outer convective envelope, even when irradiated.
We can now estimate the maximum contribution to mass transfer arising from thermal relaxation due to irradiation and the associated time scale. Thermal relaxation is maximal right at the onset of irradiation. At that time the gravitational luminosity of the star is $`L_g(R_0,s_{\mathrm{eff}})=s_{\mathrm{eff}}L_0`$. Using now the bipolytrope model in the formulation of KR92, we can write for the thermal relaxation term due to irradiation
$`\left({\displaystyle \frac{\mathrm{ln}R}{t}}\right)_{\mathrm{irr}}|_{R=R_0}`$ $`=`$ $`{\displaystyle \frac{(L_g)_{\mathrm{irr}}R_0}{GM^2}}(Q,n_1)`$ (20)
$`=`$ $`{\displaystyle \frac{s_{\mathrm{eff}}}{\tau _{_{\mathrm{KH}}}}}(Q,n_1)={\displaystyle \frac{s_{\mathrm{eff}}}{\tau _{\mathrm{ce}}}},`$
where $`\tau _{\mathrm{KH}}=GM^2/R_0L_0`$ is the Kelvin-Helmholtz time of the unirradiated star in thermal equilibrium, $`\tau _{\mathrm{ce}}=\tau _{\mathrm{KH}}/`$ the thermal time scale of the convective envelope, and the quantity $`(Q,n_1)`$ defines a dimensionless number which depends only on the relative size $`Q=R_{\mathrm{core}}/R_0`$ of and on the polytropic index $`n_1`$ in the radiative core. An explicit expression for $`(Q,n_1)`$ is given in KFKR96). In particular, for a single polytrope $`n=3/2`$, i.e. a fully convective star, where $`Q=0`$, one has $`=7/3`$. Furthermore, we show $`(Q,n_1)`$ as a function of mass for zero age main sequence stars as a full line in Fig. 1. We note that $``$ scales roughly as the inverse of the relative mass $`M_{\mathrm{ce}}/M`$ of the convective envelope: the dashed line in Fig. 1 shows that $`M_{\mathrm{ce}}/M\text{const.}`$ within better than a factor of two. Thus for the purpose of an estimate we can rewrite (20) as
$$\left(\frac{\mathrm{ln}R}{t}\right)_{\mathrm{irr}}|_{R=R_0}\frac{7}{3}\frac{s_{\mathrm{eff}}}{\tau _{\mathrm{KH}}}\left(\frac{M_{\mathrm{ce}}}{M}\right)^1,$$
(21)
Since the radius in thermal equilibrium with irradiation is larger by
$$\delta \mathrm{ln}R=\mathrm{ln}\frac{R_e(s_{\mathrm{eff}})}{R_0}=r\mathrm{ln}(1s_{\mathrm{eff}}),$$
(22)
the time scale for thermal relaxation becomes
$`\tau `$ $``$ $`\delta \mathrm{ln}R\left({\displaystyle \frac{\mathrm{ln}R}{t}}\right)_{\mathrm{irr}}^1=r{\displaystyle \frac{\mathrm{ln}(1s_{\mathrm{eff}})}{s_{\mathrm{eff}}}}\tau _{\mathrm{ce}}`$ (23)
$``$ $`r{\displaystyle \frac{\mathrm{ln}(1s_{\mathrm{eff}})}{s_{\mathrm{eff}}}}{\displaystyle \frac{7}{3}}\tau _{\mathrm{KH}}{\displaystyle \frac{M_{\mathrm{ce}}}{M}}.`$
As can be seen from (21) the maximal rate of expansion is proportional to $`s_{\mathrm{eff}}`$. Interestingly, the time over which the new thermal equilibrium is established is much shorter than if $`s_{\mathrm{eff}}=1`$, unless $`1s_{\mathrm{eff}}`$ is a small number and our model does not apply anyway.
Either (21) or (23) are to be inserted in (8) to get an estimate for the contribution of irradiation to mass transfer. Using (21) in (8) yields the peak contribution and (23) in (8) a time average.
In order to check the validity of our simple analytical model we have also performed numerical computations of full stellar models using the modified Stefan-Boltzmann law (9) as one of the outer boundary conditions and with $`s_{\mathrm{eff}}`$ as a free parameter. For our computations we have used a modified version of Mazzitelli’s (1989) stellar evolution code which is described in more detail in KR92, and have assumed a standard Pop. I chemical composition with (in the usual notation) $`X=0.70`$ and $`Z=0.02`$.
One of the basic predictions of our simple model is that (see Eq. 3) $`\mathrm{log}(R_e(s_{\mathrm{eff}})/R_0)`$ scales linearly with $`\mathrm{log}(1s_{\mathrm{eff}})`$ and that the slope $`|r|`$ is small. This is nicely confirmed by the behaviour of full stellar models shown in Fig. 2. As can be seen, the prediction is valid, at least qualitatively, over more than two orders of magnitude of $`(1s_{\mathrm{eff}})`$. Furthermore the slope in Fig. 2 is indeed small, confirming that $`|r|`$ is a small number. That the slope is different for stars of different mass is due to the fact that the effective values of the parameters $`a`$, $`b`$ and $`\nu `$ change with stellar mass.
In Fig. 3 we show in a mass radius diagram the thermal equilibrium radius $`R_e`$ as a function of mass for three different values of $`s_{\mathrm{eff}}`$, i.e. for the standard main sequence $`(s_{\mathrm{eff}}=0)`$, and for $`s_{\mathrm{eff}}=0.5`$ and $`s_{\mathrm{eff}}=0.9`$. As can be seen, the mass radius relations for $`s_{\mathrm{eff}}=0.5`$ and $`s_{\mathrm{eff}}=0.9`$ are shifted by a small amount to larger radii and run roughly “parallel”to the standard main sequence $`(s_{\mathrm{eff}}=0)`$, as is predicted by our simple model.
Finally, in Fig. 4 we show as an example the thermal relaxation with time of an $`0.4M_{}`$ star with $`s_{\mathrm{eff}}=0.5`$. If time is measured in units of $`(t/\mathrm{ln}R)|_{R=R_0}`$, as is done in Fig. 4, it is seen that the relaxation process is characterized by this time scale and that new thermal equilibrium is reached after only 0.2–0.3 of these time units. Again this confirms our analytical result (cf. Eq. 23), according to which the relaxation process lasts a few $`r\mathrm{ln}(1s_{\mathrm{eff}})`$ in these units. It is also seen that the effective temperature on the unirradiated part of the star rises only very little, as predicted, but that the relative mass of the convective envelope is reduced significantly from $`0.6`$ at the beginning to $`0.33`$ in the new thermal equilibrium.
## 4 Stability against irradiation-induced mass transfer
Let us now examine the situation in which a low-mass star transfers mass to a compact companion (of mass $`M_c`$ and radius $`R_c`$) and, in turn, is irradiated (directly or indirectly) by the accretion light source. Because irradiation can enhance mass transfer and more irradiation can give rise to even higher mass transfer, we must examine under which conditions such a situation is stable against irradiation-induced runaway mass transfer.
### 4.1 Arbitrary irradiation geometry
In this subsection we wish to keep the discussion as general as possible. Therefore, we do not specify the irradiation geometry. Specific models which do just that will be presented in the next subsection (4.2).
The component normal to the stellar surface of the irradiating flux generated by accretion can be written as
$$F_{\mathrm{irr}}(\vartheta ,\phi )=\frac{\eta }{4\pi }\frac{GM_c(\dot{M}_s)}{R_ca^2}h(\vartheta ,\phi ).$$
(24)
Here $`\dot{M}_s`$ is the mass transfer rate, $`a`$ the orbital separation, $`h(\vartheta ,\phi )`$ a dimensionless function of the position on the secondary’s surface (characterized by polar coordinates $`(\vartheta ,\phi )`$) which describes the irradiation geometry, and $`\eta <1`$ a dimensionless efficiency factor. $`\eta =1`$ only if the accretion luminosity is radiated isotropically from the compact star and if all the energy which is radiated into the solid angle subtended by the donor as seen from the accretor is absorbed below the photosphere. In a real situation $`\eta `$ will be considerably less than unity for several reasons. The most important of these are: 1) the accretion luminosity will, in general, not be emitted isotropically, e.g. if accretion occurs via a disk which radiates predominantly perpendicular to the orbital plane and casts a shadow onto the donor, or if the accretor is strongly magnetized and accretes mainly near the magnetic poles, as e.g. in AM Her systems. 2) energy emitted in certain spectral ranges, as e.g. EUV radiation and soft X-rays, is unlikely to reach the photosphere of the donor (in the case of EUV and soft X-ray radiation because of the high column density of neutral hydrogen). 3) part of the incident flux will be scattered away before penetrating into the photosphere. 4) not all of the transferred mass needs to be accreted by the compact star. Part of it may leave the system before releasing much potential energy, e.g. via a wind from the outer regions of the accretion disk.
From this it is clear that computing a reliable value for $`\eta `$ is a formidable if not nearly impossible task. Therefore, we treat $`\eta `$ as a free parameter and our goal must be to arrive at conclusions which are as independent of $`\eta `$ as possible.
For deriving the criterion for adiabatic stability against irradiation-induced mass transfer we follow Ritter (1988) (see also Ritter 1996). The only difference is that we use now Eq. (6) instead of (1) (the latter corresponding to Eq. (12) in Ritter’s (1988) paper). Accordingly the criterion for adiabatic stability becomes
$`\zeta _S\zeta _R>\zeta _{\mathrm{irr}}`$ $`=`$ $`M_s{\displaystyle \frac{}{\dot{M}_s}}\left({\displaystyle \frac{\mathrm{ln}R_s}{t}}\right)`$ (25)
$`=`$ $`M_s{\displaystyle \frac{}{L}}\left({\displaystyle \frac{\mathrm{ln}R_s}{t}}\right){\displaystyle \frac{dL}{d\dot{M}_s}}.`$
The dimensionless number $`\zeta _{\mathrm{irr}}`$ which is defined by (25) measures how sensitively the stellar radius changes in response to irradiation produced by mass transfer. Ignoring the influence of irradiation, i.e. setting $`dL/d\dot{M}_s=0`$, gives $`\zeta _{\mathrm{irr}}=0`$ and (25) reduces to the usual criterion for adiabatic stability.
We compute the derivative $`(^2\mathrm{ln}R_s/tL)`$ in (25) in the framework of the bipolytrope model (e.g. KR92) from which we obtain (cf Eq. 20)
$$\frac{}{L}\left(\frac{\mathrm{ln}R_s}{t}\right)=\frac{R_s}{GM_s^2}(Q,n_1).$$
(26)
For calculating $`dL/d\dot{M}_s`$ we shall make use of what we shall refer to as the weak irradiation assumption. Making this assumption is tantamount to assuming that at any point $`(\vartheta ,\phi )`$ lateral heat transport is negligible compared to radial transport. Lateral heat transport occurs in the form of radiative diffusion and advection because of non-vanishing lateral temperature gradients $`T/\vartheta `$ and $`T/\phi `$, and departures from strict hydrostatic equilibrium. We shall show in the Appendix that the weak irradiation assumption can be justified in those cases we are interested in and that we may safely neglect lateral heat transport. Accordingly, energy conservation requires that at any point $`(\vartheta ,\phi )`$ the stellar flux, i.e. the energy lost by the star from its interior per unit time and unit surface area is
$$F(\vartheta ,\phi )=\sigma T_{\mathrm{irr}}^4(\vartheta ,\phi )F_{\mathrm{irr}}(\vartheta ,\phi ),$$
(27)
where $`T_{\mathrm{irr}}`$ is the effective temperature of the surface element in question. With (27) the stellar luminosity, i.e. the energy loss per unit time from the interior becomes
$$L=R_s^2_0^{2\pi }_0^\pi F(\vartheta ,\phi )\mathrm{sin}\vartheta d\vartheta d\phi .$$
(28)
Before working out $`dL/d\dot{M}_s`$ in (25), we shall first examine the reaction of the stellar surface to irradiation in rather general terms. For that it is convenient to introduce the dimensionless irradiating flux
$$x=\frac{F_{\mathrm{irr}}}{\sigma T_0^4}=\frac{F_{\mathrm{irr}}}{F_0}$$
(29)
and the dimensionless stellar flux
$$G=\frac{F}{F_0}=\left(\frac{T_{\mathrm{irr}}(x)}{T_0}\right)^4x=G(x),$$
(30)
where $`T_0=T_{\mathrm{eff}}(F_{\mathrm{irr}}=0)`$ is the effective temperature and $`F_0=F(F_{\mathrm{irr}}=0)=\sigma T_0^4`$ the stellar flux in the absence of irradiation. We note that $`G(0)=1`$ and that we expect $`G(\mathrm{})=0`$, i.e. that for very high irradiating fluxes energy outflow from the stellar interior is totally blocked. Next we introduce the function
$$g(x)=\frac{dG}{dx}=\frac{dF}{dF_{\mathrm{irr}}},$$
(31)
which has the following notable properties: First
$$_0^{\mathrm{}}g(x)𝑑x=G(0)G(\mathrm{})=1$$
(32)
Second, for positive albedos
$$0<g(x)<1x0.$$
(33)
Third, if $`F`$ is a monotonically decreasing function of $`x`$ (and there is no physical reason why this should not be so), then
$$g^{}(x)<0x0.$$
(34)
From (32) and (34) we can immediately prove through integration of $`g(x)`$ by parts that
$$\text{Max}\left[xg(x)\right]<1.$$
(35)
The reason why $`xg(x)`$ attains a maximum can be understood as follows: Rewriting $`xg(x)`$ in dimensional form (using (29) and (31)) we see that
$$xg(x)=\frac{F_{\mathrm{irr}}}{F_0}\frac{dF}{dF_{\mathrm{irr}}}.$$
(36)
The second factor in (36) describes the incremental blocking of the energy loss from the interior with changing irradiating flux $`F_{\mathrm{irr}}`$. The maximum of $`xg(x)`$ arises because $`dF/dF_{\mathrm{irr}}`$ vanishes for large $`F_{\mathrm{irr}}`$. This, in turn, is a consequence of the fact that as long as the star keeps a negative temperature gradient $`dT/dr`$ in its subphotospheric layer, i.e. the superadiabatic convection zone, irradiation can not block more than the total flux $`F_0`$.
As we shall see below, the fact that $`xg(x)`$ has a maximum is very important for the stability discussion. In fact, we shall see in the next section 5 that for realistic situations the maximum of $`xg(x)`$ is smaller by about a factor of two than the strict upper limit given by (35).
We can now return to Eq. (28) and compute $`dL/d\dot{M}_s`$. This can be written as
$$\frac{dL}{d\dot{M}_s}=R_s^2_0^{2\pi }_0^\pi \frac{dF}{dF_{\mathrm{irr}}}\frac{dF_{\mathrm{irr}}}{d(\dot{M}_s)}\mathrm{sin}\vartheta d\vartheta d\phi ,$$
(37)
where we note that the first factor in the integral is equal to $`g`$. With
$$\frac{dF_{\mathrm{irr}}}{d(\dot{M}_s)}=\frac{F_{\mathrm{irr}}(\vartheta ,\phi )}{(\dot{M}_s)}$$
$`(38a)`$
$$=\frac{\eta }{4\pi }\frac{GM_c}{R_ca^2}h(\vartheta ,\phi )$$
$`(38b)`$
from (24) we have (using (38a))
$`{\displaystyle \frac{dL}{d\dot{M}_s}}={\displaystyle \frac{1}{4\pi }}{\displaystyle \frac{L_0}{(\dot{M}_s)}}\left({\displaystyle \frac{R_s}{R_0}}\right)^2\times `$
$`{\displaystyle _0^{2\pi }}{\displaystyle _0^\pi }x(\vartheta ,\phi )g(x(\vartheta ,\phi ))\mathrm{sin}\vartheta d\vartheta d\phi ,`$ (39)
where $`L_0=4\pi R_0^2\sigma T_0^4`$ is the luminosity of the star in thermal equilibrium without irradiation. Combining now Eqs. (25), (26) and (4.1) we can rewrite the stability criterion as
$`\zeta _S\zeta _R>\zeta _{\mathrm{irr}}={\displaystyle \frac{1}{4\pi }}{\displaystyle \frac{\tau _{M_s}}{\tau _{\mathrm{KH}}}}(Q,n_1)\left({\displaystyle \frac{R_s}{R_0}}\right)^3\times `$
$`{\displaystyle _0^{2\pi }}{\displaystyle _0^\pi }x(\vartheta ,\phi )g(x(\vartheta ,\phi ))\mathrm{sin}\vartheta d\vartheta d\phi `$ (40)
or
$`\mathrm{\Lambda }2(\zeta _S\zeta _R){\displaystyle \frac{\tau _{\mathrm{KH}}}{\tau _{M_s}}}^1(Q,n_1)\left({\displaystyle \frac{R_s}{R_0}}\right)^3`$
$`>{\displaystyle \frac{1}{2\pi }}{\displaystyle _0^{2\pi }}{\displaystyle _0^\pi }x(\vartheta ,\phi )g(x(\vartheta ,\phi ))\mathrm{sin}\vartheta d\vartheta d\phi I,`$ (41)
where
$$\tau _{M_s}=\frac{M_s}{\dot{M}_s}$$
(42)
is the mass loss time scale. Although the relations (4.1) and (4.1) do not show an explicit dependence on $`\eta `$ they nevertheless depend on it via $`x`$. However, because of the fact that $`xg(x)`$ has a maximum, the integral on the right-hand side of (4.1) and (4.1) must have a maximum that is smaller than Max$`(xg(x))`$. Hence we can state that systems which fulfill the condition
$$\mathrm{\Lambda }>\frac{1}{2\pi }\text{Max}\left[_0^{2\pi }_0^\pi x(\vartheta ,\phi )g(x(\vartheta ,\phi ))\mathrm{sin}\vartheta d\vartheta d\phi \right],$$
(43)
which is independent of $`\eta `$, are definitely stable against irradiation-induced runaway mass transfer.
The reason for rewriting the stability criterion (4.1) in the form of (4.1) or (43) is that in the latter conditions $`\mathrm{\Lambda }`$ does not depend on irradiation but only on the internal structure of the donor star (via $`\zeta _S`$, $`\tau _{\mathrm{KH}}`$, $`(Q,n_1)`$, $`R_0`$) and on the secular evolution model (via $`\tau _{M_s}`$, $`\zeta _R`$ and $`R_s`$). On the other hand, all the information about the irradiation model is contained in the expression on the right-hand side.
For discussing the stability of mass transfer in the limit of very small irradiating fluxes, i.e. $`x0`$, we must use (38b) instead of (38a) in (4.1). This results in the following stability criterion:
$`\zeta _S\zeta _R>\zeta _{\mathrm{irr}}={\displaystyle \frac{\eta }{4\pi }}{\displaystyle \frac{M_c}{M_s}}{\displaystyle \frac{R_s}{R_c}}\left({\displaystyle \frac{R_s}{a}}\right)^2(Q,n_1)\times `$
$`{\displaystyle _0^{2\pi }}{\displaystyle _0^\pi }g(x(\vartheta ,\phi ))h(\vartheta ,\phi )\mathrm{sin}\vartheta d\vartheta d\phi .`$ (44)
or
$`\mathrm{\Gamma }`$ $``$ $`2(\zeta _S\zeta _R){\displaystyle \frac{M_s}{M_c}}{\displaystyle \frac{R_c}{R_s}}\left({\displaystyle \frac{a}{R_s}}\right)^2^1(Q,n_1)`$ (45)
$`>`$ $`{\displaystyle \frac{\eta }{2\pi }}{\displaystyle _0^{2\pi }}{\displaystyle _0^\pi }g(x(\vartheta ,\phi ))h(\vartheta ,\phi )\mathrm{sin}\vartheta d\vartheta d\phi .`$
Because of (34) we arrive at a necessary and sufficient condition for stability
$$\mathrm{\Gamma }>\frac{\eta }{2\pi }g(0)_0^{2\pi }_0^\pi h(\vartheta ,\phi )\mathrm{sin}\vartheta d\vartheta d\phi .$$
(46)
As above, we have separated in the conditions (45) and (46) factors which do not depend on irradiation (collected in $`\mathrm{\Gamma }`$) from those which do (on the right-hand side). Like $`\mathrm{\Lambda }`$, $`\mathrm{\Gamma }`$ depends only on the internal structure of the donor star and the secular evolution model. In essence, Eqs. (45) and (46) are conditions on the value of $`\eta `$ in the sense that for a given irradiation model, i.e. given $`h(\vartheta ,\phi )`$ and $`g(x)`$, $`\eta `$ must not exceed a certain value if mass transfer is to be stable.
Furthermore, Eqs. (45) and (46) show in particular that because $`g(x)`$ is maximal for $`x=0`$ (see Eq. 34), i.e. for $`F_{\mathrm{irr}}=0`$, systems with an unirradiated donor star are the most susceptible to irradiation. In other words, if a system is stable at the turn-on of mass transfer it will remain so later, unless secular effects diminish the value of $`\mathrm{\Gamma }`$.
### 4.2 Specific irradiation models
In the following we describe two specific irradiation models, a rather simple one, hereafter referred to as the constant flux or constant temperature model, and a more realistic one, hereafter referred to as the point source model.
#### 4.2.1 The constant flux model
In this model we assume a fraction $`s(0.5)`$ of the stellar surface to be irradiated by a constant average normal flux
$$F_{\mathrm{irr}}=\frac{\eta }{8\pi }\frac{GM(\dot{M}_s)}{R_ca^2}.$$
(47)
Note that the parameter $`s`$, introduced above, and $`s_{\mathrm{eff}}`$ which we have introduced in Sect. 3 are not the same quantity. However, $`s`$ and $`s_{\mathrm{eff}}`$ are related and the corresponding relation will be given below.
Comparison of (47) with (24) shows that we may write $`h(\vartheta ,\phi )`$ as follows:
$$h(\vartheta ,\phi )=\{\begin{array}{ccc}\frac{1}{2},0\vartheta <\vartheta _{\mathrm{max}},0\phi 2\pi \hfill & & \\ 0,\vartheta _{\mathrm{max}}\vartheta \pi ,0\phi 2\pi \hfill & & \end{array},$$
(48)
where now $`\vartheta `$ is the colatitude with respect to the substellar point and $`\phi `$ the azimuth around the axis joining the two stars. If the star is assumed to be spherical, the colatitude $`\vartheta _{\mathrm{max}}`$ of the “terminator”of the irradiated part of the surface and $`s`$ are related via
$$s=\frac{1}{2}(1\mathrm{cos}\vartheta _{\mathrm{max}}).$$
(49)
With (48) and (49), and $`x=F_{\mathrm{irr}}/F_0`$, the stability conditions (4.1) and (45) become respectively
$$\mathrm{\Lambda }>2sxg(x)$$
(50)
and
$$\mathrm{\Gamma }>\eta sg(x).$$
(51)
These are the results presented earlier in Ritter, Zhang and Kolb (1995, 1996).
Because of (48) not only is the irradiating normal flux constant but also the effective temperature on the irradiated part (hence the name constant temperature model). (48) inserted in (28) yields the luminosity of the star
$$L=4\pi R_s^2\left[s(\sigma T_{\mathrm{irr}}^4F_{\mathrm{irr}})+(1s)\sigma T_0^4\right].$$
(52)
Comparing this modified Stefan-Boltzmann law with (9) yields the relation between $`s`$ and $`s_{\mathrm{eff}}`$:
$$s_{\mathrm{eff}}=s\left[1\frac{\sigma T_{\mathrm{irr}}^4F_{\mathrm{irr}}}{\sigma T_0^4}\right]=s\left[1G(x)\right].$$
(53)
#### 4.2.2 The point source model
In this model, which has already been discussed in some detail by KFKR96, we assume the donor star to be irradiated by a point source at the location of the compact star. For simplicity we assume the secondary to be spherical. The chosen geometry is axisymmetric with respect to the axis joining the two stars. Denoting again by $`\vartheta `$ the colatitude of a point on the surface of the irradiated star with respect to the substellar point (for a sketch of the geometry see Fig. 5), $`h(\vartheta ,\phi )`$ in (24) becomes
$$h(\vartheta )=\frac{\mathrm{cos}\vartheta f_s}{(12f_s\mathrm{cos}\vartheta +f_s^2)^{3/2}},$$
(54)
where
$$f_s=\frac{R_s}{a}=f_s\left(\frac{M_c}{M_s}\right)$$
(55)
is the secondary’s radius in units of the orbital separation $`a`$. Because the secondary fills its critical Roche volume, $`f_s`$ is a function only of the mass ratio $`M_c/M_s`$. The terminator of the irradiated part of the star is at the colatitude
$$\vartheta _{\mathrm{max}}=\text{arc}\mathrm{cos}(f_s).$$
(56)
Inserting (54) in (4.1) or (45), the stability criteria become respectively
$$\mathrm{\Lambda }>_0^{\vartheta _{\mathrm{max}}}x(\vartheta )g(x(\vartheta ))\mathrm{sin}\vartheta d\vartheta =I_{\mathrm{PS}}$$
(57)
and
$$\mathrm{\Gamma }>\eta _0^{\vartheta _{\mathrm{max}}}g(x(\vartheta ))h(\vartheta )\mathrm{sin}\vartheta d\vartheta .$$
(58)
With (54) inserted in (28), the luminosity of the star is
$`L`$ $`=`$ $`4\pi R_s^2\sigma T_0^4\times `$ (59)
$`\left\{{\displaystyle \frac{1}{2}}(1+f_s)+{\displaystyle \frac{1}{2}}{\displaystyle _0^{\vartheta _{\mathrm{max}}}}G(x(\vartheta ))\mathrm{sin}\vartheta d\vartheta \right\}.`$
Comparing (59) with (9) we find that the effective fraction of the stellar surface over which the energy outflow is blocked is
$$s_{\mathrm{eff}}=\frac{1}{2}\left\{1f_s_0^{\vartheta _{\mathrm{max}}}G(x(\vartheta ))\mathrm{sin}\vartheta d\vartheta \right\}.$$
(60)
## 5 The reaction of the subphotospheric layers to irradiation
From the stability analysis we have carried out in the previous section it is clear that we need to know more about the functions $`G(x)`$ or $`g(x)`$ (cf. Eqs. 30 and 31) if we wish to use the stability criteria in a quantitative way. So far we know only the properties detailed in Eqs. (32)–(35). These, however, are insufficient for our purposes. Therefore, in this section we shall first use a simple model to derive $`g(x)`$ explicitly and thereafter discuss results of numerical calculations.As in section 4 we shall adopt the weak irradiation assumption. In this approximation the relation between the stellar flux $`F`$ and the irradiating flux $`F_{\mathrm{irr}}`$ is a purely local property.
### 5.1 A one-zone model for the superadiabatic layer
For determining $`F(F_{\mathrm{irr}})`$ we need now a more detailed model of the stellar structure than the one we have assumed in Sect. 3. Whereas in Sect. 3 we have assumed with Kippenhahn and Weigert (1994) that the convective envelope remains adiabatic up to the photosphere, we shall now relax this assumption and take into account the existence of a thin superadiabatic convection zone below the photosphere where convection itself is ineffective as a means of energy transport and energy flows mainly via radiative diffusion. Sufficiently deep in the star, where convection is effective, i.e. adiabatic, the thermal and mechanical structure of the envelope remain spherically symmetric to a very good approximation even in the presence of anisotropic irradiation at the surface (cf. our discussion in Sect. 2). It is only the very thin superadiabatic layer (with a mass of typically $`10^{10}\text{M}_{}`$), where energy is mainly transported via radiation, which is strongly affected by irradiation. It is this property which allows us to make the weak irradiation assumption, i.e. to treat the effects of irradiation in a local approximation and with the following simple model.
For this we adopt for the moment the temperature-pressure stratification shown in Fig. 6. This means that we replace the “true” structure by one in which we assume convection to be adiabatic out to a point where the pressure is $`P=P_B`$ and the temperature $`T=T_B`$ (where the subscript $`B`$ stands for the base of the superadiabatic zone). The superadiabatic convection zone extends from the point $`(P_B,T_B)`$ up to the photosphere where $`P=P_{\mathrm{ph}}`$ and $`T=T_{\mathrm{eff}}`$. Because in this zone convection is ineffective, we make the simplifying assumption that energy transport is via radiation only, i.e. that $`=_{\mathrm{rad}}`$, where $`_{\mathrm{rad}}`$ is the radiative temperature gradient. Because on an irradiated part $`T_{\mathrm{eff}}=T_{\mathrm{irr}}>T_0`$, where $`T_0`$ is the effective temperature in the absence of irradiation, but $`T_B`$ is assumed to be the same irrespective of irradiation, we see that irradiation, by raising the effective temperature, reduces the temperature gradient $``$ and thus the radiative energy loss through these layers. The superadiabatic layer works like a valve which is open if there is no external irradiation and which closes in progression with the irradiating flux. As sketched in Fig. 6 we assume for simplicity that $`P_{\mathrm{ph}}`$ does not depend on $`F_{\mathrm{irr}}`$, i.e. on $`T_{\mathrm{eff}}`$.
We can now to derive $`T_{\mathrm{irr}}(F_{\mathrm{irr}})`$ by making a simple one-zone model for the superadiabatic layer. For this we shall furthermore assume that the optical depth through this layer is large enough for the diffusion approximation to hold. Then the radiative flux is given by
$$F_{\mathrm{rad}}=F=\frac{ac}{3\kappa \varrho }\frac{dT^4}{dr},$$
(61)
where $`\varrho `$ is the density, $`\kappa `$ the opacity and the other symbols have their usual meaning.
Now we consider the transported flux in two different superadiabatic layers (using subscripts 1 and 2) on an arbitrary isobar at some level $`P_{\mathrm{ph}}<P_1=P_2<P_B`$. The radiative fluxes are then
$$F_i=\frac{ac}{3\kappa _i\varrho _i}\left(\frac{dT^4}{dr}\right)_i,i=1,2.$$
(62)
Therefore
$$\frac{\kappa _1\varrho _1}{\kappa _2\varrho _2}=\frac{F_2}{F_1}\frac{(dT^4/dr)_1}{(dT^4/dr)_2}.$$
(63)
Now we use for the opacity a power-law approximation of the form (10) and the ideal gas equation
$$P=\frac{}{\mu }\varrho T,$$
(64)
where $``$ is the gas constant and $`\mu `$ the mean molecular weight. Since in the regions of interest the dominant species (H and He) are neutral, we may assume $`\mu _1=\mu _2`$. Thus with $`P_1=P_2`$ we have
$$\varrho _1T_1=\varrho _2T_2.$$
(65)
Inserting now (10) and (65) into (63) we obtain
$$\frac{F_1}{F_2}=\frac{T_1^{1b}(dT^4/dr)_1}{T_2^{1b}(dT^4/dr)_2}=\{\begin{array}{cc}\frac{(dT^{5b}/dr)_1}{(dT^{5b}/dr)_2},& 5b0\\ & \\ \frac{(d\mathrm{ln}T/dr)_1}{(d\mathrm{ln}T/dr)_2},& 5b=0\end{array}.$$
(66)
Now we make the one-zone approximation by writing
$`{\displaystyle \frac{dT^n}{dr}}={\displaystyle \frac{T^n(P=P_B)T^n(P=P_{\mathrm{ph}})}{\mathrm{\Delta }r}}={\displaystyle \frac{T_B^nT_{\mathrm{eff}}^n}{\mathrm{\Delta }r}}`$
(67)
$`{\displaystyle \frac{d\mathrm{ln}T}{dr}}={\displaystyle \frac{\mathrm{ln}T(P=P_B)\mathrm{ln}T(P=P_{\mathrm{ph}})}{\mathrm{\Delta }r}}={\displaystyle \frac{\mathrm{ln}T_B\mathrm{ln}T_{\mathrm{eff}}}{\mathrm{\Delta }r}}.`$
If we now identify layer 1 with the unirradiated one, i.e. set $`F_1=F_0`$ and $`T_{\mathrm{eff},1}=T_0`$, and layer 2 with an irradiated one, i.e. set $`F_2=F`$ and $`T_{\mathrm{eff},2}=T_{\mathrm{irr}}`$ we obtain by inserting (5.1) into (66)
$$\frac{F_0}{F}=\frac{1}{G}=\frac{\sigma T_0^4}{\sigma T_{\mathrm{irr}}^4F_{\mathrm{irr}}}=\{\begin{array}{cc}\frac{T_B^{5b}T_0^{5b}}{T_B^{5b}T_{\mathrm{irr}}^{5b}},\hfill & b50\hfill \\ & \\ \frac{\mathrm{ln}T_B\mathrm{ln}T_0}{\mathrm{ln}T_B\mathrm{ln}T_{\mathrm{irr}}},\hfill & b5=0\hfill \end{array},$$
(68)
where we have assumed for simplicity $`\mathrm{\Delta }r_1=\mathrm{\Delta }r_2`$. Equation (68) can be solved for $`T_{\mathrm{irr}}=T_{\mathrm{irr}}(T_0,T_B,F_{\mathrm{irr}})`$, thus providing $`G(x,T_B)`$. Thus (68) together with (28) or special cases thereof (Eqs. 52 or 59) can be used as an outer boundary condition for numerical computations.
Let us now briefly consider the consequences of our above assumptions that $`P_{\mathrm{ph},1}=P_{\mathrm{ph},2}`$ and $`\mathrm{\Delta }r_1=\mathrm{\Delta }r_2`$. Because the dominant opacity source is H<sup>-</sup> and therefore the opacity increases steeply with temperature, the photospheric pressure on an irradiated, hotter part of the star will be lower than on an unirradiated part (because $`P_{\mathrm{ph}}g/\kappa `$, where $`g`$ is the surface gravity). Furthermore, the hotter the surface, the further out the photospheric point will be, i.e. $`\mathrm{\Delta }r_2>\mathrm{\Delta }r_1`$ in the above calculation. Because $`P_{\mathrm{ph},2}<P_{\mathrm{ph},1}`$, the temperature gradient $`_2`$ below an irradiated part will be higher when setting $`P_{\mathrm{ph},2}=P_{\mathrm{ph},1}`$ rather than using the proper value of $`P_{\mathrm{ph},2}`$. Therefore by assuming $`P_{\mathrm{ph},2}=P_{\mathrm{ph},1}`$ we overestimate the radiative flux on the irradiated part. We obtain the same result from using $`\mathrm{\Delta }r_2=\mathrm{\Delta }r_1`$: because in reality $`\mathrm{\Delta }r_2>\mathrm{\Delta }r_1`$, the temperature gradient (5.1) on an irradiated part is lower than what our estimate with $`\mathrm{\Delta }r_2=\mathrm{\Delta }r_1`$ yields. Therefore, our very simple one-zone model, i.e. Eq. (68) underestimates the blocking effect somewhat. Since, on the one hand, this deficit can easily be compensated for by slightly increasing the value of $`b`$, and since, on the other hand, the precise value of $`b`$ which is appropriate is not exactly determined within our model (we shall later take an average value determined from published opacity tables), we consider (68) a fair approximation of the physical situation described. The really important aspect of our model is, however, that qualitatively it yields the correct behaviour of a stellar surface exposed to external irradiation, and that it is still simple enough to allow insight in the situation described.
We can also compute $`g=dG/dx`$. Differentiation of (68) yields
$`g`$ $`=\{\begin{array}{ccc}\frac{nT_0^4T_{\mathrm{irr}}^{n1}}{nT_0^4T_{\mathrm{irr}}^{n1}+4T_{\mathrm{irr}}^3\left(T_B^nT_0^n\right)},& n=5b0& \\ & & \\ \frac{T_0^4}{T_0^4+4T_{\mathrm{irr}}^4\mathrm{ln}\left(T_B/T_0\right)},& n=5b=0& \end{array},`$ (72)
if $`T_{\mathrm{irr}}<T_B`$. For consistency with Eq. (32) we require $`g=0`$ if $`T_{\mathrm{irr}}>T_B`$.
As can be seen from Eqs. (68) and (72) the functions $`G`$ and $`g`$ depend only on two parameters characterizing the unirradiated star, namely on $`T_0`$ and $`T_B`$, and on the opacity law via $`b`$. While $`T_0`$ is a well-defined quantity, $`T_B`$ is not because in real stars the run of temperature $`T`$ with pressure $`P`$ is not as simple as the one assumed in our simple model (and sketched in Fig. 6). In particular, the transition from convective to radiative energy transport is smooth and does not occur at one particular point as we have assumed in our model. Since $`T_B`$ stands for the temperature at which this transition occurs, we determine $`T_B`$ by requiring that in a full stellar model the ratio $`F_{\mathrm{conv}}/F_{\mathrm{rad}}`$ of convective flux $`F_{\mathrm{conv}}`$ to radiative flux $`F_{\mathrm{rad}}`$ reaches a prescribed value, say $`F_{\mathrm{conv}}/F_{\mathrm{rad}}=k`$. This is equivalent to the condition $`_{\mathrm{rad}}=(k+1)`$. Of course the choice of $`k`$ is somewhat arbitrary but a value $`k1`$ seems a natural choice. Our simple model is an acceptable description of the real situation only if for a given model $`T_B(k)`$ is sufficiently insensitive to $`k`$. In order to demonstrate that this is indeed the case we plot in Fig. 7 the run of $`T_B`$ as a function of the stellar mass of zero-age main-sequence stars with Pop. I chemical composition (X = 0.70, Z = 0.02) for three different values of $`k`$, i.e. $`k=1`$ (full line), $`k=0.5`$ (dashed line) and $`k=2`$ (dotted line). Fig. 7 shows two important properties of low-mass stars: The first one is that a significant superadiabatic convection zone exists only in stars with a mass $`M0.65\text{M}_{}`$. Below $`M0.6\text{M}_{}`$ the stratification is essentially adiabatic up to the photosphere. This means that application of our simple one-zone model is restricted to stars in the mass range $`0.65\text{M}_{}M1\text{M}_{}`$. The second property shown in Fig. 7 is that the run of $`T_B(M)`$ for different $`k`$ is qualitatively the same for all three values of $`k`$. This means that as long as a star has a significant superadiabatic convection zone the transition from convective to radiative energy transport occurs in a rather narrow temperature interval, thus justifying our simple approach.
5.2 Stars with a mass $`M0.6\text{M}_{}`$
As Fig. 7 shows, $`T_BT_0`$ in stars with a mass $`M0.6\text{M}_{}`$. This means on the one hand that the total optical depth between the photosphere (at $`T=T_0`$) and the point where $`T=T_B`$ is small, in fact too small for our simple one-zone model, which assumes the diffusion approximation (Eq. 61), to be applicable. On the other hand, this means also that in these stars even in the photosphere a non-negligible fraction of the flux is transported by convection. Therefore, we must ask how irradiation changes the transported flux if the top of the (adiabatic) convection zone is at low optical depth. Because the convective flux is $`F_{\mathrm{conv}}(_{ad})^{3/2}`$ and the value of $``$ is directly influenced by irradiation, this situation cannot be described by a simple model. Rather one ought to determine $``$ by solving the full set of equations describing convective energy transport, i.e. in the simplest case the equations of mixing length theory. Considering the uncertainties inherent in current convection theories it is not obvious whether it is possible to make general statements about the functions $`g`$ or $`G`$. After all it is at least conceivable that already a small irradiating flux could result in a significant reduction of $`(_{ad})`$ (which itself is a rather small number because convection is not far from adiabatic) and that therefore $`F/F_{\mathrm{irr}}`$ could attain a large negative value. However, as the following argument shows, there is a limit to how fast $`F`$ can drop in response to increasing $`F_{\mathrm{irr}}`$. If $`F`$ drops too strongly this results in an effective temperature $`T_{\mathrm{irr}}<T_0`$. This in turn means that the temperature gradient must be steeper than in the unirradiated star and, therefore, that $`F>F_0`$, in contradiction to the starting assumption $`F/F_{\mathrm{irr}}<0`$. In order to avoid this contradiction $`T_{\mathrm{irr}}`$ must not decrease with increasing $`F_{\mathrm{irr}}`$, i.e. $`T_{\mathrm{irr}}/F_{\mathrm{irr}}0`$, from which we immediately recover (33), i.e. $`g1`$. Because $`g^{}<0`$ (cf. 34), $`g`$ is maximal in the limit $`F_{\mathrm{irr}}0`$. Therefore we need to determine $`g(0)`$ for the stars in question. For this we return to the one zone model (Sect. 5.1). From Eq. (72) we find that $`g(0)`$ increases as $`T_B/T_0`$ decreases. In fact, in the limit $`T_B=T_0`$, this model yields $`g(0)=1`$. Thus the closer the adiabatic convection zone reaches to the surface the more sensitive the star is to irradiation, i.e. the larger $`g`$. On the other hand, we know that $`g1`$ in all cases. It is therefore plausible that for stars which are almost adiabatic up to the photosphere, i.e. $`M0.6\text{M}_{}`$, $`g(0)1`$. As we shall see below this is confirmed by numerical computations.
### 5.3 Results of numerical computations
Numerical computations of $`g(x)`$ and $`G(x)`$ have been carried out by HR97 for low-mass main sequence stars of Pop. I chemical composition using a full 1D stellar structure code (Hameury 1991), where the outer boundary condition was changed according to Eqs. (27) and (28) with $`F_{\mathrm{irr}}=\text{const.}`$ The results relevant for this paper are shown in Fig. 8 and can be summarized as follows:
* With $`T_B(M)`$ as shown in Fig. 7 and an appropriately chosen value for $`b`$, i.e. $`b45`$, typical for H<sup>-</sup> opacity, the predictions of our simple one-zone model are in good agreement with numerical results as long as $`T_{\mathrm{irr}}<T_B`$. In particular, the run of $`g(0)`$ shown in Fig. 8 as a full line for the numerical computations and as a short-dashed line for the one-zone model show that within its range of validity the latter reproduces the numerical results quite well.
* $`g(0)1`$ for stars with a mass $`M0.5\text{M}_{}`$, as we have argued above (Sect. 5.2).
* Max$`(xg(x))`$ $``$ 0.5 over the whole mass range of interest, i.e. $`0.1\text{M}_{}M1\text{M}_{}`$. This is shown as a dash-dotted line in Fig. 8.
* In the point source model (cf Sect. 4.2.2) the integral $`I_{PS}`$ in (57) has a maximum which is smaller than $`(1f_s)`$ Max$`(xg(x))`$. Max$`(I_{PS})`$ is shown in Fig. 8 as a long dashed line. As can be seen, Max$`(I_{\mathrm{PS}})0.20`$.
## 6 Secular evolution with irradiation
We resume the stability discussion of Sect. 4, but now with the knowledge of the function $`g(x)`$ which we have gained in Sect. 5. We shall restrict most of the following to the constant flux model (Sect. 4.2.1) and the point source model (Sect. 4.2.2). In addition, we shall assume that all the properties of a binary along a secular evolution without irradiation, in particular the functions $`\mathrm{\Gamma }`$ and $`\mathrm{\Lambda }`$ defined respectively in Eqs. (4.1) and (45), are known. Among the compact binaries we specifically discuss cataclysmic variables, and low-mass X-ray binaries. In particular, we wish to examine the following questions: a) which systems are stable (unstable) at 1) turn-on of mass transfer, or 2) during the secular evolution, against irradiation-induced runaway mass transfer, and b) what kind of evolution do systems undergo which are unstable? Because part of these questions have been dealt with extensively by King (1995), KFKR95, KFKR96, KFKR97 and MF98, we shall mainly be concerned with those aspects which have not already been treated in detail in the above papers.
### 6.1 Cataclysmic variables
In the following we set $`M_c=M_{\mathrm{WD}}`$ and $`R_c=R_{\mathrm{WD}}`$, where $`M_{\mathrm{WD}}`$ and $`R_{\mathrm{WD}}`$ are respectively the mass and the radius of the accreting white dwarf. Furthermore we shall restrict our discussion to CVs where the secondary is a low-mass main sequence star. Because this is the case for the vast majority of CVs this is not a very strong restriction. For determining the values of $`\mathrm{\Lambda }`$ and $`\mathrm{\Gamma }`$ we adopt the standard evolutionary scheme for CVs, i.e. the model of disrupted magnetic braking (e.g. King 1988 for a review) and use results of corresponding model calculations by KR92.
#### 6.1.1 Stability of mass transfer at turn-on
Because $`g(x)`$ is maximal for $`x=0`$, systems are most susceptible to irradiation at turn-on of mass transfer. The relevant stability criterion for the constant flux model is (51) and for the point source model Eq. (58). As we have already pointed out in Sect. 4 these conditions are in fact conditions on the efficiency factor $`\eta `$ because everything else is basically fixed. Given the masses of both binary components, i.e. $`M_{\mathrm{WD}}`$ and $`M_s`$, $`R_{\mathrm{WD}}`$ follows from the mass radius relation of white dwarfs, the secondary’s radius $`R_s`$ from the mass radius relation of main sequence stars, or from the evolutionary history. The orbital separation $`a`$ follows from Roche geometry via $`R_s`$ and the mass ratio, $`\zeta _S`$ and $``$ from the secondary’s internal structure, and finally $`\zeta _R`$ again from the mass ratio. Thus $`\mathrm{\Gamma }`$ (Eq. 45) is uniquely determined by $`M_{\mathrm{WD}}`$ and $`M_s`$. Given the function $`g(x)`$, and in particular $`g(0)`$, condition (51) is one for $`\eta s`$, and condition (58) one for $`\eta `$ only. Taking as an example $`g`$ from our one-zone model with $`b=4`$, $`s=0.5`$ and $`T_B/T_0`$ for $`k=1`$ from Fig. 7, we can plot a line for a given $`\eta `$ in the $`M_s`$-$`M_{\mathrm{WD}}`$ plane along which mass transfer is marginally stable. This is shown in Fig. 9 for various values of $`\eta `$. The figure is to be read as follows: a parameter combination ($`M_{\mathrm{WD}}`$, $`M_s`$) corresponds to a point in Fig. 9. If that point lies above the line corresponding to a given value of $`\eta `$, mass transfer is unstable at turn-on. What we can infer from Fig. 9 is that for typical WD masses $`M_{\mathrm{WD}}0.5\text{M}_{}`$, condition (51) is violated for surprisingly small values of $`\eta `$, i.e. $`\eta 0.1`$. This is the case, in particular, if $`M_s0.6\text{M}_{}`$. If we take instead of (51) the condition of the point source model (58), $`s`$ is no longer a parameter. Taking again the same $`g`$ as above (one-zone model with $`b=4`$) the result is qualitatively the same. The main difference is that the value of $`\eta `$ necessary for marginal stability needs to be larger by about a factor of 2.
The trends seen in the curves of Fig. 9 are easily explained: they reflect the run of $`\mathrm{\Gamma }(M_{\mathrm{WD}},M_s)`$ (cf Eq. 46). The smaller $`\mathrm{\Gamma }`$ the less stable mass transfer. Because $`\mathrm{\Gamma }`$ is proportional to $`R_{\mathrm{WD}}/M_{\mathrm{WD}}`$ and $`^1`$, and $`R_{\mathrm{WD}}/M_{\mathrm{WD}}`$ is a steeply decreasing function of $`M_{\mathrm{WD}}`$ whereas $`^1M_{\mathrm{ce}}/M_s`$ scales roughly as the relative mass of the convective envelope (cf. Fig. 1, dashed line) and thus decreases strongly with $`M_s`$, mass transfer is more likely to be unstable (stable) the higher (lower) $`M_{\mathrm{WD}}`$ and $`M_s`$. The other factors entering $`\mathrm{\Gamma }`$, i.e. $`(\zeta _S\zeta _R)`$, $`M_s/R_s`$ and $`(a/R_s)^2`$ are of comparatively minor importance. The fact that the curves in Fig. 9 are not monotonic results from the steep increase of $`T_B/T_0`$ with stellar mass near $`M=0.65\text{M}_{}`$ (cf. Fig. 7).
Because $`g^{}(x)<0`$, systems in which mass transfer is stable at turn on will also be stable against irradiation-induced runaway mass transfer for any finite value of $`F_{\mathrm{irr}}`$, i.e. $`\dot{M}_s`$. The opposite, however, is not true: not all systems which are unstable at turn-on will be so for the secular mean mass transfer rate.
#### 6.1.2 Stability of secular evolution
The appropriate stability criterion is (4.1) in its most general form, (50) for the constant flux model and (57) for the point source model. Again, the left-hand side $`\mathrm{\Lambda }`$ of these criteria is fully determined by the adopted model of secular evolution, whereas the value of the corresponding right-hand sides still depends on $`\eta `$. As we have stressed in Sect. 4, we can give a sufficient criterion for stability of mass transfer because the function $`xg(x)`$ has a maximum. The corresponding criterion is (43) in its most general form,
$$\mathrm{\Lambda }>2s\text{Max}(xg(x))$$
(73)
for the constant flux model, and
$$\mathrm{\Lambda }>\text{Max}(I_{\mathrm{PS}})$$
(74)
for the point source model. Conditions (73) and (74) can be read off Fig. 10 where we plot $`\mathrm{\Lambda }`$ as a function of secondary mass $`M_s`$ (full line) along a standard secular evolution of a CV with $`M_{\mathrm{WD}}=1\text{M}_{}`$ and an initial secondary mass $`M_{s,i}=1\text{M}_{}`$. The evolutionary data are taken from KR92. Into the same figure we plot Max$`(xg(x))`$ $`(=2s`$ Max$`(xg(x))`$ for $`s=1/2`$) and Max$`(I_{\mathrm{PS}})`$ using results by HR97 for $`g(x)`$ (dashed lines). As can be seen the values of Max$`(xg(x))`$ and Max$`(I_{\mathrm{PS}})`$ are almost independent of secondary mass, i.e. of the orbital period. For CVs above the period gap, i.e. with $`M_s0.25\text{M}_{}`$, $`\mathrm{\Lambda }`$ increases very steeply with decreasing secondary mass. This behaviour is mainly due to two factors, first to $`^1M_{\mathrm{ce}}/M_s`$ which increases strongly with decreasing mass, and second to $`\tau _{\mathrm{KH}}`$ which increases strongly towards lower masses mainly because of the mass luminosity relation $`LM^3`$ of low-mass main sequence stars. So, what $`\mathrm{\Gamma }`$ essentially represents is the thermal inertia of the convective envelope. As can be seen from Fig. 10, the intersection of $`\mathrm{\Lambda }`$ with Max$`(xg(x))`$ or Max$`(I_{\mathrm{PS}})`$ is near $`M_s=0.7\text{M}_{}`$. Because $`\mathrm{\Lambda }`$ increases so steeply, small changes in Max$`(xg(x))`$ or Max$`(I_{\mathrm{PS}})`$ do not yield a significantly different result.
When a CV approaches the period gap, i.e. the secondary becomes fully convective at $`M_s0.25\text{M}_{}`$, the system detaches and $`\mathrm{\Gamma }0`$ because $`\tau _{M_s}\mathrm{}`$. When mass transfer resumes $`\mathrm{\Lambda }`$ is smaller by typically a factor 10 than immediately above the gap, reflecting the reduced angular momentum loss below the gap and the fact that at turn-on $`R_s=R_0`$ (cf. 4.1). $`\mathrm{\Lambda }`$ starts increasing again with further decreasing mass because the Kelvin-Helmholtz time $`\tau _{\mathrm{KH}}`$ becomes very long as the secondary becomes degenerate.
Because $`\mathrm{\Lambda }`$ does not depend strongly on the mass of the white dwarf we can generalize the result found from Fig. 10: In the framework of standard CV evolution, i.e. the model of disrupted magnetic braking, CVs are stable against irradiation-induced runaway mass transfer when the mass of the secondary star is $`M_s0.7\text{M}_{}`$. If, on the other hand, $`M_s0.7\text{M}_{}`$, a system can be unstable but need not be so, subject to the value of $`\eta `$.
We note, however, that by invoking substantial consequential angular momentum loss (CAML), i.e. angular momentum loss which depends explicitly on the mass transfer rate (see King and Kolb 1995 for a discussion of CV evolution with CAML), the mass range over which systems can be unstable is much larger. This has already been pointed out by KFKR96 and confirmed in computations by MF98.
As can be seen in Fig. 10 there is also a slight chance that CVs immediately after turn-on below the period gap are unstable. According to the constant flux model (Eq. 72), some CVs could be unstable, according to the point source model, which is more realistic but still very optimistic, the stability criterion (74) is violated only marginally if at all. Therefore, our conclusion is that CVs below the period gap are very probably stable, at least in the framework of standard CV evolution.
Examining the factors which determine the value of $`\mathrm{\Lambda }`$ (Eq. 41) we see that within a given evolutionary model all factors are determined. $`\zeta _S`$, $`\tau _{\mathrm{KH}}`$ and $``$ depend only on the mass of the secondary and its evolutionary history, $`\tau _{M_s}`$ on the adopted rate of angular momentum loss, $`\zeta _S\zeta _R`$ and the evolutionary history (via $`(\mathrm{ln}R_s/t)_{\mathrm{th}}`$ in (3)), $`\zeta _R`$ on the mass ratio and $`R_s/R_0`$ on the evolutionary history (e.g. Stehle et al. 1996). We see also that making $`\mathrm{\Lambda }`$ smaller (in order to get the instability for lower secondary masses) is possible only by either increasing $`\tau _{M_s}`$, i.e. lowering the mean mass transfer rate $`\dot{M}_s`$, or by increasing $`\zeta _R`$. The former is practically impossible without upsetting the standard evolutionary paradigm for CVs, i.e. the period gap model which requires that above the period gap ($`M_s0.25\text{M}_{}`$, $`P3^h`$) $`\tau _{\mathrm{KH}}/\tau _{M_s}5`$ (e.g. Ritter 1984, King 1988; KR92; Stehle et al. 1996). Increasing $`\zeta _R`$, on the other hand, is possible only if a system experiences significant CAML.
### 6.2 Low-mass X-ray binaries
At first glance one might suspect that in LMXBs irradiation of the secondary represents a much larger threat for the stability of mass transfer than it does in CVs. However, for the following reasons this is very probably not the case: first we note the observational fact that very few of the LMXBs show X-ray eclipses. This has been interpreted as a consequence of the large vertical scale height of the X-ray irradiated accretion disk. This in turn allows the secondary to stay permanently in the disk’s shadow. If this is the case, none or at most a small part of the secondary’s surface is directly exposed to the accretion light source. Second, indirect illumination of significant parts of the donor (the high latitude regions or part of the back side) is ruled out because this would require a very extended scattering corona indeed, with a typical size of the scattering sphere (at optical depth $`\tau 1`$) of order or larger than the donor star. As a consequence one would expect X-ray eclipses to occur much more frequently than they are actually observed. Third, heat transport by currents from the hot, illuminated to the cool parts in the X-ray shadow are probably also negligible. This is because in cool stars with a deep convective envelope the superadiabatic convection zone isolates the interior from the surface. It is itself unable to transport significant amounts of heat because of its small heat capacity and the fact that the thermal time scale is much shorter than the time scale of circulation.
Heat transport by currents caused by hydrostatic disequilibrium is, however, of importance in stars with a radiative envelope. Effects of this are e.g. seen in the X-ray binaries HZ Her and V1033 Sco (see e.g. Shahbaz et al. (2000), and references therein).
Because neither indirect llumination nor heat advection can contribute significantly to the blocking of the stellar flux, the integral $`I`$ on the right-hand side of (4.1) is much smaller than for a comparable CV, despite the fact that in a LMXB there is potentially much more energy available for irradiating the secondary. The fact that $`xg(x)`$ has a maximum at $`x1`$ is the reason why even large irradiating fluxes do not help. Rather, optimal irradiation is achieved if as large a fraction as possible of the secondary’s surface is irradiated with a flux such that $`xg(x)`$ is near its maximum, i.e. if $`x1`$. This is clearly not the case in LMXBs. Not only is the surface fraction which is directly irradiated small, worse, where the surface of the secondary is directly exposed to the X-ray source, the associated flux is large, i.e. $`x1`$ unless $`\eta `$ is assumed to be very small. The latter is very unlikely considering the fact that most of the accretion luminosity emerges in form of rather hard X-rays. Thus, unless the secular evolution of LMXBs is totally unlike that of CVs as far as the nature of the secondary and the typical mass transfer rates are concerned, the value of $`\mathrm{\Lambda }`$ is virtually the same as for CVs but, as explained above, the right-hand side of (4.1) is much smaller than in CVs. Therefore we conclude that very probably LMXBs are stable against this type of irradiation-induced runaway mass transfer. As has been noted by KFKR97, systems in which accretion is intermittent rather than continuous, i.e. transient LMXBs, are even more stable.
### 6.3 Evolution of unstable systems
Let us now discuss briefly the evolution of systems (i.e. CVs) in which the stationary mass transfer given by (3) is unstable, i.e. systems for which the stability criterion (4.1) or a special form thereof (Eqs. 50 or 57) is violated. From the fact that $`g^{}(x)<0`$ we know that mass transfer is then already unstable at turn-on. Therefore, when mass transfer turns on, the mass transfer rate increases, and because (4.1) is violated, it does not settle at the secular mean $`\dot{M}_s`$ given by (3). However, because the thermal relaxation caused by irradiation saturates both in amplitude and with time (see our discussion in Sect. 3, in particular Eqs. (20) and (23), and Fig. 4) the mass transfer rate does not run away without bound. Rather, there is an upper limit: with (20) we obtain for the maximum mass transfer rate
$`\text{Max}(\dot{M}_s)`$ $``$ $`\dot{M}_s+{\displaystyle \frac{M_s}{(\zeta _S\zeta _R)}}\text{Max}\left\{{\displaystyle \frac{s_{\mathrm{eff}}}{\tau _{\mathrm{KH}}}}\right\}`$ (75)
$``$ $`\dot{M}_s+{\displaystyle \frac{s}{\zeta _S\zeta _R}}{\displaystyle \frac{M_s}{\tau _{\mathrm{KH}}}}.`$
Because mass transfer can be unstable against irradiation only if $`M_s0.7\text{M}_{}`$, i.e. when the secondary has a relatively thin convective envelope and thus $``$ is large (cf. Fig. 1), the maximum contribution of irradiation to mass transfer can be several times ($`s`$ times) the thermal time scale mass transfer rate $`M_s/\tau _{\mathrm{KH}}`$. Thus, for such systems Max$`(\dot{M}_s)\dot{M}_s`$.
After having reached the peak mass transfer rate, mass transfer cannot continue at that rate. Rather it must decrease with time for two reasons: first, thermal relaxation saturates on the time scale given by (23), i.e. $`(\mathrm{ln}R/t)_{\mathrm{irr}}`$ decreases on that time scale. Second, because mass transfer occurs at a rate above the secular mean, the binary system is driven apart, i.e. $`(d\mathrm{ln}R_R/dt)`$ $`>`$ $`(d\mathrm{ln}R_s/dt)`$. Both effects result eventually in the termination of mass transfer. The system becomes slightly detached, the secondary, in the absence of irradiation, shrinks. However, because of the absence of mass transfer the contraction of the system due to angular momentum loss is fast enough to catch up, so that mass transfer resumes and the cycle repeats again. In other words: if a system is unstable at the secular mean mass transfer rate it must undergo a limit cycle in which phases of enhanced, irradiation-driven mass transfer alternate with phases of very low or no mass transfer. The conditions for the occurrence of mass transfer cycles in semi-detached binaries have been investigated in more detail and using more general principles (in the framework of non-linear dynamics) by King (1995), KFKR95, KFKR96 and KFKR97. Their main result is that mass transfer cycles driven by radius variation of the secondary can only occur if $`(\mathrm{ln}R_s/t)_{\mathrm{th}}+(\mathrm{ln}R_s/t)_{\mathrm{nuc}}`$ in (1) depends explicitly on the instantaneous mass transfer rate. The only plausible mechanism providing such a dependence is irradiation of the donor star by radiation generated through accretion, i.e. the situation we are studying in this paper. The necessary criterion for the occurrence of mass transfer cycles found by these authors is identical to what we have found here, namely the violation of (4.1). In the framework of a linear stability analysis of mass transfer, with which we, King (1995), KFKR95, KFKR96 and KFKR97 have been concerned so far, we can not calculate the long-term evolution over time scales $`\tau _{M_s}`$ of systems under the irradiation instability. For this the full set of equations describing mass transfer and stellar structure under irradiation have to be solved. Results of such calculations and computational details will be presented in the following section. Because similar computations have already been done by HR97 and MF98, we shall concentrate here on aspects which have not been dealt with in detail by HR97 and MF98, but are, in our opinion, important for better understanding of the evolution under the irradiation instability.
## 7 Secular evolution with irradiation: numerical computations
### 7.1 Computational techniques
To compute the secular evolution of a compact binary with a low-mass donor star we have used the bipolytrope programme described in detail in KR92. With respect to the procedure described in KR92 we have, however, implemented two modifications in order to allow for a proper treatment of irradiation and its consequences for the secular evolution.
First, we compute the mass transfer rate explicitly rather than by using Eq.(3) as in KR92, i.e. we adopt the following prescription (e.g. Ritter 1988):
$$\dot{M}_s=\dot{M}_0\mathrm{exp}\left[\frac{R_RR_s}{H_P}\right]$$
(76)
Expressions for the photospheric pressure scale height $`H_P`$ and the factor $`\dot{M}_0`$ in terms of stellar parameters of the donor star and binary parameters are also given in Ritter (1988). In the context of this paper the donor star is always a low-mass main sequence star. For such stars both, $`H_P`$ and $`\dot{M}_0`$ are only weak functions of the stellar mass and have typical values $`H_P/R_s10^4`$ and $`\dot{M}_010^8\text{M}_{}`$yr<sup>-1</sup>. The reason for using (76) instead of (3) is that the latter is valid only for stationary mass transfer. However, this is not a good approximation when a system evolves through mass transfer cycles, because such cycles proceed unavoidably through phases of non-stationary mass transfer. In order to simplify the computation of $`\dot{M}_s`$ we used fixed values for $`H_P`$ and $`\dot{M}_0`$, i.e. $`H_P=10^4R_s`$ and $`\dot{M}_0=10^8\text{M}_{}`$yr<sup>-1</sup>. Furthermore we used (76) also when $`\dot{M}_s>\dot{M}_0`$, i.e. when the donor star overfills its critical Roche volume. (76) is still a reasonable approximation if $`R_sR_R<`$ few $`H_P`$, i.e. $`\dot{M}_s10^7\text{M}_{}`$yr<sup>-1</sup> (see e.g. Kolb and Ritter 1990) which is adequate for our purposes.
When computing the mass transfer rate from (76) we use for $`H_P`$ and $`\dot{M}_0`$, both of which depend on the effective temperature, the value of $`T_{\mathrm{eff}}`$ on the unirradiated part of the star, i.e. $`T_{\mathrm{eff}}=T_0`$. There are two main arguments for this choice: first, in disk-accreting systems the inner Lagrangian point $`L_1`$ is in the shadow of the disk and the cooling time of gas in the superadiabatic layer flowing from the irradiated parts towards $`L_1`$ is short compared to the flow time in the shadow region. Second, if we use $`T_{\mathrm{eff}}=T_{\mathrm{irr}}`$ instead of $`T_{\mathrm{eff}}=T_0`$, $`H_P`$ and $`\dot{M}_s`$ would react practically instantaneously and with a large amplitude to irradiation giving, in turn, rise to a runaway of the mass transfer rate on an extremely short time scale (order of hours). This is obviously not what happens in the systems we do observe.
Second we use either Eq. (52) in the constant flux model or Eq. (59) in the point source model of irradiation in place of the usual Stefan-Boltzmann law as one of the outer boundary conditions for the donor star model. For the calculations presented below we have used the one-zone model described in Sect. 5 rather than results obtained from full stellar models by HR97 described earlier. Specifically, $`T_{\mathrm{irr}}`$ in (52) or (59) was computed by solving (68) in which $`F_{\mathrm{irr}}`$ is determined for an adopted value of $`\eta `$ from (24) with $`h(\vartheta ,\phi )`$ taken respectively from (48) and (49) in the case of the constant flux model, and from (54) in the case of the point source model. $`T_B`$ in (68) is taken from the numerical results shown in Fig. 7. Specifically, we have used $`T_B(M_s)`$ for $`k=1`$. Furthermore, in order to determine $`T_{\mathrm{irr}}`$ we need to specify $`b`$, i.e. the temperature exponent of the photospheric opacity law (10). For most of our experiments we have used $`b=4`$ which is adequate for H<sup>-</sup> opacity.
The diagnostic quantities $`\zeta _{\mathrm{irr}}`$ and $`s_{\mathrm{eff}}`$ are calculated from (4.1) and respectively (53) or (60) with the appropriate choice of $`h(\vartheta ,\phi )`$, i.e. Eqs. (48) or (54), and $`g(\vartheta ,\phi )`$ from (72).
### 7.2 Computational limitations
By using the bipolytrope approximation for describing the internal structure of the donor star, we can only deal with chemically homogeneous stars which have a radiative core and a convective envelope or are fully convective, i.e. with low-mass $`(M_s1\text{M}_{})`$ zero age main sequence stars. Thus, we are unable to address chemically evolved stars, in particular subgiants. Giants, on the other hand, have been dealt with approximately by KFKR97. We note also that chemically evolved donors among CVs might be more common than hitherto thought (Kolb and Baraffe 2000; Ritter 2000; Baraffe and Kolb 2000) and therefore deserve further study. For our computations, we have adopted Pop. I chemical composition. In the context of the bipolytrope approximation, the chemical composition is of relevance mainly for determining the appropriate gauge functions, i.e. polytropic index $`n_1`$ of the radiative core and entropy jump $`h`$ in the surface layers as a function of stellar mass (for details see KR92). Even when adopting the appropriate gauge functions, we know that the value of $`\zeta _S`$ computed in the bipolytrope approximation is smaller than what one would obtain for a full stellar model if $`M0.6\text{M}_{}`$. Because of this, in the bipolytrope approximation binary systems are more likely to be unstable against irradiation than they are in reality if $`M_s0.6\text{M}_{}`$.
We have used the one-zone model described in Sect. 5. As we have discussed there, application of this model is practically restricted to stars with a mass $`M_s0.60.7\text{M}_{}`$. We have, however, carried out a few calculations of systems with a smaller secondary mass. In those cases we have assumed $`g(x)=1`$ if $`0x1`$ and $`g=0`$ otherwise. This corresponds to what (72) yields in the limit $`T_BT_0`$ and approximates results of numerical calculations, at least for small fluxes, i.e. $`x<1`$, reasonably well.
Our calculations contain a number of free or at best not well-constrained parameters. To mention just the most important ones: in the constant flux model we already make a very simplifying assumption about the function $`h(\vartheta ,\phi )`$, i.e. the irradiation geometry. This assumption results in the free parameter $`s`$. In addition, we have the efficiency factor $`\eta `$, a parameter about which we know little beyond the fact that probably $`0<\eta 1`$. Using our one-zone model (Sect. 5) introduces furthermore the parameters $`T_B`$ and $`b`$, both of which can, however, be fixed reasonably well by comparison with full stellar models. In the point source model we do not need to specify $`s`$. But all the other parameters, i.e. $`\eta `$, $`T_B`$ and $`b`$ remain in the problem. By using numerical results for the functions $`G(x)`$ and $`g(x)`$ would get rid of the parameters $`T_B`$ and $`b`$. But we would still be left with specifying $`h(\vartheta ,\phi )`$ and $`\eta `$. On top of all that we have also a number of input parameters and functions which are already needed for computing a secular evolution without irradiation. Apart from the binary’s initial parameters the most important of those are the angular momentum loss rate and parameters arising from assumptions about mass and consequential angular momentum loss from the binary system.
The purpose of the following computations, going beyond the linear stability analysis, is to illustrate the temporal evolution under the irradiation instability and a number of its specific properties which we have uncovered in the foregoing discussion.
As we have demonstrated in the previous section the irradiation instability is more likely of relevance for CVs than for LMXBs. Consequently, in the examples below we have assumed that the accretor is a white dwarf with a mass $`M_{\mathrm{WD}}=1\text{M}_{}`$ and a radius according to the mass radius relation (e.g. Nauenberg 1972) of $`R_{\mathrm{WD}}=\mathrm{5\; 10}^8`$cm. We assume that during the secular evolution the mass of the white dwarf remains constant, i.e. $`\dot{M}_{\mathrm{WD}}=0`$, and that on average the transferred matter leaves the system with the specific orbital angular momentum of the white dwarf. Other parameters characterizing the three examples which we shall discuss subsequently in detail are listed in Table 2.
### 7.3 Numerical results
For the first example of a secular evolution with irradiation, results of which are shown in Fig. 11, we have adopted the constant flux model with $`s=0.5`$, an initial secondary mass $`M_{s,i}=0.8\text{M}_{}`$ and loss of orbital angular momentum on a constant time scale $`\tau _J=J/\dot{J}=\mathrm{1.27\; 10}^8`$yr derived from the Verbunt and Zwaan (1981) prescription for magnetic braking with $`f_{\mathrm{vz}}=1`$. The value $`\eta =0.035`$ was chosen such that on the one hand initially $`\zeta _{\mathrm{irr}}>\zeta _S\zeta _R`$ and on the other $`T_{\mathrm{irr}}<T_B`$ at all times. The top panel of Fig. 11 shows the run of the mass transfer rate with irradiation (full line) and without (dashed line), the second the evolution of the secondary’s radius with irradiation (full line) and without (dashed line). In the third panel we show the run of $`\zeta _{\mathrm{irr}}`$ (full line) and of $`\zeta _S\zeta _R`$ (dash-dotted line) with time. Finally the bottom panel shows the run of $`s_{\mathrm{eff}}`$ with time. As was to be expected for a system in which (initially) $`\zeta _{\mathrm{irr}}>\zeta _S\zeta _R`$ mass transfer evolves through cycles with at least initially large amplitude. The amplitude of the radius variations, on the other hand, are small, a consequence of the small value of $`H_P/R=10^4`$. According to (76) we have $`\mathrm{\Delta }\mathrm{log}R_s=H_P/R\mathrm{\Delta }\mathrm{log}\dot{M}_s`$. We see also that during the mass transfer peaks a significant fraction of the stellar luminosity is blocked. Because $`\eta `$ has been chosen such that always $`T_{\mathrm{irr}}<T_B`$ we always have $`s_{\mathrm{eff}}<s`$. What is immediately apparent from this calculation is that though the system evolves through mass transfer cycles, these are damped on a rather short time scale, i.e. a time scale much shorter than $`\tau _J`$. The reason for this is that $`\zeta _{\mathrm{irr}}`$ decreases rapidly with time (mass of the donor). Eventually $`\zeta _{\mathrm{irr}}<\zeta _S\zeta _R`$ and mass transfer becomes stable. This is mainly a consequence of the increasing thermal inertia of the convective envelope, i.e. of a decrease of $`/\tau _{\mathrm{KH}}`$ with decreasing donor mass. The damping of the oscillations could only be overcome if at the same time $`g`$ increases sufficiently strongly with decreasing mass. We shall show below that in a restricted mass range this is indeed possible.
Given the results of the above example we now ask to what extent they are representative, i.e. whether the qualitative behaviour depends strongly on the adopted model parameters or not. With this end in view we have carried out numerous experiments, the results of which we shall now discuss.
Working with the one-zone model (Sect. 5) we ask first how the above results change with the parameter $`b`$, i.e. the adopted photospheric opacity. Inspection of Eq. (72) shows that for given values of $`T_B`$ and $`T_{\mathrm{irr}}`$ $`g(0)`$ increases with $`b`$. This means that at least for small fluxes the donor star is more sensitive to irradiation for larger $`b`$. The reason for this is easy to understand: the larger $`b`$ the more pronounced the increase of the optical depth in the superadiabatic layer in response to irradiation, i.e. of the average temperature, and thus the more effective the blocking of the energy outflow from the adiabatic interior. Thus, increasing (decreasing) $`b`$ above (below) the value $`b=4`$ we have used in the example shown in Fig. 11 results in more (less) pronounced mass transfer cycles. The time scale on which the mass transfer oscillations are damped remains, however, essentially uneffected by changes of $`b`$, reflecting the fact that $`/\tau _{\mathrm{KH}}`$ does not depend on $`b`$.
Next we compare the results obtained with the constant flux model (Fig. 11) with those obtained with the point source model. The results of a run with the latter model and parameters identical to those used for producing Fig. 11 (except of $`s`$ which is not a free parameter in this model) are qualitatively very similar to those found in Fig. 11. The amount of stellar flux blocked during a mass transfer peak is, however, systematically smaller (by about a factor of two) than in the constant flux model. The main reason for this is that the irradiated area on the donor $`s_{\mathrm{PS}}=0.5(1f_s)=0.32`$ is smaller by about a factor of 1.6 than what we have assumed in the constant flux example, i.e. $`s=0.5`$. Therefore, in order to achieve the same effect as in the constant flux model with $`s=0.5`$, $`\eta `$ in the point source model needs to be increased by about a factor of two. Otherwise the results obtained from the two models are very similar. Because of this and because the constant flux model is computationally much less demanding we have performed most of our simulations with that model.
Next we examine briefly the dependence on the initial mass of the donor star. From our extensive discussions in Sects. 4 and 6 we know already that below a critical mass which is between about 0.6 and $`0.7\text{M}_{}`$, depending on the adopted models, systems following a standard CV evolution are stable against irradiation.
Adopting the constant flux model and the one-zone model, a system with an initial secondary mass of $`0.7\text{M}_{}`$ is still unstable (whereas with the point source model and a more realistic $`g(x)`$ such a system would be stable). Results of an evolution with $`s=0.5`$ and $`\eta =0.04`$ and the remaining parameters as in Fig. 11 (cf. Table 2) are shown in Fig. 12. As can be seen this evolution differs in several respects from the one shown in Fig. 11. Initially the amplitude of the mass transfer cycles increases with time. After only a few cycles the peak mass transfer rate is so high that the irradiation effect saturates, i.e. $`T_{\mathrm{irr}}>T_B`$ and $`s_{\mathrm{eff}}=s=1/2`$. Although the one-zone model used here does not apply when $`T_{\mathrm{irr}}`$ approaches $`T_B`$, the main effect of saturation can be modelled anyway by setting $`g=0`$ if $`T_{\mathrm{irr}}T_B`$. The qualitative behaviour obtained in this way remains the same as if a more realistic and smooth function $`g(x)`$ was used. After an initial phase of increasing amplitudes they later start decreasing and die out very rapidly after about $`\mathrm{4\; 10}^7`$yr. This behaviour can be understood as follows: We have already pointed out above that increasing amplitudes of the cycles can only be expected if with decreasing mass $`g(x)`$ increases fast enough. This is exactly what happens in the evolution shown in Fig. 12. The fast increase of $`g(x)`$ is the result of the fast decrease of $`T_B/T_1`$ when going from $`M_s0.7\text{M}_{}`$ to $`M_s0.6\text{M}_{}`$ (see Fig. 7). Below $`M_s0.6\text{M}_{}`$, $`g(x)`$ does no longer change much with mass. As a result, the mass transfer cycles are then damped because of the secular increase of $`\tau _{\mathrm{KH}}/`$. When the system eventually becomes stable, irradiation is still important in blocking the energy outflow. As can be seen from the bottom panel of Fig. 12, $`s_{\mathrm{eff}}`$ is of the order 0.37 after the system has stabilized.
Going to even lower initial secondary masses, mass transfer cycles cannot occur unless either $`\zeta _S\zeta _R`$ or $`\dot{M}_s`$ is much smaller than in a standard CV evolution (cf. our discussion in Sect. 6.1). For illustrating this we show in Fig. 13 the results of a calculation for which we have assumed $`M_s=0.25\text{M}_{}`$, $`\eta =0.20`$, $`s=0.5`$ and the much smaller angular momentum loss rate of gravitational radiation. Thus these parameters mimic an unstable system just below the period gap. As we have seen at the end of Sect. 6.1, CVs just below the period gap can be unstable if the constant flux model with $`s=0.5`$ is adopted. They are, however, stable if the point source model is used, unless $`\zeta _R`$ is lowered below the standard value by invoking CAML (see MF98). If we use in addition to the constant flux model also the one-zone model in the limit $`T_BT_1`$, at low fluxes the donor star is even more susceptible to irradiation than if a more realistic form of $`g(x)`$ had been used. However, for the purpose of this exercise it does not matter whether CVs below the gap are stable or not. This example was chosen just to demonstrate that with a low enough $`\dot{M}_s`$ and depending on the value of $`\eta `$ a system could evolve through mass transfer cycles as we have concluded from our stability discussion in Sects. 4 and 6. One additional property of this run which is worth mentioning is that the cycles are only rather weakly damped. The reason for this is that a star with a mass $`M_s<0.25\text{M}_{}`$ is always fully convective and therefore $`=7/3`$ remains constant. Thus the decrease of $`\zeta _{\mathrm{irr}}`$ is mainly due to the slow increase of $`\tau _{\mathrm{KH}}`$ with time.
## 8 Discussion and conclusions
With the exception of the numerical examples presented in the previous section, where we have assumed the donor star to be on the ZAMS, we have so far not made any explicit assumption about the evolutionary status of the secondary. In fact our stability considerations are valid for any type of donor star as long as it has a sufficiently deep outer convective envelope. In the following we shall therefore briefly address the stability of mass transfer and the peak mass transfer rate in case of instability if the secondary is more or less nuclear evolved. In addition we shall briefly discuss the impact of the donor’s metallicity on the stability of mass transfer. A detailed investigation of the irradiation instability in those cases is, however, deferred to a subsequent paper.
### 8.1 Effects of evolution and metallicity on the stability of mass transfer
Returning to the stability criterion (4.1) we see that the left-hand side $`\mathrm{\Lambda }`$ depends essentially only on the ratio of two time scales, i.e. the thermal time scale of the convective envelope $`\tau _{\mathrm{ce}}`$ and the mass transfer time scale $`\tau _{M_s}`$ or, respectively, on the time scale on which mass transfer is driven $`\tau _{\mathrm{dr}}=(2/\tau _J+1/\tau _{\mathrm{nuc}})^1`$. Mass transfer is more stable the larger the ratio $`\tau _{\mathrm{ce}}/\tau _{\mathrm{dr}}`$.
Evolutionary computations such as those shown in Ritter (1994) show that at a given secondary mass $`\tau _{\mathrm{ce}}`$ is the shorter the more evolved the star. If magnetic braking according to Verbunt & Zwaan (1981) is invoked for computing the angular momentum loss rate, numerical computations (e.g. those shown in Ritter 1994, or King, Kolb & Burderi 1996) of the mass transfer from a nuclear evolved donor show that $`\tau _{\mathrm{dr}}`$ is the longer the more evolved the donor at the onset of mass transfer. Taken together this means that mass transfer is the more unstable (less stable) the more evolved the donor. As an example, evaluating $`\mathrm{\Lambda }`$ along the evolutionary tracks shown in Ritter (1994) we find that as long as the secondary has not reached the terminal age main sequence, mass transfer is stable if $`P5^h6^h`$. In the sequence starting with a $`1.2\text{M}_{}`$ secondary at the end of central hydrogen burning, systems down to an orbital period $`P2^h`$ may be unstable.
Main sequence stars of low metallicity are systematically hotter, smaller, more luminous and have a thinner convective envelope than main sequence stars of solar metallicity. Hence $`\tau _{\mathrm{ce}}`$ is systematically shorter for low metallicity stars. At the same time $`\tau _{\mathrm{M}_\mathrm{s}}`$ is the shorter the lower the metallicity. Taken together this means that mass transfer is systematically more unstable (less stable) the lower the metallicity. As an example, using results from Stehle et al. (1997), we find that in CVs with a Pop. II donor mass transfer could be unstable for orbital periods as low as $`4^h`$.
The case of systems containing a giant donor and in which mass transfer is driven by its nuclear evolution, i.e. $`\tau _{\mathrm{dr}}=\tau _{\mathrm{nuc}}`$, has already been dealt with in some detail by KFKR97. We shall mention here only that such systems are systematically more unstable against irradiation-induced runaway mass transfer than systems with a main sequence donor. There are three reasons for this: First, for a giant $`\tau _{\mathrm{ce}}`$ is systematically shorter than for a main sequence star of the same mass (because of the much larger values of $`R`$ and $`L`$), and this despite the fact that $`M_{\mathrm{ce}}/M`$ can be quite large. Second, in terms of radius changes due to irradiation, giants react much more strongly than main sequence stars. (Note that in the framework of the homology model presented in Sect. 3 the effective $`\nu `$ for giants is -3!) Third, the mass transfer time scale $`\tau _{\mathrm{M}_\mathrm{s}}`$associated with nuclear evolution of a giant is always much longer than $`\tau _{\mathrm{ce}}`$ over the entire range of core masses of interest, i.e. $`0.15\text{M}_{}M_\mathrm{c}0.5\text{M}_{}`$. For details see KFKR97 and Ritter (1999).
### 8.2 The peak mass transfer rate
Our estimate (75) of the peak mass transfer rate achieved during a mass transfer cycle can be rewritten as
$$\mathrm{Max}(\dot{M}_s)\dot{M}_s\left[1+\frac{2s}{\mathrm{\Lambda }}\right].$$
(77)
In the case of a fully developed instability it is the second term in (77) which dominates. Therefore, the peak mass transfer rate is essentially determined by the rate of expansion of the donor, i.e.
$$\mathrm{Max}(\dot{M}_s)\dot{M}_s\frac{2s}{\mathrm{\Lambda }}\frac{M_s}{\zeta _S\zeta _R}\frac{s}{\tau _{\mathrm{ce}}}.$$
(78)
Thus the lower the thermal inertia of the convective envelope the higher the peak mass transfer rate. As a consequence, very high peak rates can result if the mass of the convective envelope or $`\tau _{\mathrm{KH}}`$ is small. If this is the case the long-term evolution of the corresponding systems could be drastically changed. We note e.g. that even if the donor is a main sequence star, the peak mass transfer rate can easily exceed the value required for maintaining stable nuclear burning on the white dwarf, i.e. $`\dot{M}_{\mathrm{WD}}10^7\text{M}_{}\mathrm{yr}^1`$ (e.g. Fujimoto 1982). With a Pop. I ZAMS donor this limit is reached if $`M_s0.8\text{M}_{}`$ (see Fig. 11), with a donor at the end of central hydrogen burning if $`P10^h`$ or $`M_s0.6\text{M}_{}`$. These values are derived from the evolutionary calculations discussed in Ritter (1994).
The consequences of reaching peak mass transfer rates equal to or even in excess of the stable nuclear burning limit of the white dwarf can be far-reaching. First, if steady nuclear burning on the white dwarf is reached, such a system will no longer appear as an ordinary CV but rather look more like a supersoft X-ray source. Because such systems are bright in the EUV only, they become virtually undetectable in our Galaxy: very few such systems have indeed been found (see e.g. Greiner 1996). Second, with nuclear burning on the white dwarf, the nuclear luminosity exceeds the accretion luminosity by a factor of $`1010^2`$. With so much more irradiation luminosity available effects other than those discussed in this paper might also become important, e.g. driving of a strong wind from the donor (see van Teeseling and King 1998, King and van Teeseling 1998), and which would change the evolution of such systems altogether. Third, an unavoidable consequence of the very high peak mass transfer rates are very extended low states during which a system is practically detached and thus virtually undetectable. Fourth, from the fact that such systems are barely detectable in both the high and low state and that the transition time between high and low states and vice versa is very short (see KFKR97), one is practically forced to the conclusion that the observed long-period CVs are either stable against the irradiation instability or, for some other reason do not reach peak rates equal to or larger than the stable nuclear burning limit.
### 8.3 Conclusions
In this paper we have studied the reaction of low-mass stars to anisotropic irradiation and its implications for the long-term evolution of compact binaries. For this we have shown that the case of anisotropic irradiation in close binaries is relevant and that spherically symmetric irradiation is probably not an adequate approximation. We have studied the reaction of low-mass stars to anisotropic irradiation by means of simple homology considerations. We have shown in the framework of this model that if the energy outflow through the surface layers of a low-mass main sequence star is blocked over a fraction $`s_{\mathrm{eff}}<1`$ of its surface, it will inflate only modestly, by about a factor $`(1s_{\mathrm{eff}})^{0.1}`$ (see Eq. 3) in reaching a new thermal equilibrium, and that the maximum contribution to mass transfer due to thermal relaxation is $`s_{\mathrm{eff}}`$ times the value one obtains for spherically symmetric irradiation (i.e. $`s_{\mathrm{eff}}=1`$). The duration of the thermal relaxation phase is $`0.1\tau _{\mathrm{ce}}|\mathrm{ln}(1s_{\mathrm{eff}})|`$ (Eqs. 22 and 23), where $`\tau _{\mathrm{ce}}`$ is the thermal time scale of the convective envelope. In addition, we have carried out numerical computations of the thermal relaxation process of low-mass stars involving full stellar models and using the modified Stefan-Boltzmann law (9) as an outer boundary condition. The results of these computations (shown in Figs. 2-4) fully confirm results from homology and show that the effects caused by anisotropic irradiation are not only quantitatively but also qualitatively different from those caused by spherically symmetric irradiation.
Next we have carried out a detailed stability analysis. The criterion for stability against irradiation-induced runaway mass transfer in its most general form (arbitary irradiation geometry) is given in Eqs. (4.1) and (45). One of the remarkable results of this stability investigation is that it is not arbitrarily large irradiation fluxes which destabilize a system most effectively. Rather the most effective irradiating flux is $`F_{\mathrm{irr}}F_0`$, where $`F_0`$ is the surface flux of the unperturbed star.
The reaction of the stellar surface to irradiation is best expressed in terms of a function $`g(x)=dF/dF_{\mathrm{irr}}`$, where $`x=F_{\mathrm{irr}}/F_0`$ is the normalized flux. General considerations show that $`\mathrm{Max}(xg(x))<1`$. For determining $`g(x)`$ we used a simple, analytic one-zone model for the superadiabatic convection zone. The results of this simple model (Eq. 72) are found to be in good qualitative and satisfactory quantitative agreement with results of computations involving full stellar models (see Fig. 8).
Application of our stability analysis to CVs and LMXBs yields the following results:
CVs which evolve according to the standard evolutionary paradigm, i.e. the model of disrupted magnetic braking, are stable against irradiation-induced runaway mass transfer if the mass of the (ZAMS) donor is $`M_s0.7\text{M}_{}`$. This holds unless substantial consequential angular momentum losses greatly destabilize the systems. Systems in which the mass of the (ZAMS) secondary is $`M_s0.7\text{M}_{}`$ can be unstable but need not be so, depending on the efficiency of irradiation $`\eta `$ (defined in Eq. (24)). Substantial consequential angular momentum losses can however destabilize CVs over essentially the whole range of periods of interest.
CVs with a Pop. II or an evolved secondary are inherently less stable than CVs with a Pop. I ZAMS secondary. The latter are those stars which, for a given mass, have the highest thermal inertia, making the corresponding CVs the most stable ones.
If mass transfer is unstable, we found that it must evolve through a limit cycle in which phases of irradiation-induced mass transfer alternate with phases of small (or no) mass transfer. At the peak of a cycle mass transfer proceeds on a time scale which is roughly $`1/s_{\mathrm{eff}}`$ times the thermal time scale of the convective envelope (see Eq. 78). With decreasing mass of the secondary the amplitude of the mass transfer cycles gets smaller and the cycles eventually disappear (after a system has become stable) because the thermal inertia of the secondary (expressed in the functions $`\mathrm{\Lambda }`$ and $`\mathrm{\Gamma }`$ defined respectively in Eqs. (4.1) and (45)) increases strongly (see Figs. 10 and 11). A necessary condition for the maintenance of cycles is that the thermal time scale of the convective envelope has to be much shorter ($`0.05`$) than the time scale on which mass transfer is driven.
LMXBs are very likely to be stable because a) the donor star is strongly shielded from direct irradiation over most of the hemisphere facing the X-ray source, and b) because where this is not the case, $`xg(x)`$, i.e. the sensitivity of the stellar surface to changes in the irradiating flux, is very small.
###### Acknowledgements.
We acknowledge support from a Royal Society/Chinese Academy of Sciences Joint Project. ZZ acknowledges support from Chinese National Natural Science Foundation. Major parts of this work have been completed while ZZ was visiting the MPA Garching, funded by the Max-Planck-Gesellschaft. We thank Andrew King for improving the language of the manuscript and an anonymous referee for helpful comments.
## Appendix A Lateral heat diffusion
Here we show that in a realistic situation the lateral temperature gradient caused by irradiating the secondary anisotropically is negligible compared to the radial temperature gradient in the subphotospheric layers and that, therefore, lateral energy transport by radiation is unimportant. We demonstrate this in the framework of the point source model. In that model we estimate the lateral gradient of $`T_{\mathrm{irr}}^4`$ to be
$`(T_{\mathrm{irr}}^4)_\vartheta `$ $``$ $`{\displaystyle \frac{T_{\mathrm{irr}}^4(\vartheta =0)T_{\mathrm{irr}}^4(\vartheta =\vartheta _{\mathrm{max}})}{R_s\vartheta _{\mathrm{max}}}}`$ (79)
$`=`$ $`{\displaystyle \frac{T_{\mathrm{irr}}^4(\vartheta =0)T_0^4}{R_s\vartheta _{\mathrm{max}}}}.`$
On the other hand, the radial temperature stratification is given by the Eddington approximation (e.g. Tout et al. 1989)
$`T^4(\tau ,\vartheta )`$ $`=`$ $`{\displaystyle \frac{3}{4}}\left(T_{\mathrm{irr}}^4(\vartheta ){\displaystyle \frac{F_{\mathrm{irr}}(\vartheta )}{\sigma }}\right)\left(\tau +{\displaystyle \frac{2}{3}}\right)`$ (80)
$`+`$ $`{\displaystyle \frac{F_{\mathrm{irr}}(\vartheta )}{\sigma }},`$
where $`\tau `$ is the optical depth. Therefore
$$(T^4)_r=\frac{dT^4(\tau ,\vartheta )}{dr}=\frac{3}{4}\left(T_{\mathrm{irr}}^4(\vartheta )\frac{F_{\mathrm{irr}}(\vartheta )}{\sigma }\right)\frac{d\tau }{dr}$$
(81)
and
$$\text{Min}|(T^4)_r|=\frac{3}{4}\left(T_{\mathrm{irr}}^4(\vartheta =0)\frac{F_{\mathrm{irr}}(\vartheta =0)}{\sigma }\right)\frac{d\tau }{dr}.$$
(82)
From the definition of the optical depth we have
$$\frac{d\tau }{dr}=\kappa \varrho .$$
(83)
Inserting (83) into (82) yields the minimum radial temperature gradient in the presence of irradiation:
$$\text{Min}|(T^4)_r|=\frac{3}{4}\left(T_{\mathrm{irr}}^4(\vartheta =0)\frac{F_{\mathrm{irr}}(\vartheta =0)}{\sigma }\right)\kappa \varrho .$$
(84)
Combining now (79) and (84) we obtain
$`\text{Max}|{\displaystyle \frac{(T^4)_\vartheta }{(T^4)_r}}|`$
$`{\displaystyle \frac{4}{3\vartheta _{\mathrm{max}}}}{\displaystyle \frac{T_{\mathrm{irr}}^4(\vartheta =0)T_0^4}{T_{\mathrm{irr}}^4(\vartheta =0)F_{\mathrm{irr}}(\vartheta =0)/\sigma }}{\displaystyle \frac{1}{\kappa \varrho R_s}}.`$ (85)
Now, because $`4/3\vartheta _{\mathrm{max}}1`$ and in the subphotospheric layers $`\kappa \varrho H_P1`$, we have
$`\text{Max}|{\displaystyle \frac{(T^4)_\vartheta }{(T^4)_r}}|`$ $``$ $`{\displaystyle \frac{H_P}{R_s}}{\displaystyle \frac{T_{\mathrm{irr}}^4(\vartheta =0)T_0^4}{T_{\mathrm{irr}}^4(\vartheta =0)\frac{F_{\mathrm{irr}}(\vartheta =0)}{\sigma }}}`$ (86)
$`=`$ $`{\displaystyle \frac{H_P}{R_s}}{\displaystyle \frac{G(x(\vartheta =0))+x(\vartheta =0)1}{G(x(\vartheta =0))}}.`$
Because in the stars in question $`H_P/R_s10^4`$ is so small, the lateral temperature gradient is always much smaller than the radial one unless $`x(\vartheta =0)1`$ and thus $`G(x(\vartheta =0))`$ $`1`$. However, in the range of interest of $`x`$ and $`G(x)`$, i.e. where $`xg(x)`$ is near its maximum and $`x1`$ to a few, (86) yields that the lateral temperature gradient is of order $`(T)_\vartheta H_P/R_s(T)_r(T)_r`$.
|
warning/0005/cond-mat0005535.html
|
ar5iv
|
text
|
# Onset of dielectric modes at 110𝐾 and 60𝐾 due to local lattice distortions in non-superconducting 𝑌𝐵𝑎₂𝐶𝑢₃𝑂_6.0 crystals
## Abstract
We report the observation of two dielectric transitions at $`110K`$ and $`60K`$ in the microwave response of non-superconducting $`YBa_2Cu_3O_{6.0}`$ crystals. The transitions are characterized by a change in polarizability and presence of loss peaks, associated with overdamped dielectric modes. An explanation is presented in terms of changes in polarizability of the apical $`O`$ atoms in the $`BaO`$ layer, affected by lattice softening at $`110K,`$ due to change in buckling of the $`CuO`$ layer. The onset of another mode at 60K strongly suggests an additional local lattice change at this temperature. Thus microwave dielectric measurements are sensitive indicators of lattice softening which may be relevant to superconductivity.
It was recognized soon after the discovery of the high temperatures superconductors that the cuprates are structurally similar to the ferroelectric perovskites. The basic perovskite $`ABO_3`$ structure occurs in ferroelectrics like $`BaTiO_3`$ and incipient ferroelectrics or quantum paraelectrics such as $`SrTiO_3`$, as well as in sub-units of the superconductor $`YBa_2Cu_3O_{6.0+x}`$. The implications of this structural similarity received early support from the observation of large dielectric response in the insulating parent compound $`YBa_2Cu_3O_{6.0}`$. Furthermore, theoretical models have been proposed which include the possible competition between ferroelectricity and superconductivity .
In this paper we show some striking dielectric properties of single crystals of insulating $`YBa_2Cu_3O_{6.0}`$ which seem to have a strong bearing on the superconductivity of the doped YBCO. Traditionally the information on lattice dynamics has been obtained from inelastic neutron, XAFS, Raman and infrared spectroscopy measurements. Our microwave measurements probe consequences of lattice modes on the long wavelength ($`q=0`$) dielectric properties, and its very high sensitivity leads to the observation of features not detected by other techniques.
In addition to large dielectric strengths $`\epsilon ^{}10^210^3`$, consistent with previous measurements, we report the presence of two dielectric transitions at $`110K`$ and $`60K`$. These transitions are accompanied by the onset of polarization modes indicated by the presence of dielectric loss peaks below the transition temperatures. The transitions arise from structural distortions occurring at these temperatures, such as buckling of the $`CuO`$ plane leading to the $`110K`$ transition, which affect the electrodynamic response. Thus precision microwave measurements are shown to be a sensitive probe of lattice effects, complementing other traditional probes of lattice dynamics. Taken together with numerous reports of lattice effects at or near the superconducting transition temperature $`T_c`$ , the present results demonstrate the importance of charge and lattice dynamics in the high temperature superconducting oxides.
Ultra-pure single crystals of $`YBa_2Cu_3O_{6+x}`$ were prepared in contamination-free $`BaZrO_3`$ crucibles. The high quality of these single crystals, which have also been prepared in the entire composition range from insulating, non-superconducting material ($`x=0`$) to optimum ($`x=0.95`$) and over-doped ($`x=1.0`$), has been extensively documented in a wide range of measurements, including structural and transport studies of the superconducting and non-superconducting states. A brief list of these reports can be found in . In this paper we focus on new results on the insulating $`x=0.0`$ compound.
The high sensitivity microwave measurements were carried out in a $`Nb`$ superconducting cavity resonant at $`10GHz`$ in the $`TE_{011}`$ mode. The sample is placed at the center of the cavity at a maximum of the microwave magnetic field $`H_\omega `$. We introduce an electromagnetic susceptibility $`\stackrel{~}{\zeta }_H(T)=\zeta _H^{}(T)+i\zeta _H^{\prime \prime }(T)`$ which is related to the measured parameters, the shift in cavity resonant frequency $`\delta f(T)`$ and the resonance width $`\mathrm{\Delta }f(T)`$ by $`\delta f(T)i\mathrm{\Delta }f(T)=g(\zeta _H^{}(T)+i\zeta _H^{\prime \prime }(T))`$, where $`g`$ is a geometric factor. A detailed analysis of the relevant cavity perturbation for general sample conditions including lossy dielectric and metallic or superconducting states, has been recently carried out by us . We are able to directly measure the conductivity $`\stackrel{~}{\sigma }_{tot}`$ or the dielectric permittivity $`\stackrel{~}{\epsilon }_{tot}`$ (where $`\stackrel{~}{\sigma }_{tot}=i\omega \epsilon _0\stackrel{~}{\epsilon }_{tot}`$). The analysis shows that for arbitrary conductivity,
$`\stackrel{~}{\zeta }_H(T)`$ $`=`$ $`{\displaystyle \frac{3}{2}}\left[13/\stackrel{~}{z}^2+3\mathrm{cot}(\stackrel{~}{z})/\stackrel{~}{z}\right]\text{ };`$ (1)
$`\text{where }\stackrel{~}{z}`$ $`=`$ $`ka=k_0a\sqrt{\stackrel{~}{\epsilon }_{tot}}.`$ (2)
Note that we use time dependences $`e^{i\omega t}`$. In the limit $`\stackrel{~}{z}1`$, $`\stackrel{~}{\zeta }_H(T)(1/10)(k_0a)^2(\stackrel{~}{\epsilon }_{tot}1)`$ for a lossy dielectric. The dielectric permittivity $`\stackrel{~}{\epsilon }_{tot}`$ was extracted from the data using Eq. 1. $`\stackrel{~}{\epsilon }_{tot}`$ includes both bound polarization ($`\stackrel{~}{\epsilon }`$) and free charge conductivity $`\stackrel{~}{\sigma }`$ contributions, i.e. $`\stackrel{~}{\epsilon }_{tot}=\stackrel{~}{\epsilon }+i\stackrel{~}{\sigma }/\omega \epsilon _0`$. In the present case the conductivity is negligible and $`\stackrel{~}{\epsilon }_{tot}=\stackrel{~}{\epsilon }`$. We have earlier carried out extensive measurements of the surface impedance of a variety of superconductors, metals and insulators, and demonstrated the validity of these measurements.
The dielectric permittivity $`\epsilon ^{}(T)`$ and $`\epsilon ^{\prime \prime }(T)`$ of $`YBa_2Cu_3O_{6.0}`$ are shown in Fig. 1. Here $`H_\omega ||\widehat{c}`$-axis, so that the displacement currents are in the $`ab`$-plane, i.e. we are measuring in-plane dielectric permittivity $`\stackrel{~}{\epsilon }_{ab}`$. The large microwave dielectric permittivity observed in the present composition seems to be a characteristic of some perovskite oxides. Such large dielectric strengths $`\epsilon ^{}10^210^3`$ in non-metallic insulating $`YBa_2CuO_{6.0+x}`$ were reported by Rey, et. al., for ceramic samples which were quenched to retain the oxygen homogeneity. It is worth remarking that the present crystals are also quenched from high temperature and this may be an important requirement for the observation of this effect. We have observed similar response in the microwave dielectric permittivity of another $`YBa_2Cu_3O_{6.0}`$ crystal (Fig. 2, bottom panel) obtained from a different batch, confirming the presence of the dielectric transitions reported here.
The data in Fig.1 can be analyzed in terms of three dielectric modes, $`\stackrel{~}{\epsilon }=\stackrel{~}{\epsilon }_\alpha +\stackrel{~}{\epsilon }_\beta +\stackrel{~}{\epsilon }_\gamma `$, each of which is well described by a Debye relaxation form with respect to the temperature dependence :
$$\stackrel{~}{\epsilon }=\stackrel{~}{\epsilon }_\alpha +\stackrel{~}{\epsilon }_\beta +\stackrel{~}{\epsilon }_\gamma =\underset{i=\alpha ,\beta ,\gamma }{}\frac{\epsilon _{i0}(T)}{1i\omega \tau _i(T)}$$
(3)
$`\stackrel{~}{\epsilon }_\gamma `$ appears to represent the low $`T`$ tail of a high temperature process, with $`\epsilon _{\gamma 0}(T)=160`$, and with a relaxation time $`\tau _\gamma (T)=6.5\times 10^{13}\mathrm{sec}^1\mathrm{exp}(1000/T)`$ characterized by an activation energy $`1000K`$. The $`\stackrel{~}{\epsilon }_\gamma `$ process is dominant between $`300K`$ and approximately $`180K`$, below which it “freezes” out quasistatically as the dipole relaxation rate becomes extremely slow. A residual temperature independent dielectric contribution $`465+i125`$ remains at all temperatures. We believe $`\stackrel{~}{\epsilon }_\gamma `$ is the contribution which has been measured by several previous investigators on non-metallic $`YBa_2CuO_{6.0}`$ and represents a polarization mode formed at high temperatures $`T>300K`$. $`\stackrel{~}{\epsilon }_\alpha `$ and $`\stackrel{~}{\epsilon }_\beta `$ indicate the onset of two new dielectric modes which turn on below transition temperatures $`T_{c\alpha }=60K`$ and $`T_{c\beta }=110K`$. We describe these modes with the following parameters :
* $`\epsilon _{\beta 0}(T)=60(1(T/T_{c\beta }))`$, $`T_{c\beta }=110K`$ and $`\tau _\beta (T)=4\times 10^{10}(\mathrm{sec}K)/(T+200)`$ , for the $`\stackrel{~}{\epsilon }_\beta `$ process, and
* $`\epsilon _{\alpha 0}(T)=280(1(T/T_{c\alpha }))`$, $`T_{c\alpha }=60K`$ and $`\tau _\alpha (T)=2.5\times 10^{10}(\mathrm{sec}K)/(T+5)`$, for the $`\stackrel{~}{\epsilon }_\alpha `$ process.
$`\epsilon _{\alpha 0}`$ and $`\epsilon _{\beta 0}`$ are similar to order parameters which grow at temperatures below a transition. As $`T`$ is lowered, both $`\epsilon _\alpha ^{}(T)`$ and $`\epsilon _\alpha ^{\prime \prime }(T)`$ increase initially due to the growing polarization. However below a characteristic temperature both $`\epsilon _\alpha ^{}(T)`$ and $`\epsilon _\alpha ^{\prime \prime }(T)`$ begin to decrease because the dipoles are no longer able to follow the microwave field. The peak temperature $`T_{p\alpha }`$ $`25K`$ is determined by the condition $`\omega \tau _\alpha (T_{p\alpha })=1`$ , although the peak for $`\epsilon ^{}`$ is at a higher $`T`$ than for $`\epsilon ^{\prime \prime }`$. The peaks are so-called dielectric loss peaks. Identical arguments hold for $`\stackrel{~}{\epsilon }_\beta (T)`$ also. Here the peak is much broader and occurs at $`T_{p\beta }75K`$.
For the $`\alpha `$ and $`\beta `$ processes, the temperature dependence is too broad to be described by an activated relaxation rate. We have found that a relaxation rate which is linear in $`T`$, i.e. $`\tau _{\alpha ,\beta }^1(T)a_{\alpha ,\beta }(T+T_{(\alpha ,\beta )0})`$, describes the data very well as seen in Fig. 1, with $`a_\alpha 0.4\times 10^{10}(sec.K)^1`$, $`T_{\alpha 0}=5K`$ and $`a_\beta 0.25\times 10^{10}(sec.K)^1`$ and $`T_{\beta 0}=200K`$. Such relaxation rates with linear $`T`$ dependences are well known in the copper oxide superconductors.
We note that similar large dielectric constants have been observed in other copper oxides. In $`Bi_2Sr_2(Dy,Y,Er)Cu_2O_8`$, the parent compound of $`Bi:2212`$, large in plane $`\epsilon ^{}10^310^5`$ were reported . It is also important to note that the dielectric modes discussed here bear a strong similarity to the numerous modes observed in the dielectric response of the perovskite $`SrTiO_3`$ at $`65K,37K`$ and $`16K`$. The dielectric loss peaks reported here are similar to those observed in $`La_{5/3}Sr_{1/3}NiO_4`$ and $`Sr_{14}Cu_{24}O_{41}`$.
The present results indicate that at $`110K`$ and $`60K`$, two new polarization onset transitions occur in $`YBa_2CuO_{6.0}`$. We note that the $`110K`$ and $`60K`$ onsets cannot arise from any contamination of the sample by a superconducting phase, since then the contribution should be diamagnetic (negative $`\epsilon ^{}`$), opposite to what is observed.
A scenario leading to such dielectric transitions can be arrived at starting with the so-called Bilz model for ferroelectricity, which is based upon the nonlinear polarizability of oxygen, and originally developed for perovskite structures. This applies to a displacive type ferroelectric where dipole moments are induced during the phase transition so that soft mode concept becomes important. Above the displacive ferroelectric transition, the oxygen atoms are essentially oscillating in a potential well. Below the ferroelectric transition, a double-well is formed and the oxygen atom then locks into one of the minima - the displacement then leads to a large permanent polarization. In the present case the ferroelectricity is prevented from occurring, either due to quantum fluctuations, as was proposed for $`SrTiO_3`$ , or due to coupling between Ba-O and Cu-O layers(Fig. 3) . Consequently the $`O`$ potential is greatly softened leading to the large dielectric permittivities observed.
We use a modification of the Bilz model specifically for the oxo-cuprate superconductors implemented by Shenoy, et. al. . The equation-of-motion of the Oxygen relative ion-electron coordinate $`\stackrel{}{w}`$ is given by $`m_e(\stackrel{_{}}{\stackrel{}{w}}+\mathrm{\Gamma }\stackrel{.}{\stackrel{}{w})}+D\stackrel{}{w}=Ze\stackrel{}{E}e^{i\omega t}`$, where $`D`$ is the SCPA curvature of the anharmonic Oxygen electronic potential. For a driving electric field $`\stackrel{}{E}e^{i\omega t}`$, the susceptibility $`\alpha =Zew/E=Ze/m_eD(\omega _o^2\omega ^2i\omega \mathrm{\Gamma })`$. The dielectric constant $`\epsilon =n\alpha /\epsilon _0`$ then becomes
$$\stackrel{~}{\epsilon }=\frac{\epsilon (0)}{(1\omega ^2/\omega _0^2)i\omega \tau }$$
(4)
Here $`\epsilon (0)=nZe^2/D`$ and $`\tau =\mathrm{\Gamma }/\omega _0^2`$.
Rather large dielectric constants are feasible for soft modes. For the case of the $`O`$ atom in the $`BaO`$ layer in $`YBa_2CuO_{6.0}`$ we have $`n=1.2\times 10^{28}m^3`$. With $`Z=1,\epsilon _0=8.85\times 10^{12}F/m`$, and using $`\omega _0=2\pi f_0`$, $`f_0=3\times 10^{13}Hz`$so that $`D=3.3\times 10^2N/m`$, we get $`\epsilon (0)10^3`$ comparable to the experimental results. Note that despite the softening, the condition $`\omega \omega _0`$ is well satisfied ($`f=10^{10}Hz`$), and the so-called Drude-Lorentz form of a resonant mode given by Eq. 3 reduces to the Debye- like relaxation forms used in Eq. 2. The above estimate indicates considerable softening of the anharmonic $`O`$ potential. Indeed this can happen because the curvature is extremely sensitive to interatomic forces in these materials. A mechanism for such softening has been given by Shenoy, et. al. for the layered HTS.
There have been extensive studies of lattice dynamics in superconducting $`YBa_2Cu_3O_{6+x}`$. One of the key features that has emerged is that there are small structural distortions which occur although there is no change in the overall structure. Particularly well-established are the structural distortions reported at $`T_c`$ in $`YBCO`$, $`Hg:1201`$ and $`Tl:2212`$. In $`YBa_2Cu_3O_{6+x}`$ the coupling between apical O(1) and planar O(2) oxygen changes at the superconducting transition due to the displacements of O(2) in the directions perpendicular to the CuO2 plane (Fig. 3) changing the buckling in the plane . The nearness of the $`\beta `$ dielectric mode around $`110K`$ to the $`T_c=93K`$ ($`YBa_2Cu_3O_{6.95}`$) is striking and suggests a possible change in the O(2) dynamics in $`YBa_2Cu_3O_{6.0}`$ as well, and hence we attribute this $`\beta `$ mode to the change in the dynamics of O(2).
In ref. the possibility of dielectric modes in coupled Ba-O and Cu-O layers is described. Including the various interatomic forces, it can be shown that the curvature $`D`$ becomes a function of the buckling angle $`\theta `$. The change in buckling at $`110K`$ would lead to a change in the mixing of the $`ab`$ plane acoustic and $`c`$-axis optic mode, resulting in a new set of mixed modes which would move to lower frequency due to softening. The resulting decrease in $`D`$ would then explain the $`\beta `$ mode at $`110K`$. The calcultion of Shenoy et al. based on the mean field approach support our description of the dynamics of O(1).
The presence of the $`60K`$ mode suggests another local structural change at this temperature scale, possibly in the chain layer. It is very interesting to note that this temperature scale is present in the doped YBCO as well. In addition to the proximity of the $`110K`$ transition to the optimum superconducting $`T_c`$ of $`93K`$ noted above, equally important is the presence of a secondary temperature scale around $`60K`$ in certain measurements of $`YBa_2Cu_3O_{6+x}`$. This temperature scale has manifested itself in various experiments whose connections are becoming apparent only recently. Thus the present results may have important implications for superconductivity in these materials.
In optimally doped single crystals of $`YBa_2Cu_3O_{6.95}`$ an additional onset of pair conductivity at $`65K`$ was noted in ref., well below the main $`T_c=93K`$. The consequences of this on the thermal conductivity and vortex transport have been observed. The present temperature scales also have striking resemblance to transitions around $`110K`$ and $`65K`$ in over doped YBCO observed in NQR measurements by Grevin et al. , which have been interpreted in terms of CDW correlations. In their results a short range CDW sets in the Cu-O chains at $`110K`$ which becomes long range around $`65K`$. The formation of CDW in the chains modulate the charge in the planes leading to a transition between an inhomogeneous charge state to a low-temperature ordered charge state in the planes. Keeping in view the fact that NQR probes the electric field gradient around the Cu nucleus the NQR transitions could as well be due to changes in local structure which leads to CDW formation in the doped metallic YBCO. Together, the NQR, thermal conductivity and the present results stress the importance of the $`110K`$ and $`60K`$ temperatures scales. These results indicate that local distortions in the structure at $`110K`$ and $`60K`$ would lead to changes in polarization as we have observed in $`YBa_2Cu_3O_{6.0}`$ and charge ordering in doped YBCO. Microwave conductivity measurements on doped crystals from $`x=0`$ to $`1`$, when taken together with these other observations, strongly suggest that doping moves this onset temperature from $`60K`$ at $`x=0`$ to $`70K`$ at $`x=1`$. At the superconducting $`T_c`$ increases with $`x`$ , but never exceeds $`93K`$. An interesting implication is that the implied structural distortion transition at $`110K`$ may represent an upper limit on the superconducting transition $`T_c`$.
In conclusion we have observed three paraelectric modes with temperature onsets at $`60K`$, $`110K`$ and $`>300K`$ in the non-metallic $`YBa_2Cu_3O_{6.0}.`$ These modes are well described by the change in polarizability of apical $`O(1)`$ oxygens due to change in lattice dynamics with temperature. A striking observation is that the two low temperature modes have direct connections with the transitions observed by traditional lattice probes as well as NQR and thermal conductivity measurements on doped superconducting $`YBa_2Cu_3O_{6.0+x}`$, strongly suggesting that oxygen and lattice dynamics plays an important role in both the superconducting and non-superconducting materials.
We thank C.Kusko, R.S.Markiewicz, C. Perry and S.R.Shenoy for useful discussions.
This work was supported by ONR and NSF-ECS-9711910.
|
warning/0005/hep-th0005208.html
|
ar5iv
|
text
|
# 1 Introduction and Conclusions
## 1 Introduction and Conclusions
Non-commutative gauge field theories lately attracted a lot of attention, mainly due to the discoveries of their relation to string theory . It was also found that the perturbative structure of these theories has an interesting pattern. It was shown , in the case of scalar theory, that planar diagrams of the non-commutative theory are the same as planar diagrams of ordinary commutative theory, up to global phases. For an earlier related work see ref.. It was then suggested that non-planar graphs are UV finite, due to the oscillatory Moyal phase which regulates the integrals. It was found later that these contributions actually lead to divergences, which were interpreted as infra-red divergences. These contributions are singular in the $`\theta 0`$ limit and they occur also in gauge theories .
This paper is devoted to the study of $`U(N)`$ non-commutative gauge theory. The $`U(1)`$ case was already studied by several authors . The renormalization of the model, at the one loop approximation, was studied first in . The UV/IR mixing, in the $`U(1)`$ case was studied by and . Related works about perturbative dynamics of non-commutative field theories are . Perturbative aspects of non-commutative field theories from string theory were discussed in .
The non-commutative $`U(N)`$ Yang-Mills action is
$$d^4x\mathrm{tr}\frac{1}{2g^2}F_{\mu \nu }F^{\mu \nu }$$
(1)
where $`F_{\mu \nu }`$ is
$$F_{\mu \nu }=_\mu A_\nu _\nu A_\mu i(A_\mu A_\nu A_\nu A_\mu )$$
(2)
and $`A_\mu `$ is a $`N\times N`$ matrix. The $``$-product between two functions $`f`$ and $`g`$ is defined by
$$fg(x)=e^{\frac{i}{2}\theta ^{\mu \nu }_\mu ^{(\xi )}_\nu ^{(\eta )}}f(x+\xi )g(x+\eta )|_{\xi ,\eta 0}.$$
(3)
The action (1) is invariant under $`U(N)`$ gauge transformation
$$\delta _\lambda A_\mu =_\mu \lambda i(A_\mu \lambda \lambda A_\mu ).$$
(4)
The gauge transformation (4) is different from the commutative gauge transformation in the sense that it mixes the $`U(1)`$ gauge boson with the $`SU(N)`$ gauge bosons. In fact, the non-commutative Yang-Mills action (1) also mixes the $`U(1)`$ and the $`SU(N)^{}s`$. It cannot be written as a sum of a $`SU(N)`$ and a $`U(1)`$ theories as the ordinary YM theory, since there are interaction terms between the $`SU(N)`$ gluons and the $`U(1)`$ ’photon’. In order to demonstrate this point we list in figure 1 below the Feynman rules which describe the 3-gluons interaction (the full list of Feynman rules is written in Appendix A)
The complicated structure of the action (1) raises the question of gauge invariance and consistency of the non-commutative Yang-Mills theory at the quantum level. The action (1) consists of many interaction terms with a single coupling $`g`$ \- due to gauge invariance. It is not clear, a-priori, that the relations among the various couplings in the action is kept at the quantum level. There are two limits of the theory which hint that the full $`U(N)`$ gauge invariance might be broken. The first limit is $`\theta 0`$. In this case, the theory is expected to reduce to the ordinary commutative theory. However, the commutative theory has a $`SU(N)\times U(1)`$ gauge symmetry and the $`SU(N)`$ coupling is not related to the $`U(1)`$ coupling by gauge symmetry. Moreover, at the quantum level the $`SU(N)`$ coupling runs and the $`U(1)`$ coupling is kept fixed. The second limit, is the planar limit. Since it looks as if the non-commutative theory and the commutative are identical at the planar level, the same question about the $`U(1)`$ coupling should be raised.
As we shall see the $`U(N)`$ gauge symmetry is not broken quantum mechanically. The renormalization procedure does not violate the relations between the various couplings (at least at the one loop level). The resolution of the puzzles mentioned above, is the following: the limit $`\theta 0`$ does not lead to the ordinary commutative theory. Though the resulting action looks like the ordinary YM action (note that the $`U(1)`$ and the $`SU(N)`$ seems to decouple in the $`\theta 0`$ limit, see figure 1b), the $`U(1)`$ and $`SU(N)`$ couplings have exactly the same beta function.
The fact that the limit $`\theta 0`$ of the $`U(1)`$ theory is singular was already pointed out in . It was with relation to the non-planar contributions, which are manifestly singular in $`\theta `$. We claim, however, that the theory is not smooth in $`\theta `$ even in the planar limit. In this case, indeed the $`SU(N)`$ sector theory looks like the commutative theory, but the interaction of the $`U(1)`$ with the $`U(N)^{}s`$ survives the limit. In particular, the $`U(1)`$ gauge coupling runs. Thus, the planar sector of the $`\theta 0`$ theory does not correspond to the planar sector of the commutative theory. In this way, the puzzle about $`U(N)`$ gauge invariance at the quantum level is also resolved.
The main results of the paper are the following: in section 2 we calculate the counter terms which are needed to regulate the divergences in the planar graphs of the $`SU(N)`$ and $`U(1)`$ gluons propagators. We find that they are the same and equal to the ordinary commutative counter term of the $`SU(N)`$ propagator. The non-planar contributions, however, are different. There is a non-planar finite contribution to the $`U(1)`$ propagator, but there is no such contribution for the $`SU(N)`$. In section 3 we calculate the counter terms of the various 3 gluons vertices. Our results in this section are similar to those of section 2. The divergent (planar) part of the various 3 gluons vertices is the same, but the finite (non-planar) part is different. Finally, in section 4, we calculate the beta function, discuss our results and more general cases where also matter fields are present.
We shall use the following conventions: capital letters $`(A,B,C,\mathrm{})`$ denote $`U(N)`$ indices, small letters $`(a,b,c,\mathrm{})`$ denote $`SU(N)`$ indices. The $`U(1)`$ generator is normalized as follows $`t^0=\frac{1}{\sqrt{2N}}`$, such that $`\mathrm{tr}t^At^B=\frac{1}{2}\delta ^{AB}`$. Finally, $`[t^a,t^b]=if^{abc}t^c`$ and $`\{t^a,t^b\}=\frac{1}{N}\delta ^{ab}+d^{abc}t^c`$. Thus $`d^{abc}`$ represents the symmetric tensor for the fundamental representation. In addition, we shall use the notation $`\stackrel{~}{p}_\mu =\frac{1}{2}\theta _{\mu \nu }p^\nu `$.
## 2 Corrections to the gluon propagator
In order to check the renormalizability and gauge invariance at the quantum level, let us start with the one loop correction to the gluon propagator. The various contribution are drawn in figure 2 below
We consider first the case where the external legs carry $`U(1)`$ indices (’photons’). The calculation is a straightforward generalization of the calculations which were performed for the $`U(1)`$ non-commutative Yang-Mills theory. The three contributes are drawn in figure 2. Let us focus on diagram 2a. The only difference in comparison with the $`U(1)`$ theory is that now all the $`U(N)`$ gluons can circulate in the loop. We will use the Feynman rules 1b and 1c. We denote the external momentum by $`p`$ and the internal momentum by $`q`$. The resulting expression is
$`A^{\mu \nu }`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle }{\displaystyle \frac{d^4q}{(2\pi )^4}}{\displaystyle \frac{i}{q^2}}{\displaystyle \frac{i}{(p+q)^2}}\times T_1\times `$ (5)
$`(g^{\mu \rho }(pq)^\sigma +g^{\rho \sigma }(2q+p)^\mu +g^{\sigma \mu }(q2p)^\rho )\times `$
$`(\delta _\rho ^\nu (qp)_\sigma +g_{\rho \sigma }(2qp)^\nu +\delta _\sigma ^\nu (q+2p)_\rho ),`$
$`T_1`$ $`=`$ $`{\displaystyle \frac{2g^2}{N}}\delta ^{AB}\delta ^{AB}\mathrm{sin}^2\stackrel{~}{p}q`$ (6)
By using the identity
$$\mathrm{sin}^2\stackrel{~}{p}q=\frac{1}{2}(1\mathrm{cos}2\stackrel{~}{p}q)$$
(7)
we can isolate the planar contribution which comes from the $`\frac{1}{2}`$, from the non-planar contribution (the cosine)
$$T_1=g^2N+\text{non-planar term.}$$
(8)
The planar part of the contribution is divergent and its value is exactly the same as the value of the divergent part of the gluon propagator in ordinary $`SU(N)`$ Yang-Mills theory. The same pattern occurs in the other diagrams in figure 2. Indeed, after the summation of the three diagrams in figure 2 we find that in order to cancel the divergent part of the $`U(1)`$ propagator the following counter term is needed
$$\delta _3^{(11)}=\frac{g^2N}{(4\pi )^2}\times \frac{5}{3}\times \frac{2}{ϵ},$$
(9)
where dimensional regularization was used and $`ϵ=4d`$. Note that the counter term does not depend on $`\theta `$. As long as $`\theta `$ is non zero, a counter term (9) is needed. Otherwise the $`U(1)`$ theory is free. Therefore, though the Feynman rules of the theory are smooth in $`\theta `$, the limit $`\theta 0`$ is singular.
For completeness let us quote the result for the finite part of the correction to the $`U(1)`$ propagator . It is calculated by replacing the cosine of (7) by an exponent and by looking at the high momentum regime in the integrals of (5) and the two other diagrams in figure 2. The result is
$$A_{finite}^{\mu \nu }=2g^2N\frac{d^4q}{(2\pi )^4}\frac{2q^\mu q^\nu g^{\mu \nu }q^2}{q^4}e^{i2\stackrel{~}{p}q}g^2N\frac{\stackrel{~}{p}^\mu \stackrel{~}{p}^\nu }{\stackrel{~}{p}^4}.$$
(10)
Note that this term is singular in $`\theta `$. When inserted into higher loops it behaves as ordinary infra-red divergences . Thus, an effect which was originally due to high momentum turns out to be an IR effect. This is the UV/IR mixing which was found in .
Let us turn now to the calculation of the correction to the $`SU(N)`$ part of the gluon. Again, let us start with diagram 2a. The coupling of the $`SU(N)`$ bosons which circulate in the loop contains the symmetric tensor $`d^{abc}`$. The integral is the same as (5), but $`T_1`$ is replaced by $`T_2`$
$$T_2=g^2(f^{xya}\mathrm{cos}\stackrel{~}{p}q+d^{xya}\mathrm{sin}\stackrel{~}{p}q)(f^{xyb}\mathrm{cos}\stackrel{~}{p}q+d^{xyb}\mathrm{sin}\stackrel{~}{p}q)$$
(11)
By using the $`SU(N)`$ identities (see Appendix B)
$`f^{axy}f^{bxy}=N\delta ^{ab}`$ (12)
$`d^{axy}d^{bxy}=(N{\displaystyle \frac{4}{N}})\delta ^{ab}`$ (13)
$`T_2`$ can be written as
$$T_2=g^2N(1\frac{4}{N^2}\mathrm{sin}^2\stackrel{~}{p}q).$$
(14)
Interestingly there is another contribution to the gluon propagator which doesn’t occur in ordinary Yang-Mills theory, due to the existence of new vertices (fig. 1b). It is possible to exchange a $`U(1)`$ boson in half of the loop and $`SU(N)`$ boson in the other half. It contributes
$$T_2^{}=2\times g^2\frac{2}{N}\mathrm{sin}^2\stackrel{~}{p}q$$
(15)
(the factor 2 in (15) represents two possible exchanges of the $`U(1)`$). Collecting the two terms (14) and (15), we find that the one loop correction to the $`SU(N)`$ propagator is exactly the same as in the commutative case. Thus the divergences can be compensated by the commutative counter term
$$\delta _3^{(NN)}=\frac{g^2N}{(4\pi )^2}\times \frac{5}{3}\times \frac{2}{ϵ}.$$
(16)
Remarkably (9) is identical to (16) (except that (16) in needed also when $`\theta =0`$, in contrast to (9)). This fact is crucial to ensure gauge invariance of the model at the quantum level, as we shall see later.
It is interesting to note that the corrections to the $`SU(N)`$ propagator does not contain a non-planar finite part. Therefore, though the propagators are identical in their divergent part, they differ in their finite part. It is due to non-planar graphs which exist in the corrections to the $`U(1)`$ propagator but do not exist for the $`SU(N)`$ one.
## 3 Corrections to the 3-gluons vertex
In this section we calculate the one-loop corrections to the 3 gluons vertex. The relevant diagrams are listed in figure 3 below. The external momenta are $`p_1,p_2,p_3`$ with $`p_1+p_2+p_3=0`$.
We begin with the simplest case in which all external legs are $`U(1)^{}s`$. We focus on diagram 3a. The gluons which circulate in the loop belongs to $`U(N)`$. Similar calculations were made in refs..
$`M^{\mu \nu \rho }`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle }{\displaystyle \frac{d^4q}{(2\pi )^4}}{\displaystyle \frac{i}{q^2}}{\displaystyle \frac{i}{(qp_2)^2}}{\displaystyle \frac{i}{(q+p_1)^2}}\times V_1\times `$
$`(g^{\sigma _1\mu }(qp_1)^{\sigma _2}+g^{\mu \sigma _2}(2p_1+q)^{\sigma _1}+g^{\sigma _2\sigma _1}(p_12q)^\mu )\times `$
$`(\delta _{\sigma _1}^\nu (p_2+q)^{\sigma _3}+\delta _{\sigma _1}^{\sigma _3}(2q+p_2)^\nu +g^{\sigma _3\nu }(q2p_2)_{\sigma _1})\times `$
$`(g_{\sigma _3\sigma _2}(2qp_1+p_2)^\rho +\delta _{\sigma _2}^\rho (p_1+qp_3)_{\sigma _3}+\delta _{\sigma _3}^\rho (p_3+p_2q)_{\sigma _2}),`$
$`V_1`$ $`=`$ $`g^3({\displaystyle \frac{2}{N}})^{\frac{3}{2}}\delta ^{XY}\delta ^{YZ}\delta ^{ZX}\mathrm{sin}\stackrel{~}{p}_1q\mathrm{sin}\stackrel{~}{p}_2q\mathrm{sin}\stackrel{~}{p}_3(q+p_1)`$ (18)
By using the following identity
$`\mathrm{sin}\stackrel{~}{p}_1q\mathrm{sin}\stackrel{~}{p}_2q\mathrm{sin}\stackrel{~}{p}_3(q+p_1)=`$
$`{\displaystyle \frac{1}{4}}\mathrm{cos}\stackrel{~}{p}_3p_1(\mathrm{sin}2\stackrel{~}{p}_1q+\mathrm{sin}2\stackrel{~}{p}_2q+\mathrm{sin}2\stackrel{~}{p}_3q)`$
$`{\displaystyle \frac{1}{4}}\mathrm{sin}\stackrel{~}{p}_3p_1(1\mathrm{cos}2\stackrel{~}{p}_1q\mathrm{cos}2\stackrel{~}{p}_2q+\mathrm{cos}2\stackrel{~}{p}_3q)`$ (19)
we can isolate the divergent parts from the finite parts. The divergent part comes from the $`\frac{1}{4}\mathrm{sin}\stackrel{~}{p}_3p_1`$ contribution. Since this part does not depend on $`q`$ it would lead to a contribution which is similar to the commutative case. The other diagrams in figure 3 are similar. Thus the needed counter term to cancel the divergent part of the $`U(1)U(1)U(1)`$ vertex is
$$\delta _1^{(111)}=\frac{2g^2N}{(4\pi )^2}\times \frac{1}{4}\times \frac{4}{3}\times \frac{2}{ϵ}=\frac{g^2N}{(4\pi )^2}\times \frac{2}{3}\times \frac{2}{ϵ}$$
(20)
The finite part of the correction, which arise from the $`\frac{1}{4}\mathrm{cos}\stackrel{~}{p}_3p_1\times \mathrm{sin}2\stackrel{~}{p}_iq`$ terms in (19), is calculated by replacing the sin by an exponent. The procedure is exactly the same as in the $`U(1)`$ case . The result is
$`M_{finite}^{\mu \nu \rho }=`$ (21)
$`g^3\sqrt{2N}\mathrm{cos}\stackrel{~}{p}_3p_1{\displaystyle }{\displaystyle \frac{d^4q}{(2\pi )^4}}{\displaystyle \frac{1}{q^6}}(4q^\mu q^\nu q^\rho q^2(q^\mu g^{\nu \rho }+q^\nu g^{\mu \rho }+q^\rho g^{\mu \nu }))\times `$
$`(e^{i2\stackrel{~}{p}_1q}+e^{i2\stackrel{~}{p}_2q}+e^{i2\stackrel{~}{p}_3q})`$
$`g^3\sqrt{N}\mathrm{cos}\stackrel{~}{p}_3p_1\left({\displaystyle \frac{\stackrel{~}{p}_1^\mu \stackrel{~}{p}_1^\nu \stackrel{~}{p}_1^\rho }{\stackrel{~}{p}_1^4}}+{\displaystyle \frac{\stackrel{~}{p}_2^\mu \stackrel{~}{p}_2^\nu \stackrel{~}{p}_2^\rho }{\stackrel{~}{p}_2^4}}+{\displaystyle \frac{\stackrel{~}{p}_3^\mu \stackrel{~}{p}_3^\nu \stackrel{~}{p}_3^\rho }{\stackrel{~}{p}_3^4}}\right)`$
The calculation of the correction to the 3-gluons vertex when the external legs are in $`SU(N)`$, is a bit more complicated, as there are many contributions in the non-commutative case. The first contribution is when $`SU(N)`$ gluons circulate in the triangle of figure 3a. We will use the Feynman rules in figure 1a. The calculation of the diagram is performed by replacing $`V_1`$ by
$`V_2`$ $`=`$ $`g^3(f^{axy}\mathrm{cos}\stackrel{~}{p}_1q+d^{axy}\mathrm{sin}\stackrel{~}{p}_1q)\times `$ (22)
$`(f^{byz}\mathrm{cos}\stackrel{~}{p}_2q+d^{byz}\mathrm{sin}\stackrel{~}{p}_2q)\times `$
$`(f^{czx}\mathrm{cos}\stackrel{~}{p}_3(q+p_1)+d^{czx}\mathrm{sin}\stackrel{~}{p}_3(q+p_1))`$
In order to simplify (22) we use the following $`SU(N)`$ identities (see Appendix B for derivation)
$`f^{axy}f^{byz}f^{czx}f^{axy}d^{byz}d^{czx}`$
$`d^{axy}f^{byz}d^{czx}d^{axy}d^{byz}f^{czx}=2N(1{\displaystyle \frac{3}{N^2}})f^{abc},`$ (23)
$`d^{axy}d^{byz}d^{czx}d^{axy}f^{byz}f^{czx}`$
$`f^{axy}d^{byz}f^{czx}f^{axy}f^{byz}d^{czx}=2N(1{\displaystyle \frac{3}{N^2}})d^{abc},`$ (24)
and trigonometric identities similar to (7). Hence, $`V_2`$ can be written as follows
$$V_2=g^3(f^{abc}\mathrm{cos}\stackrel{~}{p}_3p_1+d^{abc}\mathrm{sin}\stackrel{~}{p}_3p_1)\frac{N}{2}(1\frac{3}{N^2})+\text{other terms},$$
(25)
where ’other terms’ means additional contributions which do not lead to divergences. Apart from the $`V_2`$ contribution, there is another contribution to the $`SU(N)SU(N)SU(N)`$ vertex. It is due to $`SU(N)`$ bosons flowing in two of the sides of the triangle in figure 3a and a $`U(1)`$ boson in the third side. The contribution is
$$V_2^{}=3\times g^3\frac{2}{N}(f^{axy}\mathrm{cos}\stackrel{~}{p}_1q+d^{axy}\mathrm{sin}\stackrel{~}{p}_1q)\times \delta ^{xb}\mathrm{sin}\stackrel{~}{p}_2q\times \delta ^{yc}\mathrm{sin}\stackrel{~}{p}_3(q+p_1)$$
(26)
The part that leads to divergences in (26) can be written as follows
$$V_2^{}=g^3(f^{abc}\mathrm{cos}\stackrel{~}{p}_3p_1+d^{abc}\mathrm{sin}\stackrel{~}{p}_3p_1)\frac{N}{2}\frac{3}{N^2}$$
(27)
Thus, collecting the two contributions $`V_2`$ and $`V_2^{}`$, we find that the counter term which is needed to cancel the divergences in the 3 gluons vertex with external legs in $`SU(N)`$ is
$$\delta _1^{(NNN)}=\frac{g^2N}{(4\pi )^2}\times \frac{2}{3}\times \frac{2}{ϵ},$$
(28)
as in ordinary commutative Yang-Mills theory. Note that the interaction with the $`U(1)^{}s`$ was needed to cancel the $`\frac{1}{N^2}`$ terms in (22). Another comment is that the finite contribution (21) in the $`U(1)U(1)U(1)`$ cancels in the present case.
We turn now to the renormalization of 3-gluons vertex with one external leg in $`U(1)`$ and two external legs in $`SU(N)`$ (figure 1b). The first contribution to the diagram 3a is when $`SU(N)`$ bosons circulate in the loop. We should use the Feynman rules 1a and 1b. The contribution is
$`V_3=g^3\sqrt{{\displaystyle \frac{2}{N}}}(f^{axy}\mathrm{cos}\stackrel{~}{p}_1q+d^{axy}\mathrm{sin}\stackrel{~}{p}_1q)\times `$
$`(f^{byz}\mathrm{cos}\stackrel{~}{p}_2q+d^{byz}\mathrm{sin}\stackrel{~}{p}_2q)\times \delta ^{zx}\mathrm{sin}\stackrel{~}{p}_3(q+p_1)`$ (29)
which can be simplified (by using (12), (13) and (19)) and rewritten as
$$V_3=g^3\sqrt{\frac{2}{N}}N(2\frac{4}{N^2})\frac{1}{4}\mathrm{sin}\stackrel{~}{p}_3p_1\delta ^{ab}+\text{other terms}$$
(30)
In addition there are two other diagrams which correct the $`SU(N)SU(N)U(1)`$ vertex. In one of the diagrams there are two $`U(1)`$ bosons and one $`SU(N)`$ bosons which flow in the triangle (fig. 3a) and in the other there are two $`SU(N)`$ bosons and one $`U(1)`$. The two diagrams contributes the same. Their contribution is
$`V_3^{}`$ $`=`$ $`2\times g^3({\displaystyle \frac{2}{N}})^{\frac{3}{2}}\mathrm{sin}\stackrel{~}{p}_1q\mathrm{sin}\stackrel{~}{p}_2q\mathrm{sin}\stackrel{~}{p}_3(q+p_1)\delta ^{ab}`$ (31)
$`=`$ $`2\times g^3\sqrt{{\displaystyle \frac{2}{N}}}{\displaystyle \frac{2}{N}}{\displaystyle \frac{1}{4}}\mathrm{sin}\stackrel{~}{p}_3p_1\delta ^{ab}+\text{other terms}`$
The contribution $`V_3^{}`$ exactly compensate the $`\frac{1}{N^2}`$ part in (30). Hence the needed counter term is
$$\delta _1^{(NN1)}=\frac{g^2N}{(4\pi )^2}\times \frac{2}{3}\times \frac{2}{ϵ},$$
(32)
exactly as (20) and (28).
The ’other terms’ in eqs.(30),(31) leads to finite terms which take exactly the same form as (21).
The calculation of the 4-gluon vertices is straightforward, though tedious. Adding matter in the adjoint representation is also straightforward. The counter terms which are needed in all these cases are exactly the same as the ones which are needed in ordinary $`SU(N)`$ theory.
## 4 Renormalizability and gauge invariance
In the previous sections we calculated the counter terms which are needed to renormalize the theory. Since we are dealing with a gauge theory, gauge symmetry imposes some constraints on the various counter terms. In ordinary Yang-Mills theory the three gluons vertex and the four gluons vertex are multiplied by $`g`$ and $`g^2`$ respectively. Gauge invariance tells us that the two couplings should be the same - also at the quantum level. In the present case the situation is even more involved. A-priori, there are two types of propagators with different wave functions renormalization. There are also three types of vertices (even four, if we consider the $`f^{abc}\mathrm{cos}`$ and the $`d^{abc}\mathrm{sin}`$ parts of the $`SU(N)`$ vertex as two independent vertices).
Gauge invariance imposes the following relations, at one loop
$$\delta _1^{(111)}\frac{3}{2}\delta _3^{(11)}=\delta _1^{(NNN)}\frac{3}{2}\delta _3^{(NN)}=\delta _1^{(NN1)}\delta _3^{(NN)}\frac{1}{2}\delta _3^{(11)}.$$
(33)
We have found that in fact all $`\delta _1^i`$ are equal and $`\delta _3^i`$ are equal. Clearly, (33) is satisfied.
The calculation of the beta function is also straightforward. The $`\frac{2}{ϵ}`$ in the expressions for $`\delta `$ should be replaced by $`\mathrm{log}\frac{\mathrm{\Lambda }^2}{\mu ^2}`$ and the beta function is computed by
$$\beta (g)=g\mu \frac{}{\mu }(\delta _1+\frac{3}{2}\delta _3).$$
(34)
The result is
$$\beta (g)=\frac{g^3}{(4\pi )^2}\frac{11}{3}N,$$
(35)
as expected. Note that our result for the $`U(1)`$ case differs by a factor of $`2`$ from due to a different definition of the $`U(1)`$ coupling.
Let us comment about various limits and some special cases. In contrast to the commutative theory, where the $`U(N)`$ theory contains two couplings: a $`U(1)`$ coupling which doesn’t run and an asymptotically free $`SU(N)`$ coupling, we showed that non-commutative theory can (and as we shall see in a moment - must) contain a single coupling. The theory is asymptotically free and the value of the beta function is independent of $`\theta `$. It is a bit unusual at first sight, since the commutative theory should be a limit of the non-commutative theory. However, this limit is singular. As long as $`\theta `$ is non-zero the $`U(1)`$ coupling runs, independently of the value of $`\theta `$ and exactly as the $`SU(N)`$ coupling. When $`\theta `$ is zero, the $`U(1)`$ is frozen. Thus though the Feynman rules of the non-commutative theory are smooth in $`\theta `$, the renormalization procedure makes the limit singular.
The planar limit is also interesting. It was suggested that the planar limit of the non-commutative theory is the ordinary theory. However, the planar limit of the $`U(1)`$ theory is not the ordinary commutative theory but rather an interacting theory and the counter term (9) is still needed. Similarly, the planar limit of the $`U(N)`$ theory is not the planar limit of the $`SU(N)\times U(1)`$ ordinary theory. In order to be more precise, let us give an example which clarifies the difference between the commutative and the non-commutative planar theories. Correlation functions which involve only $`\mathrm{tr}F_{\mu \nu }`$ would yield trivial answers in the commutative theory, since the $`U(1)`$ part is decoupled and free. On the other hand, such correlation functions are highly non-trivial in the planar non-commutative case.
Another remark is about the $`SU(N)`$ theory. It was argued in (see also and for a derivation from string theory) that the non-commutative version of this theory is not consistent, since the closure of the Moyal commutator is violated. Here we find another evidence for the inconsistency of the $`SU(N)`$ theory. Had we ignored the $`U(1)`$ part of the theory, we would have found that the value of the beta function is gauge dependent. In order to see that one should calculate the counter terms in a general gauge and to observe that a $`\frac{1}{N^2}`$ gauge dependent piece is left in the beta function of the $`SU(N)`$ theory.
Finally let us comment about the $`𝒩=4`$ theory. Since this theory is finite in the ordinary commutative case, it seems that for this specific case the limit $`\theta 0`$ is smooth. Let us focus on the planar theory first. Both the $`U(1)`$ and the $`SU(N)`$ gauge couplings take their classical value and no counter terms are needed. Therefore the $`\theta 0`$ limit is the same as the $`\theta =0`$ theory. The non-planar sector of the theory is more subtle. The finite UV effects are manifestly singular in $`\theta `$. It was suggested in , that in the specific case of $`𝒩=4`$ these contributions cancel and thus also this sector of the theory is smooth in $`\theta `$.
We would like to note that there is another class of theories which are UV finite and maybe even smooth in $`\theta `$. These are the orbifold truncations of $`𝒩=4`$. These theories share the same planar diagrams as $`𝒩=4`$ . Therefore, this sector of the theory is finite. The non-planar sector is anyways UV finite in non-commutative theories. Moreover, since these theories admit Bose-Fermi degeneracy, it is likely that the non-planar contributions cancel as in the $`𝒩=4`$ case.
ACKNOWLEDGEMENTS
I would like to thank C. Angelantonj, I. Antoniadis, E. Gardi, F. Hassan and R. Minasian for discussions and comments. This research was supported in part by EEC under TMR contract ERBFMRX-CT96-0090.
## 5 Appendix A - Feynman rules for the non-commutative $`U(N)`$ Yang-Mills theory
The non-commutative Yang-Mills action including gauge fixing and ghosts takes the following form
$$S=d^4x\mathrm{tr}\left(\frac{1}{2}F^{\mu \nu }F_{\mu \nu }+\xi (^\mu A_\mu )^2\overline{c}^\mu D_\mu c+^\mu D_\mu c\overline{c}\right)$$
(36)
We use the Feynman-’t Hooft gauge $`\xi =1`$.
## 6 Appendix B - $`SU(N)`$ identities
In this section we derive $`SU(N)`$ identities which are used in the paper.
Identity A $`f^{axy}f^{bxy}=N\delta ^{ab}`$.
We denote the adjoint representation by capital letters. We use $`\mathrm{tr}T^aT^b=N\delta ^{ab}`$. Also $`T_{xy}^a=if^{axy}`$. Therefore $`T_{xy}^aT_{yx}^b=if^{axy}\times if^{byx}=f^{axy}f^{bxy}=N\delta ^{ab}`$.
Identity B $`d^{axy}d^{bxy}=(N\frac{4}{N})\delta ^{ab}`$.
$$\mathrm{tr}t^xt^xt^at^b=\frac{N^21}{2N}\frac{1}{2}\delta ^{ab}.$$
(37)
By using $`t^at^b=\frac{1}{2}(if^{abc}t^c+\frac{1}{N}\delta ^{ab}+d^{abc}t^c)`$, (37) reads
$`=\mathrm{tr}{\displaystyle \frac{1}{2}}(if^{xay}t^y+{\displaystyle \frac{1}{N}}\delta ^{xa}+d^{xay}t^y)t^bt^x`$
$`={\displaystyle \frac{1}{2}}(if^{xay}+d^{xay})\mathrm{tr}t^yt^bt^x+{\displaystyle \frac{1}{4}}{\displaystyle \frac{1}{N}}\delta ^{xa}\delta ^{bx}`$
$`={\displaystyle \frac{1}{8}}(f^{axy}f^{bxy}+d^{axy}d^{bxy})+{\displaystyle \frac{1}{4N}}\delta ^{ab}`$ (38)
Thus (37) and (38) leads to
$$f^{axy}f^{bxy}+d^{axy}d^{bxy}=(2N\frac{4}{N})\delta ^{ab},$$
(39)
and by using identity A, identity B is proven.
Identity C
$`f^{axy}f^{byz}f^{czx}f^{axy}d^{byz}d^{czx}`$
$`d^{axy}f^{byz}d^{czx}d^{axy}d^{byz}f^{czx}=2N(1{\displaystyle \frac{3}{N^2}})f^{abc}`$ (40)
$`d^{axy}d^{byz}d^{czx}d^{axy}f^{byz}f^{czx}`$
$`f^{axy}d^{byz}f^{czx}f^{axy}f^{byz}d^{czx}=2N(1{\displaystyle \frac{3}{N^2}})d^{abc}`$ (41)
we begin with
$$\mathrm{tr}t^xt^xt^at^bt^c=\frac{N^21}{2N}\frac{1}{4}(if^{abc}+d^{abc}).$$
(42)
Eq. (42) can be written also as follows
$`=\mathrm{tr}{\displaystyle \frac{1}{2}}(if^{xay}t^y+{\displaystyle \frac{1}{N}}\delta ^{xa}+d^{xay}t^y)t^bt^ct^x`$
$`={\displaystyle \frac{1}{2}}(if^{xay}+d^{xay})\mathrm{tr}t^yt^bt^ct^x+{\displaystyle \frac{1}{2N}}\mathrm{tr}t^bt^ct^a`$
$`={\displaystyle \frac{1}{16}}(if^{xay}+d^{xay})(if^{ybz}+d^{ybz})(if^{zcx}+d^{zcx})+{\displaystyle \frac{2}{8N}}(if^{abc}+d^{abc}).`$ (43)
By equating (42) and (43) we arrive at
$$(if^{axy}+d^{axy})(if^{byz}+d^{byz})(if^{czx}+d^{czx})=(2N\frac{6}{N})(if^{abc}+d^{abc}).$$
(44)
The real and imaginary parts of (44) prove identity C.
|
warning/0005/cond-mat0005216.html
|
ar5iv
|
text
|
# Multi-Choice Minority Game
## I Introduction
Considerable progress in the theoretical understanding of market phenomena has been achieved by the study of the minority game. This prototypical model describes a system of agents interacting through a market mechanism . The game is based on the idea that the behavior of the agents is determined by the economic rule of supply and demand. According to this rule, given the available options (such as buy/sell), an agent wins if he chooses the minority action. The research of this game has been focused on cases in which each agent can choose between two options using its most efficient strategy, where the strategies remain unchanged throughout the game . However, in the real world, many situations of interest involve more than two decision options as well as agents with dynamic strategies. Making decisions like where to spend the summer vacation or which server to choose while surfing the web (or more generally, how to distribute data traffic in computer networks ) are only two among many common problems with more than two options. Therefore, it is tempting to investigate cases with more than two possible choices provided to agents with dynamic strategies. In a recent study of an extension in which each agent is equipped with a neural network for making his decision it was shown that a certain updating rule of the strategies of the agents improves the efficiency of the market, which is measured by the global profit of the agents. In this paper we generalize the aforementioned work to a multi-choice minority game, namely a game with general $`K`$ decision states.
The multi-choice minority game consists of $`N`$ players (agents) and $`K`$ possible decisions. In each step, each one of the players chooses one of the $`K`$ states, aiming to choose the state with the smallest number of agents. For example, a situation may arise, in which there are several possible roads which lead from place A to place B, and each driver who wants to get from A to B chooses one of the available roads. Because drivers want to avoid traffic jams, they try to choose the least traveled roads, assuming that all the roads are of the same length. Similarly, one usually prefers to go to the bar with the smallest number of people in it. Occurring over and over again, the minority decisions in these and other similar situations generate time series whose term at time $`t`$, $`x_t`$, has an integer value between $`1`$ and $`K`$ according to the minority decision. In the original game, the information provided to each player is the history vector of size $`M`$, whose components are the last $`M`$ minority states.
The paper is organized as follows: In section II a multi-layer neural network and the dynamic evolution of its weights are introduced. For the clarity of the rest of the paper which is somewhat technical we briefly discuss the main findings and results. In section III the reference case of players with random strategies is solved analytically. In section IV the global profit of the players for the network with optimal strategies (weights) is solved analytically in the thermodynamic limit, and shown to be superior to a random decision. The analytical results are compared with simulations on finite systems. In section V, the suggested updating rules for the weights are examined analytically and are found to saturate asymptotically the optimal global profit. Finally, section VI is devoted to a short summary and an outlook.
## II The model
While many strategies for the multi-choices minority game are conceivable, we study the following model which uses neural networks: each one of the $`N`$ players is represented by a perceptron of a size $`M`$. The weights belonging to the $`i`$th player are $`\{w_{ij}\}`$ where $`j=1,\mathrm{},M`$. All $`N`$ perceptrons have a common input which consists of $`M`$ components $`x_1,\mathrm{},x_M`$, where each one of the components can take one of the $`K`$ integers, $`1,2,\mathrm{},K`$, with equal probability.
The dynamics are defined by the following steps. In the first step, each one of the perceptrons calculates the $`K`$ induced local fields. For instance, the field $`h_{im}`$ induced by the $`m`$th state on player $`i`$ is defined as the summation over all weights belonging to the $`i`$th perceptron with input equal to $`m`$:
$$h_{im}=\underset{j=1}{\overset{M}{}}w_{ij}\delta _{x_j,m}.$$
(1)
In the second step, each player chooses its state, $`\{\sigma _i\}`$, following the maximal induced field:
$$\sigma _i=\{k_1\underset{m=1,..,K}{\mathrm{max}}h_{im}=h_{ik_1}\}.$$
(2)
where $`\sigma _i`$ is the output (chosen state) of $`i`$th perceptron. In the third step, the occupancy of each state is calculated:
$$N_\rho =\underset{j=1}{\overset{N}{}}\delta _{\sigma _j,\rho },$$
(3)
where it is clear that $`_\rho N_\rho =N`$. The output $`min`$ of the network is the minority decision
$$min=\{\rho \underset{m=1,..,K}{\mathrm{min}}N_m=N_\rho \},$$
(4)
The game can also be represented by a feedforward network $`M:N:1`$ ($`M`$ input units, $`N`$ hidden units and $`1`$ output). All units (input, hidden, output) are represented by $`K`$-states Potts-spins. The weights $`\{w_{ij}\}`$ are from the input units to the hidden units, and the weights from the hidden units to the output are all equal to $`1`$. The dynamics of hidden and output units are similar to zero temperature dynamics of Potts-spin systems , following the maximal induced field. The free parameters in our game are the $`MN`$ weights, $`\{w_{ij}\}`$, from the input to the hidden units. Their values will be determined by the strategy adopted by each one of the players. Our local dynamic rules are based on the generalization of the on-line Hebbian learning rule for $`K=2`$ to general $`K`$-states Potts model with the following updating rule;
$$w_{ij}^+=w_{ij}+\frac{\eta }{M}(K\delta _{x_j,\mathrm{min}}1)$$
(5)
where $`\eta `$ is the learning rate and the sign $`+`$ indicates the next time step. Note that all agents use the same rule for updating their strategy.
The score of the game is determined similarly to the Ising case. Players belonging to the minority ($`N_{min}`$ players) gain $`Q_+`$, while the other $`NN_{min}`$ players gain $`Q_{}`$, where $`Q_+>Q_{}`$. Note that in most previous works $`Q_+`$ was chosen to be $`1`$ and $`Q_{}`$ was chosen to be either $`0`$ or $`1`$. The global profit in such cases is
$$U=Q_{}N+(Q_+Q_{})N_{min}.$$
(6)
It is clear that the maximization of the global profit $`U`$ is equivalent to the maximization of $`N_{min}`$, which is bounded from above by $`N/K`$. Note that in the Ising case each player belongs either to the minority or to the majority, where in the Potts case the situation is more complex. The score may depend on the exact values of $`\{N_i\}`$ (the score decreases with $`N_\rho `$), hence the total profit $`U=U(\{N_i\})`$. In such a case the maximization of the total profit may differ from the maximization of $`N_{min}`$, and will be discussed briefly in the end of this paper.
Before we turn to discuss the guideline of the derivation of the results, which are more involved than for the Ising case, let us present the main results: (a) The score and the dynamics are formulated analytically for general $`K`$, the number of possible decisions. Exact results are obtained for $`K6`$ and asymptotically for $`K\mathrm{}`$. Results for intermediate values of $`K`$ are obtained from simulations. (b) A relaxation to the optimal score is achieved for small learning rates. (c) Regarding the optimal case, the deviation of minority group size from $`N/K`$ is found to be non-monotonic with $`K`$. (d) The total score is independent of the size of the history ($`M`$, the size of the input) available for the agents. (e) All agents are using the same type of dynamic strategy and gain on average (over time) the same profit. Our system does not undergo a phase transition to a state where the symmetry among the agents is broken into losers and winners . Throughout the investigation of the game we assume that the memory size $`M`$ is larger than the number of players $`N`$ (otherwise the completely symmetric Potts configuration is geometrically impossible). Albeit, simulations of the same dynamic for systems with $`M<N`$ show even better results for the global profit.
## III The random case
In case where the maximization of the global profit $`U`$ is identical to the maximization of $`N_{min}`$, the quantity of interest is
$$<ϵ_{min}^{}{}_{}{}^{2}>=\frac{1}{N}<(N_{min}N/K)^2>,$$
(7)
where the symbol $`<>`$ indicates an average over input patterns, and $`N/K`$ is the average number of players in each state. Note that in our calculations the input vector presented to the players at each step of the game consists of random components, instead of the true history. Nevertheless, simulations indicate that the system behavior is only slightly affected by the randomness of the inputs and the game properties remain similar.
For random players, each weight (among the $`MN`$ weights $`\{w_{ij}\}`$) is chosen from a given unbiased distribution and a variance $`1/M`$. Hence, the distribution of the overlap $`R`$ between weights belonging to any two players $`\rho `$ and $`\varphi `$
$$R_{\rho \varphi }=\underset{j=1}{\overset{M}{}}w_{\rho j}w_{\varphi j}$$
(8)
is a Gaussian with zero mean and variance $`1/M`$. In the thermodynamic limit and for $`M>N`$, one can show that in the leading order the overlap between each pair is an independent random variable. For random players and $`K=2`$ one finds $`<ϵ^2>=<_\rho (N_\rho N/K)^2/(NK)>=1/4`$; however, for general $`K`$ even the derivation of a similar quantity is non-trivial. The two cornerstones of the calculations below are the probability of a microscopic configuration $`P(\{\sigma _i\})`$, and the degeneracy $`D(\{N_\rho \})`$ of a macroscopic configuration $`\{N_\rho \}`$, which is given by the multinomial coefficient:
$$D(\{N_\rho \})=\frac{N!}{_\rho N_\rho !}.$$
(9)
In the large $`N`$ limit, the typical deviation of the size of each group from $`N/K`$ is expected to scale with $`\sqrt{N}`$. Hence we define:
$$N_\rho =N/K+ϵ_\rho \sqrt{N}$$
(10)
where it is clear that $`_\rho ϵ_\rho =0`$ and without the loss of generality we assume $`N_{min}=N_1N_\rho \rho >1`$. Applying the Stirling approximation to Eq. (9) yields the degeneracy as a function of $`\{ϵ_\rho \}`$:
$$D_K(\{ϵ_\rho \})K^N\mathrm{exp}(\frac{K}{2}\underset{\rho =1}{\overset{K}{}}ϵ_\rho ^2)\delta (\underset{\rho =1}{\overset{K}{}}ϵ_\rho ).$$
(11)
If the average over $`R_{\rho \varphi }`$, which we denote by $`R`$, is $`0`$, the agents make their choice independently and randomly, so each microscopic configuration has the same probability $`P_K=(1/K)^N`$. Now the average over $`ϵ_{min}^2`$ can be evaluated:
$$<ϵ_{min}^2>_R=\frac{_{\mathrm{}}^0ϵ_{1}^{}{}_{}{}^{2}𝑑ϵ_1\mathrm{\Pi }_{\rho >1}_{ϵ_1}^{\mathrm{}}𝑑ϵ_\rho D_K(\{ϵ_\rho \})P_K(\{ϵ_\rho \})}{_{\mathrm{}}^0𝑑ϵ_1\mathrm{\Pi }_{\rho >1}_{ϵ_1}^{\mathrm{}}𝑑ϵ_\rho D_K(\{ϵ_\rho \})P_K(\{ϵ_\rho \})}.$$
(12)
The quantity $`<ϵ_{min}^{}{}_{}{}^{2}>_{R=0}`$ was calculated numerically for $`K=3,4,5,6`$ and found to be equal to $`0.313,0.322,0.320,0.309`$, respectively (see Fig. 1). Results obtained from simulations with $`N=5000`$ and $`K6`$ are in an excellent agreement with Eq. (12). For $`K>6`$ the reported results in Fig. 1 were derived only from simulations and are in an excellent agreement with the asymptotic behavior of Eq. (12), $`<ϵ_{min}^2>_{R=0}2\mathrm{log}(K)/K`$. Another quantity of interest is the average deviation of the average number of players in each state from $`N/K`$, $`<ϵ^2>=<\frac{1}{K}_\rho ϵ_\rho ^2>`$. Similarly to Eq. (12), this quantity can be derived analytically and gives
$$<ϵ^2>_{R=0}=\frac{K1}{K^2}.$$
(13)
## IV The optimal case
So far we have compared $`<ϵ_{min}^{}{}_{}{}^{2}>`$ and $`<ϵ^2>`$ for random players, where the average overlap is zero. Without breaking the symmetry among the players, the weights can be represented by $`N`$ weight vectors which are symmetrically spread around their center of mass. More precisely, we denote the weight vector of the $`i`$th perceptron as $`𝐰_i`$, and assume that it can be expressed as
$$𝐰_i=𝐂+𝐠_i,$$
(14)
where the center of mass $`𝐂\frac{1}{N}_i𝐰_i`$, and $`\{𝐠_i\}`$ are $`N`$ unit vectors of rank $`M`$ obeying the symmetry
$$𝐠_i𝐠_j=(1+\frac{1}{N1})\delta _{ij}+\frac{1}{N1}.$$
(15)
Hence, the total profit and $`N_{min}`$ are functions of only one parameter, $`C`$. It is clear that the maximization of the total profit or $`N_{min}`$ (as for the case $`K=2`$) is obtained when $`C=0`$, which is the maximal achievable homogeneous repulsion among $`N`$ vectors of rank $`M>N`$. The repulsion is the natural tendency of each player in the minority game, since the goal is to act differently than other players. Without a cooperation which breaks the players into sub-groups, the maximal homogeneous repulsion is $`R=1/(N1)`$.
The two questions of interest are the following: (a) What is $`<ϵ^2>`$ and $`<ϵ_{min}^{}{}_{}{}^{2}>`$ as a function of $`K`$ for the optimal solution, $`C=0`$ and $`R=1/(N1)`$? (b) Is the optimal solution achievable by local dynamic rules for each one of the players? We first examine the former question regarding the optimal solution, and then we turn to study the dynamic behavior of the players.
The average deviation of the number of players in each state from $`N/K`$ at $`C=0`$ and for $`R=O(1/N)`$ can be calculated analytically. The main idea is that this quantity can be calculated similarly to Eq. (12), or via $`<ϵ^2>=1/(NK)<(_{\rho =1}^K_{i=1}^N\delta _{\sigma _i,\rho }N/K)^2>`$. The simplification of the later expression is such that an average over only a pair of players has to be done. The result as a function of $`K`$ gives
$$<ϵ^2>_R=\frac{K1}{K^2}+R(N1)(K1)K\mu ,$$
(16)
where $`\mu =[_{\mathrm{}}^{\mathrm{}}\frac{e^{h^2}}{2\pi }(1H(h))^{K2}𝑑h]^2`$ and $`H(x)=0.5\text{erfc}(\frac{x}{\sqrt{2}})`$.
Regarding the optimal score, the quantity of a particular interest is $`<ϵ_{min}^2>_{R=\frac{1}{N1}}`$. This quantity has to be compared with $`<ϵ_{min}^2>_{R=0}`$ in order to estimate the improvement in the average global gain relative to the random case. Note that the calculation of Eq. (12) for $`R0`$ is nontrivial since $`P_K(\{ϵ_\rho \})`$ is no longer independent of the configuration $`\{ϵ_\rho \}`$. However, we can overcome this difficulty in the following way. For $`R=O(1/N)`$ one can show that in the leading order $`P_K(\{ϵ_\rho \})`$ has the same form as $`D_K(\{ϵ_\rho \})`$:
$$P_K(\{ϵ_\rho \})(1/K)^N\mathrm{exp}(A(R)\underset{\rho =1}{\overset{K}{}}ϵ_\rho ^2)\delta (\underset{\rho =1}{\overset{K}{}}ϵ_\rho ),$$
(17)
where the exact value of $`A(R)`$ is unknown. The observation that both $`P_K(\{ϵ_\rho \})`$ and $`D_K(\{ϵ_\rho \})`$ have the same dependence on $`\{ϵ_\rho \}`$ (Eqs. (11) and (17)) indicates that the ratio $`<ϵ_{min}^2>/<ϵ^2>`$ is independent of $`R`$ if $`R=O(1/N)`$, and in particular:
$$\frac{<ϵ^2>_{R=0}}{<ϵ_{min}^{}{}_{}{}^{2}>_{R=0}}=\frac{<ϵ^2>_{R=\frac{1}{N1}}}{<ϵ_{min}^{}{}_{}{}^{2}>_{R=\frac{1}{N1}}}=\beta _K.$$
(18)
This property can be easily derived by rescaling $`ϵ_\rho \sqrt{A(R)}ϵ_\rho `$ in the integral representation (Eq. (12)) of each one of the four terms in Eq. (18). The same prefactor appears both in the denominator and in the numerator, and the dependence of $`\beta _K`$ on $`R`$ via $`A(R)`$ is cancelled out. Using Eq. (18), $`<ϵ_{min}^{}{}_{}{}^{2}>_{R=\frac{1}{N1}}`$ can be obtained indirectly from the knowledge of the other three terms, which are given by Eqs. (12), (13), and (16). Results for $`<ϵ_{min}^{}{}_{}{}^{2}>_{R=\frac{1}{N1}}`$ are presented in Fig. 1. In order to confirm our analytical results we performed simulations for the optimal case, Eqs. (16) and (18), with $`C=0`$. The simulations were done in two stages. In the first stage, $`N`$ normalized vectors of rank $`M`$, obeying the constraints that the overlap among each pair is equal to $`1/(N1)`$, are generated using a recursive process. The details of the algorithm will be given elsewhere . In the second stage, $`<ϵ_{min}^{}{}_{}{}^{2}>`$ and $`<ϵ^2>`$ were averaged over about $`10^5`$ randomly chosen inputs for a system with $`N=400`$ and $`M=5000`$. An excellent agreement between simulations and analytical results was obtained (see Fig. 1). The improvement in the global gain can be measured by the ratio $`\mathrm{\Gamma }_K=<ϵ_{min}^{}{}_{}{}^{2}>_{R=0}/<ϵ_{min}^{}{}_{}{}^{2}>_{R=\frac{1}{N1}}`$. This ratio decreases monotonically with $`K`$ such that its maximal value $`\mathrm{\Gamma }_2=2.7548`$ and for $`K\mathrm{}`$ $`\mathrm{\Gamma }_K1`$ (inset of Fig. 1).
## V The dynamics which lead to the optimal solution
So far we derived the properties of the optimal solution for different values of $`K`$. Now we are turning to the second question: is the optimal solution achievable by local dynamic rules (Eq. (5))? After averaging Eq. (5) over $`j`$ and in the limit where the number of examples, $`\alpha M`$, scales with the number of input units $`M`$, one can find the following equation of motion for the center of mass
$$\frac{dC^2}{d\alpha }=2\eta K<\underset{j}{}C_j\delta _{x_j,min}>+\eta ^2(K1),$$
(19)
where $`<>`$ denotes an average over the random examples. For large $`M`$, in the leading order each input vector divides each weight vector into $`K`$ equal groups of size $`M/K`$. The minority state is the one whose group of weights gives the minimal sum. Using Eq. (19) and $`M,N\mathrm{}`$, $`<_jC_j\delta _{x_j,min}>`$ is the average minimal sum of a set of $`M/K`$ center of mass components, $`\{C_j\}`$. These $`M/K`$ quantities are random variables with zero mean and variance $`C^2/M`$ ( $`<_{j=1}^{M/K}C_j>=0`$ and $`<(_{j=1}^{M/K}C_j)^2>=C^2/K`$). One can find that $`<_jC_j\delta _{x_j,min}>`$ is equal to
$$\frac{C}{2}(K1)\sqrt{\frac{K}{\pi }}_{\mathrm{}}^{\mathrm{}}\frac{e^{y^2/2}}{\sqrt{2\pi }}[H(\frac{y}{\sqrt{2}})]^{K2}𝑑y.$$
(20)
Hence, for a given $`K`$, Eqs. (19) and (20) indicate a linear relation between the fixed point value of $`C`$ and the learning rate $`\eta `$ with corrections of $`O(1/\sqrt{N})`$. As $`\eta 0`$, $`C0`$ and the system approaches the optimal configuration. The interplay between $`C`$ and $`\eta `$ was confirmed by simulations, where finite size effects decay as the size of the system becomes larger. This effect is depicted in Fig. 2 for $`K=3`$. The explicit dependence of $`<ϵ_{min}^{}{}_{}{}^{2}/N>_R`$ on $`C`$ can be found for $`RO(1/N)`$ via the relation
$$R=\frac{C^2\frac{1}{N1}}{C^2+1}.$$
(21)
Results of simulations for $`<ϵ_{min}^{}{}_{}{}^{2}>_R`$ as a function of $`C`$ for $`N=103`$ and $`M=200`$ are presented in the inset of Fig. 2. An excellent agreement between the analytical prediction and simulations was obtained in the regime of $`CO(1/\sqrt{N})`$ (corresponding to $`RO(1/N)`$).
Note that although the global gain $`U`$ which corresponds to the Boolean case is monotonic with $`K`$, the non-monotonic behavior of $`ϵ_{min}`$ implies that for non-Boolean cases non-monotonic behavior of $`U`$ may be obtained.
## VI summary and outlook
In this paper we introduced a generalization of the minority game to the case of multi-choice. The problem was applied to a multilayer network with updating rules for the weights (strategies). Static and dynamic properties of the strategies were solved analytically for various $`K`$’s and were found to be in a good agreement with simulations on finite systems. This modification of the minority game to the case of multi-choice open a manifold of new questions, which certainly deserve future research. We have chosen two of those questions to briefly discuss here.
Firstly, as we have pointed out before, the function according to which the profit is awarded is not necessarily Boolean as in Eq. (6). In fact, the model is more realistic when the profit of a player is related to the size of his group, as well as to the size of the other groups . Our analysis can be applied to these cases if the maximization of the global gain is equivalent to the maximization of the minority group. However, other scores may not fulfill this required condition. In these cases, it has to be determined whether the optimal symmetric configuration remains the maximal repulsion.
Secondly, the other strategies for the minority game that have been studied can be generalized to multi-choice situations in a straightforward manner: in the original game where each player has several decision tables, each table entry is now a value between $`1`$ and $`K`$. In Johnson’s stochastic strategy , each player has a probability of choosing the outcome that was successful the last time, or to pick one of the others with equal probability. In the strategy of Reents , players who were not in the minority could switch to some other action with a small probability in the next time step. Similarly, other conceivable strategies can also be generalized. Preliminary checks imply that all these modified strategies show similar behavior compared to that of the binary-choice game, even though their theoretical treatment probably becomes more involved. While outcomes of these games certainly have to be measured against the reference values given in Eqs. (12) and (13), it is not clear under what circumstances relations like Eq. (18) hold for other strategies.
I. K., W. K. and R. M. acknowledge a partial support by GIF.
2
|
warning/0005/cond-mat0005006.html
|
ar5iv
|
text
|
# Naudts-like duality and the extreme Fisher information principle
## I Introduction
We are going to be concerned in what follows with the workings of two information measures that have received much attention lately, those of Fisher and Tsallis , respectively. Our goal is to show that their interplay naturally yields a type of Naudts’ duality .
Fisher’s information measure (FIM) was advanced already in the twenties, well before the advent of Information Theory (IT), being conventionally designed with the symbol $`I`$ (see Eq. (1) below for the pertinent definition). Much interesting work has been devoted to the physical applications of FIM in recent times (see, for instance, and references therein). Frieden and Soffer have shown that Fisher’s information measure provides one with a powerful variational principle, the extreme physical information (EPI) one, that yields the canonical Lagrangians of theoretical physics . Additionally, $`I`$ has been shown to provide an interesting characterization of the “arrow of time” alternative to the one associated with Shannon’s $`S`$ .
Tsallis’ measure is a generalization of Shannon’s one. Notice that IT was created by Shannon in the forties . One of its fundamental tenets is that of assigning an information content (Shannon’s measure) to any normalized probability distribution. The whole of statistical mechanics can be elegantly re-formulated by extremization of this measure, subject to the constraints imposed by the a priori information one may possess concerning the system of interest . It is shown in that a parallel process can be undertaken with reference to Tsallis’ one, giving rise to what is called Tsallis’ thermostatistics, responsible for the successful description of an ample variety of phenomena that cannot be explained by appeal to the conventional one (that of Boltzmann-Gibbs-Shannon) .
## II A brief Fisher primer
Fisher’s information measure $`I`$ is of the form
$$I=𝑑xf(x,\theta )\left[\frac{1}{f(x,\theta )}\frac{f}{\theta }\right]^2,$$
(1)
where $`x`$ is a stochastic variable and $`\theta `$ a parameter on which the probability distribution $`f(x,\theta )`$ depends. The Fisher information measure provides a lower bound for the mean-square error associated with the estimation of the parameter $`\theta `$. No matter what specific procedure we chose in order to determine it, the associated mean square error $`e^2`$ has to be larger or equal than the inverse of the Fisher measure . This result, i.e., $`e^2\frac{1}{I},`$ is referred to as the Cramer-Rao bound, and constitutes a very powerful statistical result .
The special case of translation families deserves special mention. These are mono parametric families of distributions of the form $`f(x\theta )`$ which are known up to the shift parameter $`\theta `$. Following Mach’s principle, all members of the family possess identical shape (there are no absolute origins), and here Fisher’s information measure adopts the appearance
$$I=𝑑x\frac{1}{f}\left[\frac{f}{x}\right]^2.$$
(2)
The parameter $`\theta `$ has dropped out. $`I=I[f]`$ becomes then a functional of $`f`$.
At this point we introduce the useful concept of escort probabilities (see and references therein), that one defines in the fashion
$$F_q(x)=\frac{f(x)^q}{f(x)^q𝑑x},$$
(3)
$`q`$ being any real parameter, $`F_q(x)𝑑x=1`$, and, of course, for $`q=1`$ we have $`F_1f`$.
The concomitant “escort-FIM” becomes
$$I[F_q]=𝑑xF_q(x)\left[\frac{1}{F_q(x)}\frac{F_q(x)}{x}\right]^2,$$
(4)
that, in terms of the original $`f(x)`$ acquires the aspect
$$I[F_q]=q^2\frac{𝑑xf(x)^{q2}\left[\frac{f(x)}{x}\right]^2}{𝑑xf(x)^q}.$$
(5)
We shall denote with $`I_q`$ the new “escort-FIM”
$$I_q=\frac{𝑑xf(x)^{q2}\left[\frac{f(x)}{x}\right]^2}{𝑑xf(x)^q}.$$
(6)
(Notice that for $`q=0`$ the integration range must be finite in order to avoid divergences in the denominator.)
The parameter $`q`$ can be identified with Tsallis’ nonextensivity index , which allows one to speak of “Fisher measures in a nonextensive context”. Their main properties have been discussed in .
## III The extreme physical information principle (EPI)
The Principle of Extreme Physical Information (EPI) is an overall physical theory that is able to unify several sub-disciplines of Physics . In Ref. Frieden and Soffer (FS) show that the Lagrangians in Physics arise out of a mathematical game between an intelligent observer and Nature (that FS personalize in the appealing figure of a “demon”, reminiscent of the celebrated Maxwell’s one). The game’s payoff introduces the EPI variational principle, which determines simultaneously the Lagrangian and the physical ingredients of the concomitant scenario.
FS envision the following situation, involving Fisher’s information for translation families: some physical phenomenon is being investigated so as to gather suitable, pertinent data. Measurements must be performed. Any measurement of physical parameters appropriate to the task at hand initiates a relay of information $`I`$ (or $`I_q`$ in a non-extensive environment) from Nature (the demon) into the data. The observer acquires information, in this fashion, that is precisely $`I`$ (or $`I_q`$). FS assume that this information can be elicited via a pertinent experiment. Nature’s information is called, say, $`J`$ .
Assume now that, due to the measuring process, the system is perturbed, which in turn induces a change $`\delta J`$. It is natural to ask ourselves how the data information $`I_q`$ will be affected. Enters here FS’s EPI: in its relay from the phenomenon to the data no loss of information should take place. The ensuing new Conservation Law states that $`\delta J=\delta I_q`$, or, rephrasing it
$$\delta (I_qJ)=\mathrm{\hspace{0.17em}0},$$
(7)
so that, defining an action $`𝒜_q`$
$$𝒜_q=I_qJ,$$
(8)
EPI asserts that the whole process described above extremizes $`𝒜_q`$. FS conclude that the Lagrangian for a given physical environment is not just an ad-hoc construct that yields a suitable differential equation. It possesses an intrinsic meaning. Its integral represents the physical information $`𝒜_q`$ for the physical scenario. On such a basis some of the most important equations of Physics can be derived for $`q=1`$ . For an interesting Quantum Mechanical derivation see . A cosmological application of the nonextensive ($`q1`$) conservation law (7) is reported in . Mechanical analogs that can be built up using this law are discussed in . Notice, however, that the last two references use an old Tsallis’ normalization procedure (advanced in ), that cannot be assimilated within the framework of the escort distribution concept.
## IV Solutions to the variational problem
According to EPI, $`J`$ is fixed by the physical scenario . We adopt here a more modest posture by assuming that $`J`$ embodies only the normalization constraint, and say nothing regarding a specific physical scenario. $`J`$ is just
$$J=\lambda f(x)𝑑x,$$
(9)
where $`\lambda `$ is the pertinent Lagrange multiplier. Such a $`J`$ has been successfully employed in with reference to a quantum mechanical problem. Playing the Frieden-Soffer game, i.e., performing the variation (7), leads then to
$$2f\ddot{f}+(q2)\dot{f}^2+qI_qf^2+\lambda Qf^{3q}=0$$
(10)
a $`q`$-dependent, non-linear differential equation that should yield our “optimal” probability distribution $`f`$ (we set $`Q=f^q𝑑x`$). Now, one should demand that, for $`q=1`$, (10) become identical to the differential equation that arises in such circumstances (see that equation in , for instance, and call $`\lambda ^{^{}}`$ the concomitant Lagrange multiplier used there). This requirement is fulfilled if we set $`\lambda =\lambda ^{^{}}qI_q`$. The $`q=1`$-expression becomes then
$$2f\ddot{f}\dot{f}^2+\lambda ^{}f^2=0,$$
(11)
where, of course, one has $`Q=1`$. The solution of Eq. (11) is of the form
$$f_{q=1}(x)=A^2\mathrm{cos}^2k(xx_0)$$
(12)
where $`k`$ is a constant to be determined below and $`A`$, $`x_0`$ are arbitrary integration constants.
It easy to show that (10) has, as a first integral,
$$\dot{f}^2+I_qf^2+\lambda Qf^{3q}=cf^{2q},$$
(13)
where $`c`$ is an integration constant. This equation involves Fisher’s generalized information for translation families. We must solve it having (6) in mind. In order to establish the consistency between (13) and (6) we introduce a set of normalized variables
$$z=\sqrt{I_q}𝑑x,\overline{\lambda }=\frac{\lambda Q}{I_q},\overline{c}=\frac{c}{I_q},$$
(14)
(the integral is an indefinite one) in terms of which Eqs. (6), (9), (10), and (13) are transformed into
$$1=\frac{f^{q2}f_{}^{}{}_{}{}^{2}𝑑z}{f^q𝑑z},$$
(15)
$$J_q=\overline{\lambda }\frac{I_q}{f^q𝑑z}f(z)𝑑z,$$
(16)
(an indefinite integral),
$$2ff^{\prime \prime }+(q2)f^{}{}_{}{}^{2}+qf^2+\overline{\lambda }f^{3q}=0,$$
(17)
and
$$f^{}{}_{}{}^{2}+f^2+\overline{\lambda }f^{3q}=\overline{c}f^{2q}.$$
(18)
Inserting (18) into (6) we conclude that the integration constant acquires the aspect
$$\overline{c}=\frac{2Q+\overline{\lambda }}{x_2x_1},$$
(19)
where $`x_2`$ and $`x_1`$ are the integration limits, to be fixed by the remaining parameters of the theory. A quite interesting point is that the general solution of (18) can be given in closed form as
$$^zdz=zconst.=\pm \frac{f^{\frac{q}{2}1}}{\sqrt{\overline{c}\overline{\lambda }ff^q}}df,$$
(20)
where the constants $`\overline{c},\overline{\lambda }`$ must be of such nature that a real $`f`$ ensues.
## V Symmetry properties of the EPI probability distribution
We start by changing variables in (17) to
$$u=\frac{f^{}(z)}{f(z)},$$
(21)
and obtaining
$$u^{\prime \prime }+\alpha uu^{}+\beta u^3+\gamma u=0,$$
(22)
with
$$\alpha =(2q1),\beta =\frac{1}{2}q(q1),\gamma =\beta .$$
(23)
(A complete study of the properties of equation (22) is found in ). Further, we effect the transformation
$`f`$ $``$ $`1/f,`$ (24)
so that
$$uu,u^{}u^{},u^{\prime \prime }u^{\prime \prime }.$$
(25)
If we require that equation (22) be invariant under this transformation, the parameters $`\alpha `$, $`\beta `$ and $`\gamma `$ must change according to $`\alpha \alpha `$, $`\beta \beta `$ and $`\gamma \gamma `$ respectively. This entails that the parameter $`q`$, that characterizes the degree of non-extensivity of the system, transform as $`q1q`$. A property of this type has been called “duality” by Naudts , although in his case the relationship is of the form $`q\frac{1}{q}`$ (duality between $`q>1`$ statistics and $`q<1`$ one). In our case, the duality arises between two $`q`$-values whose sum adds up to unity.
Introducing now into (17) the new variable
$$h=\frac{1}{f},$$
(26)
we get
$$2hh^{\prime \prime }(q+2)h^2qh^2\overline{\lambda }h^{q+1}=0,$$
(27)
which under the substitution $`q1q`$, becomes
$$2hh^{\prime \prime }+(q3)h^2+(q1)h^2\overline{\lambda }h^{2q}=0.$$
(28)
This equation can be rewritten, if we first define
$$w(q)=(h^2h^2\overline{\lambda }h^{2q}+\overline{c}h^{3q}),$$
(29)
as
$$2hh^{\prime \prime }+(q2)h^2+qh^2\overline{c}h^{3q}+w(q)=0,$$
(30)
where the terms in $`w(q)`$ correspond to the (transformed) first integral of (17)
$$f^2+f^2+\overline{\lambda }f^{3q}=\overline{c}f^{2q},$$
(31)
which under (26) becomes
$$h^2+h^2+\overline{\lambda }h^{2q}=\overline{c}h^{3q}.$$
(32)
As a consequence, $`w(q)`$ in (30) vanishes and the equation (17), under the transformation (26), turns out to retain its form, changing $`q1q`$ and $`\overline{c}\overline{\lambda }`$. It is convenient at this point to effect a slight change of notation and denote by $`f_q`$ the solution to (17) that obtains when the nonextensivity index is $`q`$. The above symmetry argument entails
$$f_q(\overline{c},\overline{\lambda })\frac{1}{f_{1q}(\overline{\lambda },\overline{c})}.$$
(33)
Using this symmetry property we can re-obtain the probability distribution (12) for $`q=1`$, i.e., the ordinary, extensive one, in term of the probability distribution for $`q=0`$, that can be easily calculated from (18)
$$f^{}_{}{}^{}2=(\overline{c}1)f^2\overline{\lambda }f^3,q=0.$$
(34)
The solutions are
$$f_0(z)=\frac{\overline{c}1}{\overline{\lambda }}\left\{1\mathrm{tanh}^2\frac{\sqrt{\overline{c}1}}{2}(zz_0)\right\}\overline{c}>1,$$
(35)
and
$$f_0(z)=\frac{\overline{c}1}{\overline{\lambda }}\left\{1+\mathrm{tan}^2\frac{\sqrt{1\overline{c}}}{2}(zz_0)\right\}\overline{c}<1,$$
(36)
where the last solution must be normalized in a finite interval. The symmetry transformation (33) yields now the general solution for $`q=1`$
$$f_1(\overline{c},\overline{\lambda })\frac{1}{f_0(\overline{\lambda },\overline{c})}.$$
(37)
This is to be compared with the result (12). We start with (36), effect the transformation (37) and reach
$$f_1(z)=\frac{\overline{c}}{1+\overline{\lambda }}\mathrm{cos}^2\frac{\sqrt{1+\overline{\lambda }}}{2}(zz_0)$$
(38)
which, after a little algebra that involves also going back to the $`x`$ variable adopts indeed the form (12) with $`A^2=c/\lambda ^{}`$ and $`k=\sqrt{\lambda }/2`$. A similar analysis can be performed for (35).
We have thus found the general solution for the (extensive) EPI variational treatment corresponding to a $`J`$ that entails just normalization of the probability distribution. Notice that, within the context of Naudts’ effort , the extensive thermostatistics $`q=1`$ is self-dual. Instead, according to the present Fisher framework, the self-dual instance obtains for $`q=1/2`$.
## VI Conclusions
We have shown that the EPI principle, used in conjunction with a Fisher measure constructed with escort distributions that depend upon the Tsallis index $`q`$, renders a probability distribution endowed with a remarkable symmetry: a Naudts’-like duality .
Tsallis enthusiasts had thought, before the advent of Naudts work , that a different statistics obtains for each different value of the nonextensivity index $`q`$. The duality concept is then important because it ascribes the same statistics to a given pair of (suitably related) $`q`$-values. We have shown here that such a pair can be selected in two distinct manners, i.e., à la Naudts or à la Fisher, and have detailed the prescription corresponding to the latter choice.
Finally, we have also ascertained which is the general (normalized) probability distribution that extremizes the physical information.
|
warning/0005/hep-ph0005115.html
|
ar5iv
|
text
|
# Hyperon Beta–Decay and Axial Charges of the Lambda in view of Strongly Distorted Baryon Wave–Functions
## I Introduction
The study of the axial current matrix elements (or quark spin structure) of the $`\mathrm{\Lambda }`$ hyperon, which is an interesting subject in its own, has gained further attraction as it has been suggested that polarized $`\mathrm{\Lambda }`$’s could be utilized to gain information about the proton spin structure, i.e. the nucleon axial vector matrix elements. For this to be a sensible program it is necessary that large polarizations of the up and down quarks, $`\mathrm{\Delta }U_\mathrm{\Lambda }`$ and $`\mathrm{\Delta }D_\mathrm{\Lambda }`$, in the iso–singlet $`\mathrm{\Lambda }`$ carry over to the corresponding fragmentation functions. Although this is an assumption one would expect the non–strange $`\mathrm{\Lambda }`$ fragmentation functions to be small if model calculations undoubtly showed that $`\mathrm{\Delta }U_\mathrm{\Lambda }`$ and $`\mathrm{\Delta }D_\mathrm{\Lambda }`$ were small. On the other hand, if model calculations indicated that $`\mathrm{\Delta }U_\mathrm{\Lambda }`$ and $`\mathrm{\Delta }D_\mathrm{\Lambda }`$ were large, it would provide sufficient motivation to study and measure these fragmentation functions . Using results on the axial current matrix elements from deep–inelastic scattering as well as hyperon beta–decay data together with flavor covariance indeed results in sizable polarizations for the non–strange quarks, $`\mathrm{\Delta }U_\mathrm{\Lambda }=\mathrm{\Delta }D_\mathrm{\Lambda }0.20`$ together with $`\mathrm{\Delta }S_\mathrm{\Lambda }0.60`$ for the strange quark . The use of flavor covariance is motivated by the feature that the Cabibbo scheme utilizing the $`F`$&$`D`$ parameterization for the flavor changing axial charges works unexpectedly well as the comparison in table I exemplifies. In the present note we will study an approach which allows the incorporation of deviations from the flavor symmetric formulation, after all $`SU(3)`$–flavor is not an exact symmetry. Clearly, any model that reproduces the data equally well as the Cabibbo scheme with a minimal set of parameters can be regarded as a reasonable description of hyperon beta–decay.
Our treatment of flavor symmetry breaking is based on the Skyrme model approach to describe baryons as solitons in an effective meson theory. In this type of models baryon states are obtained by quantizing the large amplitude fluctuations (zero modes) of the soliton. In the proceeding section we will briefly review the quantization procedure with the inclusion of flavor symmetry breaking. This approach is of great fundamental interest not only for the large $`N_C`$ treatment of QCD but certainly also in its own right; especially because this approach has been (at least quantitatively) successful in understanding the proton spin problem . Furthermore, in the framework of quantizing the soliton the study of flavor symmetry breaking in these models is very interesting. This is even more the case as some of the difficulties encountered earlier (such as the overall scale in the predicted baryon mass differences or the unexpectedly large strangeness contribution to nucleon matrix elements ) have been largely understood and solved . It is thus appealing to also study the $`\mathrm{\Lambda }`$ axial charges in such a framework, especially because they might be accessible experimentally . Here we will focus on a description with the symmetry breaking mainly residing in the baryon wave–functions, including important higher order contributions. The order parameter is the strange current quark mass, $`m_s`$. In the effective meson Lagrangian it emerges via the meson properties, e.g. $`m_K^2m_\pi ^2=𝒪(m_s)`$. Sizable deviations from flavor symmetric (octet) wave–functions are needed in the chiral soliton approach to account for the pattern of the baryon mass–splittings . The proposed picture implies that the strange quark component in the sea is suppressed, a scenario which has also been considered in ref . On the other hand we will assume that the current operators, from which the charges are computed, are dominated by their flavor symmetric components. We will find that the proposed approach approximately reproduces the data with no (or minimal) explicit symmetry breaking in the axial charge operator. The present studies represent a refinement of some earlier calculations as we now include contributions to the axial charge operator which were omitted in ref and are subleading in the $`1/N_C`$ counting. A systematic expansion in $`1/N_C`$ would also require a careful treatment of the allowed representations in flavor space for the baryon wave–functions. We do not attempt such an expansion but rather assume the physical value $`N_C=3`$. In addition we present the results obtained from a complete calculation in a realistic vector meson soliton model. That calculation supports the suggested picture.
## II Symmetry Breaking in the Baryon Wave–Functions
Here we briefly review the energy eigenvalue problem for the low–lying $`\frac{1}{2}^+`$ and $`\frac{3}{2}^+`$ baryons as it arises in the collective coordinate treatment of chiral soliton models. This approach was initiated in ref . In the model framework it leads to exact eigenstates for an arbitrary strength of the flavor symmetry breaking. The collective coordinates for flavor rotations are introduced via
$`U(\stackrel{}{r},t)=A(t)U_0(\stackrel{}{r})A^{}(t),A(t)SU(3).`$ (1)
$`U_0(\stackrel{}{r})`$ describes the soliton field configuration embedded in the isospin subgroup of flavor $`SU(3)`$. A prototype model Largangian for the chiral field $`U(\stackrel{}{r},t)`$ would consist of the Skyrme model supplemented by the Wess–Zumino–Witten terms as well as suitable symmetry breaking pieces. In the action notation it reads
$`\mathrm{\Gamma }={\displaystyle d^4x\left\{\frac{f_\pi ^2}{4}\text{Tr}\left[_\mu U(^\mu U)^{}\right]+\frac{1}{32ϵ^2}\text{Tr}\left[[U^{}_\mu U,U^{}_\nu U]^2\right]\right\}}+\mathrm{\Gamma }_{WZ}+\mathrm{\Gamma }_{SB}.`$ (2)
Here $`f_\pi `$ is the pion decay constant and $`ϵ`$ is the dimensionless Skyrme parameter. $`\mathrm{\Gamma }_{WZ}`$ is the Wess-Zumino action :
$`\mathrm{\Gamma }_{WZ}`$ $`=`$ $`{\displaystyle \frac{iN_C}{240\pi ^2}}{\displaystyle _{M_5}}d^5xϵ^{\mu \nu \rho \sigma \tau }\text{Tr}[L_\mu L_\nu L_\rho L_\sigma L_\tau ]\mathrm{with}M_5=M_4.`$ (3)
Here we have used $`L_\mu =U^{}_\mu U`$. The flavor symmetry breaking terms are contained in
$`\mathrm{\Gamma }_{SB}`$ $`=`$ $`{\displaystyle }d^4x\{{\displaystyle \frac{f_\pi ^2m_\pi ^2+2f_K^2m_K^2}{12}}\text{Tr}[U+U^{}2]+{\displaystyle \frac{f_\pi ^2m_\pi ^2f_K^2m_K^2}{2\sqrt{3}}}\text{Tr}\left[\lambda _8(U+U^{})\right]`$ (5)
$`+{\displaystyle \frac{f_K^2f_\pi ^2}{4}}\text{Tr}\left[\widehat{S}(U(_\mu U)^{}^\mu U+U^{}_\mu U(^\mu U)^{})\right]\},`$
where $`\widehat{S}=\mathrm{diag}(0,0,1)`$ is the strangeness projector. It should be emphasized that many of the arguments presented below apply to more general chiral Lagrangians, though.
An appropriate parameterization of the collective coordinates in terms of eight “Euler–angles” is given by
$`A=D_2(\widehat{I})\mathrm{e}^{i\nu \lambda _4}D_2(\widehat{R})\mathrm{e}^{i(\rho /\sqrt{3})\lambda _8},`$ (6)
where $`D_2`$ denote rotation matrices of three Euler–angles for each, rotations in isospace ($`\widehat{I}`$) and coordinate–space ($`\widehat{R}`$). Substituting this configuration into the model Lagrangian yields upon canonical quantization the Hamiltonian for the collective coordinates $`A`$:
$`H=H_\mathrm{s}+\frac{3}{4}\gamma \mathrm{sin}^2\nu .`$ (7)
The symmetric piece of this collective Hamiltonian only contains Casimir operators and may be expressed in terms of the $`SU(3)`$–right generators $`R_a,`$ with $`[A,R_a]=(1/2)A\lambda _a,`$ where $`a=1,\mathrm{},8:`$
$`H_\mathrm{s}=M_{\mathrm{cl}}+{\displaystyle \frac{1}{2\alpha ^2}}{\displaystyle \underset{i=1}{\overset{3}{}}}R_i^2+{\displaystyle \frac{1}{2\beta ^2}}{\displaystyle \underset{\alpha =4}{\overset{7}{}}}R_\alpha ^2.`$ (8)
$`M_{\mathrm{cl}},\alpha ^2,\beta ^2`$ and $`\gamma `$ are functionals of the soliton, $`U_0(\stackrel{}{r})`$. The field theoretical problem has been transformed into a quantum mechanical problem for the collective coordinates which are parameterized by the ‘Euler–angles’ (6). The symmetry breaking term in the Hamiltonian (7) depends on only one of the eight ‘Euler–angles’. This suggests the following parameterization of the baryon eigenfunctions ,
$`\mathrm{\Psi }_{I,I_3,Y;J,J_3,Y_R}(A)={\displaystyle \frac{1}{\sqrt{N}}}{\displaystyle \underset{M_L,M_R}{}}D_{I_3,M_L}^{(I)}(\widehat{I})f_{M_L,M_R}^{(I,Y;J,Y_R)}(\nu )\mathrm{e}^{iY_R\rho }D_{M_R,J_3}^{(J)}(\widehat{R}).`$ (9)
The unit baryon number sector constrains the right hypercharge to $`Y_R=1`$. This constraint stems form $`\mathrm{\Gamma }_{WZ}`$ and is valid for $`N_C=3`$. The flavor hypercharge quantum number emerges via the constraint $`YY_R=2(M_LM_R)`$ for the intrinsic (iso–)spin projections $`M_L`$ and $`M_R`$.
The generators $`R_a`$ can be expressed in terms of derivatives with respect to the ‘Euler–angles’. Then the eigenvalue problem $`H\mathrm{\Psi }=ϵ\mathrm{\Psi }`$ reduces to sets of ordinary second order differential equations for the isoscalar functions $`f_{M_L,M_R}^{(I,Y;J,Y_R)}(\nu )`$. The product $`\omega ^2=\frac{3}{2}\gamma \beta ^2`$ appears as a continuous parameter in the eigenvalue equation. Hence the eigenfunctions (9) parametrically depend on $`\omega ^2`$ which is thus interpreted as the effective strength of the flavor symmetry breaking. A value in the range $`5\stackrel{<}{}\omega ^2\stackrel{<}{}8`$ is required to obtain reasonable agreement with the empirical mass differences for the $`\frac{1}{2}^+`$ and $`\frac{3}{2}^+`$ baryons . In particular, reproducting the observed spacing $`(M_\mathrm{\Lambda }M_N):(M_\mathrm{\Sigma }M_\mathrm{\Lambda }):(M_\mathrm{\Xi }M_\mathrm{\Sigma })=1:0.43:0.69`$ demands a sizable $`\omega ^2`$ since a leading order treatment of the eigenvalue equation (7) incorrectly yields $`1:1:\frac{1}{2}`$. In the exact treatment we get significantly closer to the empirical values, e.g. for $`\omega ^2=6.0`$ and $`\omega ^2=8.0`$ we find the ratios $`1:0.69:0.70`$ and $`1:0.61:0.77`$, respectivelyOne might want to add other symmetry breaking operators to (7) but it should be reminded that they are of higher order in $`1/N_C`$..
The symmetry breaking piece in eq (7) has non–zero matrix elements when sandwiched between baryon states that differ only by their respective $`SU(3)`$ representation. Hence the exact as well as the perturbative treatments lead to baryon states which are not pure octet (or decuplet) states. Rather they have admxitures of states that are members of higher dimensional $`SU(3)`$ representations but otherwise have identical quantum numbers. Baryons in these representations cannot be constructed as three quark states, rather additional quark–antiquark pairs are required. Hence such admxitures to the octet (or decuplet) baryon wave–functions can be interpreted as an effective parameterization of the meson cloud. In general those admixtures reduce the baryon matrix elements associated with operators like $`\overline{s}s`$<sup>§</sup><sup>§</sup>§For example, the normalized nucleon matrix element $`N|\overline{s}s|N/N|\overline{u}u+\overline{d}d+\overline{s}s|N`$ reduces from 23% for the pure octet wave–function to 17% at $`\omega ^2=5.0`$.. Hence the meson cloud dominantly consists of pions. This merely reflects the fact that due to the smaller masses the non–strange degrees of freedom are easier to excite than those related to strange quarks. Here we are interested in the consequences that arise from the exact treatment of the collective coordinates. We therefore compare the admxitures of states from higher dimensional representations as they result from the exact calculation outlined above with those obtained in the first order approximation. As an example we list the admixture of the nucleon wave–functions with states carrying nucleon quantum numbers dwelling in higher dimensional $`SU(3)`$ representations in table II.
We observe that in the relevant range for $`\omega ^2`$ the first order approximation has only limited validity. In particular the $`\overline{\mathrm{𝟏𝟎}}`$–amplitude is overestimated by this approximation.
The feature that the effective symmetry breaking parameter also contains the moment of inertia, $`\beta ^2`$ for rotations into strangeness direction allows the possibility that the symmetry breaking in the wave–functions, which is measured by $`\omega ^2`$, to be large albeit the explicit symmetry breaking, measured by $`\gamma `$, is not (and vice versa). Furthermore this allows for the scenario of having large deviations from flavor symmetric wave–functions without even having symmetry breaking components in the current operators since almost all symmetry breaking can eventually be included in non–derivative terms of $`\mathrm{\Gamma }_{SB}`$ which do not contribute to currents. In the next section we will study whether such a picture can be consistent with the observations on hyperon beta–decays. These decays are well parameterized by the Cabibbo scheme which is obtained by applying the Wigner–Eckart theorem to the $`SU(3)`$ symmetric baryon octet wave–functions.
## III Charge Operators
In this section we present an investigation based on the covariance in the collective coordinate approach. In this context it is not necessary to detail the model Lagrangian. In the proceeding section we will nevertheless present an analysis which utilizes a specific Lagrangian as an example. It will be found that this example essentially verifies the results obtained from the covariant treatment.
Recently a chiral soliton model motivated analysis of the axial charges of the hyperons has been performed . Up to linear order in the strange current quark mass, $`m_s`$ (next–to–leading order in flavor symmetry breaking) all operators for the respective matrix elements were collected. Their coefficients were determined from known data on hyperon beta–decayNote that the standard definition for this decay parameter differs from that in ref by a factor $`\sqrt{6}/2`$.. A model result was used to relate octet and singlet currents because they are not related by group theory. Then the polarization for the non–strange quarks in the $`\mathrm{\Lambda }`$ was predicted to be small, $`\mathrm{\Delta }U_\mathrm{\Lambda }=0.02\pm 0.17`$ in contrast to $`\mathrm{\Delta }S_\mathrm{\Lambda }=1.21\pm 0.54`$; with errors of the $`\mathrm{\Xi }`$ decay data penetrating through this analysis. Some of the results (for the central values) raise questions in view of the study representing a perturbation expansion in flavor symmetry breaking: The axial singlet matrix element of the $`\mathrm{\Lambda }`$, $`\mathrm{\Delta }\mathrm{\Sigma }_\mathrm{\Lambda }`$, turned out to be about twice as large as that of the nucleon, $`\mathrm{\Delta }\mathrm{\Sigma }_N`$. Also, the $`𝒪(m_s)`$ terms contributed almost 50% to $`\mathrm{\Delta }S_\mathrm{\Lambda }`$. This indicates that at this order the expansion has not converged (if it does at all) or that in chiral soliton models the flavor symmetric point may not be the most suitable one to expand about. This may be perceived from the observation that in chiral soliton models the effect of the derivative type symmetry breaking terms is mainly indirect. They provide the splitting between the various decay constants and thus significantly increase $`\gamma `$ because it is proportional to $`f_K^2m_K^2f_\pi ^2m_\pi ^21.5f_\pi ^2(m_K^2m_\pi ^2)`$. Besides this indirect effect the derivative type symmetry breaking terms in (5) may be omitted. Whence there are no symmetry breaking terms in current operators and the octet axial charge operator may be parameterized as
$`{\displaystyle d^3rA_i^{(a)}}=c_1D_{ai}c_2D_{a8}R_i+c_3{\displaystyle \underset{\alpha ,\beta =4}{\overset{7}{}}}d_{i\alpha \beta }D_{a\alpha }R_\beta ,a=1,\mathrm{},8,i=1,2,3.`$ (10)
Under flavor transformations (parameterized by changes of the collective coordinates) these charge operators behave like members of an octet. The expression (10) stems from subsituting the parameterization (1) into the covariant expression of the current operator obtained from a model Lagrangian using Noether’s theorem and subsequently applying the quantization rules for the collective coordinates. Here we have furthermore introduced the adjoint representation of the collective rotations, $`D_{ab}=\frac{1}{2}\mathrm{tr}\left(\lambda _aA\lambda _bA^{}\right)`$. In principle, the constants $`c_n,n=1,2,3`$, are functionals of the soliton which can be computed within the adopted model. The $`c_2`$–term originates solely from the abnormal parity terms in the action, e.g. $`\mathrm{\Gamma }_{WZ}`$, while the $`c_3`$–term additionally acquires contributions from field components which are induced by the collective rotations. Both, $`c_2`$ and $`c_3`$ are subleading in $`1/N_C`$ as the appearance of the generators, $`R_a`$ suggests. A well–known problem of many chiral soliton models is the too small prediction for the axial charge of the nucleon, $`g_A`$ when the constants $`c_n`$ are computed using the soliton solution. In this section we will not address that problem but rather use the empirical value $`g_A=1.258`$ as an input to determine the $`c_n`$. That is, we consider the constants $`c_n`$ as free parameters, alike $`F`$&$`D`$ in the Cabibbo scheme.
It turns out that for pure octet wave–functions the matrix elements of the operators multiplying the constants $`c_1`$ and $`c_3`$ have the same ratio $`F/D=5/9`$ while the operator associated with $`c_2`$ has $`F/D=5/3`$. This suggests to put $`c_1+c_3/2=(3F+5D)/2`$ and $`c_2=(9F5D)/\sqrt{3}`$ with the empirical values $`g_A=F+D=1.258`$ and $`F/D=0.575\pm 0.016`$, i. e. $`c_1+c_3/22.69`$ and $`c_20.09`$. Of course, these relations are correct only for $`\omega ^2=0`$. To see that the parameterization of the axial current matrix elements in terms of $`F`$&$`D`$ Clebsch–Gordan coefficients becomes invalid already at moderate $`\omega ^2`$ we consider the ratios $`B|D_{a3}|B^{}/B|_{\alpha ,\beta =4}^7d_{3\alpha \beta }D_{a\alpha }R_\beta |B^{}`$ in figure 1.
The fact that the operators $`D_{a3}`$ and $`_{\alpha ,\beta =4}^7d_{3\alpha \beta }D_{a\alpha }R_\beta `$ have the same $`F/D`$ ratio is reflected by all ratios assuming the same value when flavor covariant wave–functions are used ($`\omega ^2=0`$). However, we see that already at moderate symmetry breaking the description of the axial current matrix elements in terms of $`F/D`$ ratios becomes inadequate as these operators evolve quite differently. With these significant dependencies on the effective symmetry breaking of matrix elements of the various operators contributing to the axial charges on $`\omega ^2`$ it seems difficult to imagine that the empirical results for the hyperon decays, which are well described by the symmetric formulation, can be reasonably reproduced at realistic $`\omega ^2\stackrel{>}{}5`$.
Before attempting such a fit we can get more insight into the relevance of the constants $`c_n`$ from the axial singlet current. Although it is not related to the octet current (10) by group theoretical means, the fact that within the collective coordinate approach we can consider flavor symmetry breaking as a continuous parameter provides further information. In the limit $`\omega ^2\mathrm{}`$ (integrating out strange degrees of freedom) the model should reduce to the two flavor formulation. In particular the strangeness contribution to the axial charge of the nucleon should vanish in that limit. Noting that $`N|D_{83}|N0`$ and $`N|_{\alpha ,\beta =4}^7d_{3\alpha \beta }D_{8\alpha }R_\beta |N0`$ while $`N|D_{88}|N1`$ for $`\omega ^2\mathrm{}`$, we demand
$`{\displaystyle d^3rA_i^{(0)}}=2\sqrt{3}c_2R_ii=1,2,3.`$ (11)
for the axial singlet current because it leads to the strangeness projection
$`{\displaystyle d^3rA_i^{(s)}}`$ $`=`$ $`{\displaystyle \frac{1}{3}}{\displaystyle d^3r\left(A_i^{(0)}2\sqrt{3}A_i^{(8)}\right)}`$ (12)
$`=`$ $`{\displaystyle \frac{2}{\sqrt{3}}}\left\{c_1D_{8i}+c_2(1D_{88})R_i+c_3{\displaystyle \underset{\alpha ,\beta =4}{\overset{7}{}}}d_{i\alpha \beta }D_{8\alpha }R_\beta \right\}.`$ (13)
Actually all model calculations in the literature are consistent with this requirement. It is simply a consequence of embedding the soliton in the isospin subgroup of flavor $`SU(3)`$. The analysis of the famous proton spin puzzle yielding $`\mathrm{\Delta }\mathrm{\Sigma }_N=N|d^3rA_i^{(0)}|N=0.20\pm 0.10`$ then suggests $`c_2=0.12\pm 0.06`$ in agreement with the above estimate from the flavor symmetric description of hyperon decays.
In order to completely describe the hyperon beta–decays we also demand matrix elements of the vector charges. These are obtained from the operator
$`{\displaystyle d^3rV_0^{(a)}}={\displaystyle \underset{b=1}{\overset{8}{}}}D_{ab}R_b=L_a,`$ (14)
which introduces the left $`SU(3)`$ generators $`L_a`$. Again, this relation is obtained by substituting (1) into the covariant expression for the vector current operator extracted from the model Lagrangian. The relevant matrix elements are protected by the Ademollo–Gatto theorem stating that deviations from the $`SU(3)`$ relations start at order $`(\omega ^2)^2`$. In the collective coordinate appraoch this theorem is reproduced as the slope of these curves vanishes at $`\omega ^2=0`$, cf. figure 1. Consequently, symmetry breaking in the vector currents is not only ignored in the Cabibbo scheme but also in the linear treatment of ref . However, for the strongly distorted wave–functions, which we are utilizing, the deviations from the $`SU(3)`$ relations is sizable as is clearly shown in figure 1. Of course, we will take into account these deviations when computing the vector charges.
We now attempt to determine the constants $`c_n`$ to reasonably fit the ratios $`g_A/g_V`$ for the hyperon beta–decays (only $`g_A`$ for $`\mathrm{\Sigma }^+\mathrm{\Lambda }e^+\nu _e`$). The values for $`g_A`$ and $`g_V`$ are obtained from the appropriate matrix elements of respectively the operators in eqs (10) and (14), sandwiched between the eigenstates of the full Hamiltonian (7). We first have to fix a value, $`\omega _{\mathrm{fix}}^2`$ for which we want to obtain the best fit. We adopt the following strategy: we choose $`c_2`$ according the proton spin puzzle and subsequently determine $`c_1`$ and $`c_3`$ at $`\omega _{\mathrm{fix}}^2=6.0`$ such that the nucleon axial charge, $`g_A`$ and the $`g_A/g_V`$ ratio for $`\mathrm{\Lambda }pe^{}\overline{\nu }_e`$ are reproduced. For example, setting $`\mathrm{\Delta }\mathrm{\Sigma }=0.2`$ yields $`c_1=1.97`$, $`c_2=0.12`$, and $`c_3=1.38`$. This is not too different from the above consideration in the symmetric case as $`c_1+c_3/2=2.66`$. The matrix elements for the $`np`$ and $`\mathrm{\Lambda }p`$ transitions enter this determination of the $`c_n`$. The comparision with figure 1 tells us that the deviations from the symmetric limit have turned out unexpectedly small. We are now left with predictions not only for the decay parameters of the other decay processes but we can also study the variation with symmetry breaking of all relevant decays. This is shown in figure 2.
Obviously the dependence on flavor symmetry breaking is very moderate, on the order of only a few percent. In view of the model being an approximation this dependence may be considered irrelevant and the results can be viewed as being in reasonable agreement with the empirical data, cf. table I. The observed independence of $`\omega ^2`$ shows that these predictions are not sensitive to the choice of $`\omega _{\mathrm{fix}}^2`$. In addition, since we observe this approximate independence of $`\omega ^2`$, we essentially have a two parameter ($`c_1`$ and $`c_3`$, $`c_2`$ is fixed from $`\mathrm{\Delta }\mathrm{\Sigma }_N`$) fit of the hyperon beta–decays. This is alike the $`F`$&$`D`$ parameterization in the Cabibbo scheme. We remark that the two transitions, $`np`$ and $`\mathrm{\Lambda }p`$, which are not shown in figure 2, exhibit a similar neglegible dependence on $`\omega ^2`$ and, by construction, they match the empirical data at $`\omega ^2=6.0`$. It should be noted that the use of the exact eigenfunctions of (7), which leads to the non–linear behavior is important in this regard. A linearized version (in $`\omega ^2`$) would not have necessarily yielded this result. In particular a first order description would fail for the process $`\mathrm{\Xi }\mathrm{\Sigma }`$, for which $`g_A/g_V`$ is a non–monotonous function of $`\omega ^2`$. Comparing the results shown in figure 2 with the data in table I we see that the calculation using the strongly distorted wave–functions agrees approximately as well with the empirical data as the flavor symmetric $`F`$&$`D`$ fit.
We also observe that the singlet current does not get modified. Hence we have the simple relation
$`\mathrm{\Delta }\mathrm{\Sigma }_N=\mathrm{\Delta }\mathrm{\Sigma }_\mathrm{\Lambda }`$ (15)
for all values of $`\omega ^2`$.
In figure 3 we display the flavor components of the axial charge of the $`\mathrm{\Lambda }`$ hyperon. We see that also the various contributions to the axial charge of the $`\mathrm{\Lambda }`$ only exhibit a moderate dependence on $`\omega ^2`$. The non–strange component, $`\mathrm{\Delta }U_\mathrm{\Lambda }=\mathrm{\Delta }D_\mathrm{\Lambda }`$ slightly increases in magnitude. The strange quark piece, $`\mathrm{\Delta }S_\mathrm{\Lambda }`$ then grows with symmetry breaking since we keep $`\mathrm{\Delta }\mathrm{\Sigma }_\mathrm{\Lambda }`$ fixed. It should be remarked that the results shown in figure 3 agree nicely with an $`SU(3)`$ analysis applied to the data : $`\mathrm{\Delta }U_\mathrm{\Lambda }=\mathrm{\Delta }D_\mathrm{\Lambda }0.20`$ and $`\mathrm{\Delta }S_\mathrm{\Lambda }0.60`$. Finally we remark that the observed independence on the symmetry breaking does not occur for all matrix elements of the axial current. An interesting counter–example is the strange quark component in the nucleon, $`\mathrm{\Delta }S_N`$. For $`\mathrm{\Delta }\mathrm{\Sigma }=0.2`$, say, it is significant at zero symmetry breaking, $`\mathrm{\Delta }S_N=0.131`$ while it decreases (in magnitude) to $`\mathrm{\Delta }S_N=0.085`$ at $`\omega ^2=6.0`$.
Within this class of models the order of the expansion in symmetry breaking is measured by differences like $`\omega ^2m_K^2m_\pi ^2`$, $`f_K^2f_\pi ^2`$ or $`m_K^{}^2m_\rho ^2`$ which are linear in the strange current quark mass $`m_s`$. In an systematic expansion one could add symmetry breaking components to the currents to eliminate the small deviations from the empirical data. As these corrections are potentially small it might well be that this could be accomplished by a single operator of $`𝒪(m_s^2)`$ or even higher. In turn this would make the approach quite unpredictable. In addition the errors in the empirical data (cf. table I) may penetrate to the fitted coefficients $`c_n`$. It seems thus more appropriate to revert to realistic models in which we can calculate the coefficients of the next–to–leading order terms and which have been tested at other instances.
## IV Spin Content of the $`\mathrm{\Lambda }`$ in a Realistic Model
We consider a realistic soliton model which contains pseudoscalar and vector meson fields. It has been established for two flavors in ref and been extended to three flavors in ref where it has been shown to fairly describe the parameters of hyperon beta–decay (cf. table 4 in ref ).
Starting point is a three–flavor chirally invariant theory for pseudoscalar and vector mesons. The model Lagrangian contains terms which involve the Levi–Cevita tensor $`ϵ_{\mu \nu \rho \sigma }`$, to accommodate processes like $`\omega 3\pi `$ . These terms contribute to $`c_2`$ and $`c_3`$. A minimal set of symmetry breaking terms is included to account for different masses and decay constants. This effective theory contains topologically non–trivial static solutions, which are constructed by imposing ansätze in the isospin subgroup
$`\xi (\stackrel{}{r})=\mathrm{exp}\left(\frac{i}{2}\widehat{\stackrel{}{r}}\stackrel{}{\tau }F(r)\right),\omega _0(\stackrel{}{r})=\omega (r)\mathrm{and}\rho _{i,a}(\stackrel{}{r})={\displaystyle \frac{G(r)}{r}}ϵ_{ija}\widehat{r}_j,`$ (16)
while all other field components vanish classically. Here $`\xi =\mathrm{exp}\left(i\stackrel{}{\pi }\stackrel{}{\tau }/2f_\pi \right)`$ refers to the non–linear realization of the pion fields. The radial functions are determined by extremizing the static energy functional subject to boundary conditions appropriate to unit baryon number. Introducing collective coordinates for this configuration induces field components which are classically absent. From this, eight real radial functions emerge. They solve inhomogeneous linear differential equations with the soliton profiles (16) acting as sources. In regard of the discussion in the preceding section it is interesting to note that despite of strong symmetry breaking in the baryon wave–functions the model predictions for the magnetic moments approximately obey the respective $`SU(3)`$ relations .
Covariant expressions for the (axial–)vector currents are obtained by introducing appropriate sources. Substituting the above described ansätze and applying the quantization rules for the collective coordinates yields the charges as linear combinations of functionals, $`c_n[F,\omega ,G,\mathrm{}]`$ of the meson profile functions and operators in the space of the collective coordinates $`A`$. In this model the derivative type symmetry breaking terms add symmetry breaking pieces to the axial charge operator,
$`\delta A_i^{(a)}=c_4D_{a8}D_{8i}+c_5{\displaystyle \underset{\alpha ,\beta =4}{\overset{7}{}}}d_{i\alpha \beta }D_{a\alpha }D_{8\beta }+c_6D_{ai}(D_{88}1)\mathrm{and}\delta A_i^{(0)}=2\sqrt{3}c_4D_{8i}.`$ (17)
The identical coefficient $`c_4`$ in the octet and singlet currents arises from the model calculations, it is not demanded by the above mentioned consistency condition of having vanishing strangeness contribution in the nucleon for $`\omega ^2\mathrm{}`$ since we find $`N|D_{88}D_{83}|N0`$ as well as $`N|D_{83}|N0`$.
Once the model parameters are agreed on, the coefficients $`c_1,\mathrm{},c_6`$ are uniquely determined as are the parameters in the collective Hamiltonian, which in this model is more involved than eq (7). Thus the baryon wave–functions as well as the current operators are fixed and all relevant decay parameters can be computed. Unfortunately the model parameters cannot be completely determined in the meson sector . We use the remaining freedom to accommodate baryon properties in three different ways as shown in table III. The set denoted by ‘b.f.’ refers to an overall best fit to the spectrum of the low–lying spin $`\frac{1}{2}`$ and spin $`\frac{3}{2}`$ baryons. It predicts the axial charge somewhat on the low side, $`g_A=0.88`$. The set named ‘mag.mom.’ labels a set of parameters which yields magnetic moments which are close to the respective empirical data (with $`g_A=0.98`$) and finally the set labeled ‘$`g_A`$’ reproduces the axial charge of the nucleon and also reasonably accounts for hyperon beta–decay .
For all three sets the effective symmetry breaking is sizable, $`\omega ^210`$. However, its effect is somewhat mitigated by additional symmetry breaking terms ($`_{i=1}^3D_{8i}R_i`$, $`_{\alpha =4}^7D_{8\alpha }R_\alpha `$) in the collective Hamiltonian (7). We observe that in particular the predictions for the axial properties of the $`\mathrm{\Lambda }`$ are quite insensitive to the model parameters. The variation of the model parameters only seems to influence the isovector part of the axial charge operator. Surprisingly the singlet matrix element of the $`\mathrm{\Lambda }`$ hyperon is smaller than that of the nucleon, although this effect is tiny. As this difference emerges solely from the $`c_4`$ term this ordering is a reflection of $`c_4`$ being positive in this model. It should be noted that in other models $`c_4`$ is predicted to be negative , although small in magnitude as well; suggesting that $`\mathrm{\Delta }\mathrm{\Sigma }_\mathrm{\Lambda }\mathrm{\Delta }\mathrm{\Sigma }_N`$ in general.
Similar to the fit of the previous section the full model calculation predicts sizable polarizations of the up and down quarks in the $`\mathrm{\Lambda }`$ which are slightly smaller in magnitude but nevertheless comparable to those obtained from the $`SU(3)`$ symmetric analysis. One wonders whether the significant up–quark polarization $`\mathrm{\Delta }U_\mathrm{\Lambda }=\mathrm{\Delta }D_\mathrm{\Lambda }0.16`$ has an experimental signature. The Gribov–Lipatov reciprocity relation suggests that the quark fragmentation functions follow the quark distribution functions. Although there is no direct connection this hypothesis may nevertheless serve as an estimate . As the integrated polarized distribution functions are just $`\mathrm{\Delta }Q_B`$ (for quark $`Q`$ inside baryon $`B`$) it is suggestive that the predicted value for $`\mathrm{\Delta }U_\mathrm{\Lambda }`$ goes along with a significant up–quark fragmentation function for the $`\mathrm{\Lambda }`$. In electroproduction the individual quark contributions to oberservables are weighted by the square of the respective charges. For the $`\mathrm{\Lambda }`$ this elevates the non–strange contribution by a factor five. Equations (2)–(4) in ref give the polarization of a $`\mathrm{\Lambda }`$ that is produced in the current fragmentation region of deep inelastic electron–proton scattering. In addition to the $`(\mathrm{charge})^2`$ factor the strange–quark contribution is suppressed in such processes as its distribution in the nucleon is presumably small. In essence the significant negative prediction for $`\mathrm{\Delta }U_\mathrm{\Lambda }`$ should result in a negative polarization of $`\mathrm{\Lambda }`$’s produced in that a reaction. In view of the self–analyzing decay $`\mathrm{\Lambda }p\pi ^{}`$ this should be detectable. However, it is subject to the (reasonable) assumption that fragmentation and distribution functions are closely related.
## V Conclusions
In the collective coordinate approach to chiral solitons large deviations from flavor symmetric (octet) wave–functions are required to accommodate the observed pattern of the baryon mass–splitting. Especially, contributions which arise beyond next–to–leading order in the effective symmetry breaking are needed for this purpose. In the QCD language these are of the order $`m_s^2`$ or higher. In this report we have suggested a picture for the axial charges of the low–lying $`\frac{1}{2}^+`$ baryons which manages to reasonably reproduce the empirical data without introducing (significant) flavor symmetry breaking components in the corresponding operators. Rather, the sizable symmetry breaking resides almost completely in the baryon wave–functions. This scenario is especially motivated by the Yabu–Ando treatment of the Skyrme model which has the major symmetry breaking components in the potential part of the action and thus no (or only minor) symmetry breaking pieces in the current operators. The empirical data for these decay parameters are as reasonably reproduced as in the Cabibbo scheme of hyperon beta–decay. Repeatedly we emphasize that the present picture is not a re–application of the Cabibbo scheme since in the present calculation the ‘octet’ baryon wave–functions have significant admixture of higher dimensional representations (cf. table II). Furthermore the individual matrix elements which enter this calculation may strongly vary with the effective symmetry breaking (or strange current quark mass), cf. figure 1; only when combining them to the full $`g_A/g_V`$ ratios the strong dependence on the strength of symmetry breaking cancels.
In the present treatment we may consider symmetry breaking as a continuous parameter. Taking this parameter to be infinitely large the two flavor model must be retrieved for the nucleon. This consistency condition relates coefficients in the axial singlet current operator to the respective octet components, which are not otherwise related to each other by group theory. In turn we are enabled to completely disentangle the quark flavor components of the axial charge. It results in sizable up and down quark polarizations in the $`\mathrm{\Lambda }`$. Again, a picture emerged which, after some cancellations, agrees with that of the flavor symmetric treatment for known data. These results were obtained utilizing a parameterization of a charge operator which did not contain any symmetry breaking component.
We have also considered a realistic model, wherein the parameters entering the charge operators are actually predicted. These operators contain non–vanishing symmetry breaking pieces, whose matrix elements are, however, small. Essentially this model calculation confirmed the results obtained in the parametrically treatment.
### Acknowledgments
The author would like to thank G. R. Goldstein, R. L. Jaffe and J. Schechter for helpful conversations and useful references.
This work is supported in part by funds provided by the U.S. Department of Energy (D.O.E.) under cooperative research agreement #DF-FC02-94ER40818 and the Deutsche Forschungsgemeinschaft (DFG) under contract We 1254/3-1.
|
warning/0005/hep-th0005255.html
|
ar5iv
|
text
|
# 1. Introduction
## 1. Introduction
The electroweak Standard Model (SM) has been extremely successful in the interpretation of LEP/SLC data and higher order effects typically amount to 10 $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\sigma }$}`$ deviations if not taken into account . These precise predictions are only possible due to the renormalizability of the SM and the by now very precise knowledge of the relevant input parameters. Last but not least the relevant coupling constants are small enough such that perturbation theory mostly works very well.
The formal proofs of renormalizability of the SM often relied on the assumption that a gauge invariant regularization exists. The question whether such a regularization exists is non–trivial because of the chiral structure of the fermions involved. At present the only regularization, which makes elaborate computations of radiative corrections feasible, is the dimensional regularization (DR) scheme which is well-defined for field theories with vectorial gauge symmetries only. However, in theories exhibiting chiral fermions, like the electroweak SM, problems with the continuation of the Dirac matrix $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ to dimensions $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$ remain open within this context and several modifications of the ’t Hooft–Veltman DR have been proposed . It turns out that starting from the standard SM-Lagrangian and using a $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$, which does not anticommute with the other Dirac matrices $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}`$, leads to “spurious anomalies” which violate chiral symmetry and hence gauge invariance. These anomalies would spoil renormalizability if we would not get rid of them by imposing “by hand” the relevant Ward-Takahashi (WT) and Slavnov-Taylor (ST) identities order by order in perturbation theory . At first sight this might not look to be a serious problem, however, violating the symmetries of the SM makes practical calculations much more difficult and tedious than they are anyway.
The problems of course are related to the existence of the Adler–Bell–Jackiw (ABJ) anomaly , which must cancel in the SM in order not to spoil its renormalizability .
Surprisingly, the prescriptions proposed and/or used by many authors continue to be controversial , and hence it seems to be necessary to reconsider the problem once again. We shall emphasize, in particular, the advantage of working with chiral fields. The consequences of working as closely as possible with chiral fields, it seems to me, has not been stressed sufficiently in the literature so far.
As a matter of principle it is important to mention two other approaches which both work in $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$ dimensions. i) In quantum field theories on the lattice a recent breakthrough was the discovery of exact chiral invariance on the lattice which circumvents the Nielsen–Ninomiya no–go theorem . A well defined regularization which preserves simultaneously chiral–and gauge–symmetries is thus known and could be applied to the SM. ii) The algebraic renormalization of the electroweak SM to all orders within the Bogoliubov-Parasiuk-Hepp-Zimmermann (BPHZ) framework is a mathematically well defined scheme, which is much more involved because it breaks the symmetries at intermediate stages and hence leads to much longer expressions which are extremely tedious to handle in practice. In cases of doubt this is the only known scheme which is free of ambiguities and works directly in 4–dimensional continuum field theory.
For perturbative calculations in the continuum we have to stick as much as possible to the more practical route of dimensional regularization. In the following tensor quantities in $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$ dimensions are supposed to be defined by interpolation of $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{n}$}`$ ($`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{n}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}`$, integer ) dimensions to dimensions below $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$. It is well known that the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}`$–algebra, the so called “naive dimensional regularization” (NDR) <sup>*</sup><sup>*</sup>*I read it as “normal dimensional regularization”
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{𝟏}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}$$
(1)
for dimensions of space–time $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}`$, $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$ is inconsistent with
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{Tr}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\rho }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\sigma }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(2)
The latter condition is often considered to be necessary, however, for an acceptable regularization since at $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$ we must find
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{Tr}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\rho }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\sigma }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{i}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\epsilon }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\rho }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\sigma }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(3)
Generally, for $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ odd traces one obtains trace conditions from the cyclic property of traces. They are not fulfilled automatically, as we shall see, and hence the algebra is ill-defined in general. Considering $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{Tr}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\underset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{j}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}}{\overset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{j}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ cyclicity requires
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{Tr}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\underset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{j}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}}{\overset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{j}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\underset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}}{\overset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{Tr}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\underset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{j}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{j}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}}{\overset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{j}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(4)
Contraction with the metric tensor $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}}`$ yields
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{Tr}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\underset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{j}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\overset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{j}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{Tr}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\underset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{j}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\overset{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{j}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}$$
(5)
with $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$. Thus $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$ together with (2) implies $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$. However, non-anti-commutativity of $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ is in conflict with the chiral structure and hence with gauge invariance of the SM, in general. It is the purpose of this note to study the possibility of restoring gauge invariance by employing chiral fields systematically.
## 2. Formally gauge invariant Feynman rules
Obviously only terms involving $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}`$ in the standard SM Lagrangian can be affected by a non–anticommuting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$. As an example we consider the leptonic part, given by
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}`$ $`\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{R}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{B}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{R}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{R}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{R}}$}`$ (6)
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}`$ $`\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{L}$}}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}{\displaystyle \frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{B}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}{\displaystyle \frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\tau }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{a}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{W}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{a}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{L}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}`$
using standard notation. As usual the chiral fields
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{R}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{R}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{L}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\begin{array}{c}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\\ \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\end{array}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{L}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\begin{array}{c}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\\ \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\end{array}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}`$ (11)
may be represented in terms of the lepton fields $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{x}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ and the neutrino field $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{x}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ with the help of the chiral projectors
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\pm }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{𝟏}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\pm }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(12)
In order that $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\pm }$}`$ are Hermitean projection operators $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ must have the properties
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{𝟏}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(13)
Furthermore, we demand $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\pm }$}`$ to be chiral projectors also for the adjoint $`\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$ of a Dirac field $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}`$. This implies
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(14)
By Lorentz covariance in the 4–dimensional physical subspace the latter condition extends to
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{for}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(15)
It is easy to verify that $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}`$ is invariant under local $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{S}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{U}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{L}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{U}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{Y}}$}`$ gauge transformations, irrespective of $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$. Since the chiral fields have the simple transformation properties
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{L}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{exp}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\tau }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{a}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\omega }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{a}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{L}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{exp}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\tau }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{a}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\omega }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{a}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{L}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}`$
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{R}}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{exp}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{R}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{exp}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{R}}$}`$
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{R}}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{R}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}`$ (16)
the invariance of $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}`$ follows immediately from the properties of $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\pm }$}`$ alone.
We notice that in utilizing chiral fields there seems to be no conflict with the non–anti-commutativity of $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ and the formal validity of the ST–identities.
Usually, one prefers to write Feynman rules in terms of the Dirac fields $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}`$ and $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}`$. The standard Feynman rules are obtained using the relations
$$\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\pm }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\pm }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}$$
(17)
which are valid only, provided $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$.
If $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$ in $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$ dimensional space–time, the above relations no longer hold and hence the standard Feynman rules manifestly violate gauge invariance. The correct relations, replacing (17), read
$$\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\pm }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\pm }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}$$
(18)
with
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}$$
(19)
and
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{[}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{]}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(20)
We notice that by definition all $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}`$’s are anticommuting with $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(21)
According to (18) the proper expressions for the vector current and for the axial–vector current read
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{V}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{x}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}$$
(22)
and
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{x}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}$$
(23)
respectively. It might be worthwhile to point out that the standard form of the axial current $`\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}`$ is not Hermitean when $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$. The above consideration also shows how anomalies may come about in the vector current when $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}`$.
The fermion kinetic term changes to
$$\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(24)
Correspondingly, the free massless fermion fields must satisfy the field equation
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(25)
This formally implies that the conserved canonical Noether currents are precisely the ones given above.
By the field equation the fermion spinors satisfy
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}{\displaystyle \frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{u}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{s}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}{\displaystyle \frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{v}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{s}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$ (26)
and the free fermion propagator reads ($`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$)
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{S}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{K}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}}$$
(27)
with
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{K}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(28)
Formally, we have obtained chiral and gauge invariant Feynman rules for non-anticommuting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$. Eqs. (22), (23) and (27) replace the standard expressions valid for $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$.
## 3. Non-existence of a chirally invariant DR
The gauge invariant Feynman rules presented in the preceding section do not permit a regularization by continuation in the dimension $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}`$ when $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ is chosen compatible with the trace condition (2). This can be proven as follows. First we consider the Dirac algebra extended to $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{n}$}`$ ($`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{n}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}`$, integer). In this case $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{n}$}`$–dimensional representations of the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}`$–algebra are well known . A basis for the algebra is given by the set of matrices $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{𝟏}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ and the antisymmetrized products $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{[}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{]}$}}`$ associated with $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}`$–dimensional subspaces of $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{M}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}`$. We will split the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{S}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{O}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ vectors (tensors) into 4–dimensional vectors $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\widehat{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$, in the physical subspace $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{M}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$, and their orthogonal complements $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$. If we impose the trace condition (2) in the physical subspace (see Eq. (15) above) we obtain the ’t Hooft–Veltman algebra :
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\begin{array}{cc}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{;}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\hfill \\ \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{;}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\hfill \end{array}$$
(29)
with $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{!}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\epsilon }$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\rho }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\sigma }$}}\widehat{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\widehat{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\widehat{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\rho }$}\widehat{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\sigma }$}`$. Here, it is important to notice that $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ is a matrix of rank $`\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$. The matrix–elements themselves are of order $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{O}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$. As a consequence higher products of AC-terms are not of higher order in $`\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}}`$ for $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$. This is the reason why the extra terms needed to restore the Ward-Takahashi identities cannot be considered as perturbations. They affect the free part of the Lagrangian! and hence the form of the fermion propagators, as shown above. The symmetry at the end can only be there if the free and the interacting parts of the Lagrangian match appropriately.
We are now ready to reconsider the fermion propagator (27). Using (29), we get for the scalar product (28)
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{K}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\widehat{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}$$
(30)
and thus
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{S}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\frac{\widehat{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}}{\widehat{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}}$$
(31)
takes its 4–dimensional form, independent of $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}`$! It is then impossible to regularize fermion-loop integrals by continuation in $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}`$. The crucial point is that the consistency with the trace condition requires that in (28) the extra term proportional to $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}`$ like AC is a matrix of rank $`\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$ and not a correction of order $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{O}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ in the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}`$–expansion!
The problem may be reconsidered in terms of the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}`$–algebra defined by (19), which may be associated to any $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}`$–algebra:
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{𝟏}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}{\displaystyle \frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}`$ (32)
For any $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}`$–algebra in order to be closed, we must require
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{G}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{𝟏}}$}`$ (33)
for some symmetric $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\times }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}`$–matrix $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{G}$}`$, which satisfies
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{G}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\rho }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{G}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\rho }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(34)
The trace condition (4) must hold with the replacements
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{G}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}$$
(35)
which implies
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{G}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{G}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(36)
Assuming $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{G}$}`$ to have block–diagonal form
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{G}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\begin{array}{cc}\widehat{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\text{0}}$}\\ \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\text{0}}$}& \overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}}\end{array}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}$$
(37)
the condition (2) can be satisfied with a singular metric $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{G}$}`$ only:
$$\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{G}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\widehat{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}}$$
(38)
where $`\widehat{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{g}$}}`$ must be the Minkowski metric. Thus, starting from the ’t Hooft–Veltman scheme, we are lead to a dimensional reduction (DRED) scheme by adding just some terms in the Feynman rules which vanish in $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$.
As a result, the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}`$–form of the ’t Hooft–Veltman algebra is identical to the 4–dimensional Dirac algebra. In other words, using the ’t Hooft–Veltman algebra (in its $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}`$–dimensional form) together with the chiral fields, which are adapted to the gauge symmetry, “non-regularization” of fermion–loops is implied. Again, a regularization can only be obtained by giving up either the trace condition (2) or gauge invariance.
This last statement, of course, is not terribly new. What we have shown is that the Dirac algebra assuming anticommuting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ on the one hand and the ’t Hooft–Veltman algebra on the other hand are not really different, since the latter can always be rewritten in the anticommuting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}`$–form by means of the relations (19) and (20). In any case, for theories involving $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$, “dimensional regularization” compatible with (4), does not provide well–defined integrals for loops involving fermion lines. This has been noticed by ’t Hooft and Veltman in their original paper where they state: “the usual ambiguity of choice of integration variables is replaced in our formalism by the ambiguity of location of $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ in the trace”. Statements to the contrary, frequently found in the literature, are misleading. Usually, extra “prescriptions” about where to put the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ in a particular calculation are proposed. These prescriptions, however, do not resolve the problem of mathematical inconsistencies, i.e., they still require an explicit check and the restoration of the Ward-Takahashi identities.
The use of chiral fields provides an unambiguous rule for the proper location of the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$–matrices before generalization to $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$. Unfortunately, this has lead to the “non-regularization” by dimensional continuation when the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$ trace condition (2) is imposed, which in turn essentially implies the ’t Hooft–Veltman scheme.
If we violate gauge invariance by the naive application of the ’t Hooft–Veltman prescription, we have to restore the symmetry by imposing the relevant Ward–Takahashi identities and fixing appropriate counter terms. But this precisely amounts to including the extra $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ terms given in Eqs. (22) and (23). Which in turn is nothing but another way of utilizing the naive anticommuting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$.
## 4. Conclusion for the practitioner
According to our considerations above we are left with two possible strategies:
### i) $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$: the chirally improved ’t Hooft–Veltman scheme
If we insist on the trace condition (2) the gauge invariance must be manifestly broken in order to obtain the “pseudo regularization” by dimensional continuation. Again we start at the level of the chiral fields but must avoid the non–regularization by treating the AC–terms in the free part of the Lagrangian as interaction terms, i.e., we use the standard $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}`$–dimensional Fermi propagator
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{S}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}}$$
(39)
together with the chiral currents (22, 23) as our “chiral Feynman rules”. Since $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$, the choice of the Fermi propagator (39) amounts to adding the symmetry breaking term
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Delta }}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{SB}}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\psi }$}$$
(40)
to the Lagrangian. Besides the fact that this operator has no 4–dimensional representation, it is not a higher order term for $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$ as it would be necessary for treating it as a counter-term (perturbation). Expanding $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Delta }}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{SB}}$}}`$ perturbatively amounts to the assumption that $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{O}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ in the sense of matrix elements, which conflicts with (2). As we have mentioned earlier, (2) requires $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ to be a matrix of rank $`\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$ with matrix elements of order $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{O}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$. A mathematically satisfactory way out of the dilemma within the framework of DR is not possible as a result of the existence of the ABJ–anomaly.
Our considerations show that “quasi gauge invariant” Feynman rules may be obtained for non-anticommuting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ provided $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ is treated as a perturbation i.e. $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{O}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$. Examples are briefly considered in the Appendix. Results turn out to be AC–independent in this case. AC–invariance may be used as a helpful tool for checking the gauge invariance of fermionic loop contributions to amplitudes. Usually such checks are possible only by explicit consideration of WT- and/or ST-identities. We stress, once again, that any approach which treats the AC–term as a perturbation conflicts with the trace condition (2) at some point. Ignoring this point leads to “standard” confusions, frequently appearing in the literature. While working with the ’t Hooft–Veltman prescription in the standard form requires the subsequent check of the Ward-Takahashi identities,after utilizing the chiral version of the Feynman rules we may restrict ourselves to check the hard anomaly diagrams.
Since amplitudes exhibiting spurious anomalies only may be chiralized either by our chirally improved Feynman rules or by imposing the Ward-Takahashi identities which makes them AC–invariant we obviously may directly choose the scheme $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$, which is our second and preferred option:
### ii) $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$: the quasi self-chiral scheme
From a practical point of view an acceptable computational scheme should avoid spurious anomalies in the first place. This is possible only if the trace condition (2) is given up. Gauge invariance can be preserved then by using an anticommuting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$. This has been noticed in (see also ).
We observe that taking chiral fields seriously on a formal level, the only consistent way to avoid the above non-regularization is the simple one: use anticommuting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ from the very beginning, i.e., choose the NDR algebra (1). Since $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Gamma }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}`$ in this case we do not get the non-regularization of the fermion propagators. The ABJ–anomaly must be considered separately as we are going to discuss now The terminology introduced in which calls a scheme “consistent” if it respects the trace condition (2) and “inconsistent” otherwise is definitely misleading by the considerations presented in this paper. Since we cannot satisfy the Ward-Takahashi identities and the trace condition simultaneously we have the choice which one we want to consider more fundamental. Something has to be restored at the end by hand in any case. To put into place the model independent ABJ–anomalies, is by far simpler, than restoring the chiral symmetry which is broken by non–NDR schemes..
In the gauge invariant approach, closed fermion loops exhibiting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ odd traces and hard anomalies, cannot be obtained by dimensional continuation, merely, $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ odd traces are to be considered as intrinsically 4-dimensional quantities. Since charge conjugation properties and the related Bose symmetry are not automatically satisfied one has to account left- and right-circulation of the fermions in closed loops separately. In any case Adler’s approach can be utilized to resolve the remaining ambiguities. For this purpose, let us briefly consider the ABJ–anomaly exhibited by the current correlator $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{<}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{|}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{T}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{V}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{x}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{V}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{x}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{y}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{|}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{>}$}`$ of two vector currents and an axial–vector current. The one–loop diagrams are shown in Fig. 1.
In $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$, working as usual in momentum space, we may perform a covariant decomposition of the third rank pseudotensor which depends on the two independent momenta $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}`$ and $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}`$:
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\epsilon }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}`$ (41)
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\epsilon }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}`$
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\epsilon }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{6}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}`$
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\epsilon }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{7}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{8}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}`$
where the amplitudes $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}`$ are Lorentz scalars. We now impose
* Bose symmetry (i.e. consider the sum of the two diagrams of Fig. 1):
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}$$
which implies
$`\begin{array}{cccccc}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{6}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\\ \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{7}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}& \colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{8}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}\end{array}`$ (44)
* Vector current conservation:
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}$$
which implies
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}`$
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{6}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}`$ (45)
We thus find that the amplitudes $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}`$ and $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}`$ are determined uniquely in terms of the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}`$, $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{6}$}`$. The crucial observation, made by Adler long time ago , is that the amplitudes $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}`$, $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{8}$}`$, have dimension $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{d}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{eff}}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}`$ and hence are represented by convergent integrals. In contrast, $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}`$, $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{i}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}`$, have dimension $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{d}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{eff}}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$ (logarithmically divergent) and thus require regularization and renormalization. However, imposing Bose symmetry and vector current conservation uniquely determines the two regularization/renormalization dependent amplitudes in terms of the other convergent and hence unambiguous ones, i.e., the result is unique without need to refer to a specific renormalization scheme. The divergence of the axial–vector current takes the form
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\lambda }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{R}$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{8}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\pi }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\epsilon }$}^{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\alpha }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\beta }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\nu }$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}$$
where the first term on the r.h.s. is the normal term which vanishes for vanishing fermion mass $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}`$ while the second term is the mass independent anomaly. Formal axial–vector current conservation in the limit of vanishing fermion mass would require
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{7}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{8}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{?}}$}$$
with $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}`$ and $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}`$ fixed already by vector current conservation, this expression as we know does not vanish but yields the famous axial–vector current anomaly. All true anomalies, i.e.,quantum effects like the triangle anomaly which cannot be removed by adding a corresponding counter term to the Lagrangian, are well known to be related to the triangle diagram. Besides the triangle diagram itself they appear by tensor reduction from one–loop box and pentagon diagrams and diagrams which contain the one–loop anomalous graphs as subgraphs.
The Adler–Bardeen non-renormalization theorem of the one–loop anomalies implies that matters are under control provided Bose symmetry and vector current conservation are imposed, if necessary by hand. In DR it has been reconsidered in . Last but not least we must have the anomaly cancelation, possible by virtue of the quark lepton duality, in order to have the SM renormalizable .
Summary: we have shown that different $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$–schemes may be related by adding suitable terms in the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}`$–dimensional Lagrangian which vanish at $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$. In any scheme we can mimic chiral fields by the appropriate choice of the Feynman rules. We consider this to be crucial since the physical SM derives via a Higgs mechanism from a symmetric phase which exhibits chiral fermions only. The corresponding “chiral completion” (see (22,23)) of the Feynman rules cannot make a consistent scheme inconsistent or vice versa. Avoidable (often called “spurious”) anomalies are then absent. Our arguments strongly support the application of the NDR scheme (1), i.e., the $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}`$–dimensional $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}`$–algebra together with a strictly anticommuting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$, together with the simple 4–dimensional treatment of the hard anomalies discussed above. The NDR is easily implemented into computer codes and is by far the most convenient and efficient approach in calculations of radiative corrections. Removable anomalies are avoided and hence a tedious procedure of restoration of WT- and ST-identities is not needed.
The rules advocated here have been utilized successfully in the last twenty years by many authors at the one– and the two–loop level and beyond. Most SM calculations of higher order effects adopted the NDR scheme without encountering any inconsistencies. Of course, the NDR scheme has been advocated by several authors (see also ) in the past. I hope the present paper contributes to clarify part of the ongoing controversy.
Acknowledgements
Part of the ideas presented here have been developed a long time ago at the begining of the ongoing collaboration with Jochem Fleischer. I gratefully acknowledge in particular his involvement in the sometimes tedious explicit checks we have performed for many SM calculations. I also thank Christopher Ford and Oleg Tarasov for helpful discussions and for carefully reading the mansuscrpt.
## Appendix: <br>Calculations with $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$ in the SM: Two examples.
We have verified explicitly that all spurious anomalies disappear from fermion propagators and fermion form factors at one-loop order for the case where we use Feynman rules as proposed in Sec. 4. in case $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$. As explained earlier, in order to avoid the “non-regularization” of fermion lines, we must treat AC as a perturbation $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{O}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ and work to linear order in AC. All calculations have been performed in the ’t Hooft gauge with an arbitrary gauge parameter $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\xi }$}`$, which makes possible direct analytical checks of gauge invariance. We only summarize the structure of the results.
The irreducible self-energy $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Sigma }}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ we obtained has the following form
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Sigma }}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{[}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{]}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{B}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{C}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(46)
This implies that the mass- and wave-function renormalization are completely AC–independent:
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\delta }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{c}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\sqrt{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{Z}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{a}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{b}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(47)
Here the wave-function renormalization constant is given by the matrix
$$\sqrt{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{Z}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\sqrt{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{Z}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{R}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\sqrt{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{Z}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{L}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}}$$
(48)
where $`\sqrt{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{Z}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{R}$}}`$ and $`\sqrt{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{Z}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{L}$}}`$ are the independent wave–function renormalizations of the right–handed and left–handed fields, respectively. Thus the renormalized self–energy reads
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Sigma }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{r}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{a}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{[}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{]}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{B}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{b}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{C}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{c}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}$$
(49)
with $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{a}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$, $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{B}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{b}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$ and $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{C}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{c}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}`$ finite, and hence
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Sigma }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{r}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{a}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{B}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{b}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{C}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{c}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{0}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{O}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}$$
(50)
By contrast, using standard Feynman rules, we obtain
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Sigma }}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{A}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}{\displaystyle \frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{[}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{]}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{B}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{C}$}`$ (51)
$`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}`$ $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{D}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{[}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{]}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{E}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{[}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{k}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{/}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{]}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}`$
for the bare self-energy. In this case it is not possible to perform the renormalization in the standard way without imposing the Ward-Takahashi identities first, which must lead to the form (46).
Similar results can be found for form-factors. The following applies to the $`\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}`$ and $`\overline{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{Z}$}`$ vertices. The general form of the irreducible vertices reads
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{[}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{]}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{.}$}$$
(52)
We notice that the only surviving AC–term is $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ which appears in the canonical from (19) as in the Born term. Thus the vertex renormalization can be performed in an AC–independent way, i.e., the renormalized vertex is given by
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{r}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\left(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{m}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\right)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{c}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\frac{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}}{\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{[}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{]}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{c}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}$$
(53)
with $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{c}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}`$, $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{c}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}`$, $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}`$ and $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}`$ finite. Hence, we have
$$\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{\Pi }}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{r}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{=}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{c}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{c}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{1}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{3}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{p}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{2}$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{F}$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{4}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{+}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{O}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{ϵ}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}$$
(54)
independent of any AC–term. In contrast, by applying standard Feynman rules, we find additional terms of the form $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$, $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{[}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{]}$}`$ and $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\{}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}^\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mu }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{,}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\mathrm{AC}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{(}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{)}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\}}$}\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ which cannot be removed by renormalization, unless we impose the Ward-Takahashi identities first. In the chiral scheme we obtain gauge invariant form factors directly without imposing Ward-Takahashi identities by hand. Calculations in this “chiral” scheme in fact look very similar to the ones performed with anticommuting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$.
As a result of these findings we decided to work with an anti-commuting $`\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{\gamma }$}_\colorbox[rgb]{1,1,1}{$\textcolor[rgb]{0,0,0}{5}$}`$ henceforth, first at the one–loop level , later at the two–loop level . In most of these calculations we worked in the ’t Hooft gauge with a free gauge parameter which allowed us to check explicitly the gauge invariance of on-shell matrix elements.
|
warning/0005/hep-th0005086.html
|
ar5iv
|
text
|
# 1 Introduction and summary
## 1 Introduction and summary
The paper is devoted to a special multiplet of four dimensional N=2 supersymmetry containing two real scalar fields, two 2-form gauge potentials and two Weyl fermions. This multiplet is particularly relevant to four dimensional type IIB superstring vacua with N=2 supersymmetry because there the dilaton is a member of such a multiplet, see e.g. . The multiplet has been termed double tensor multiplet in the literature. However, to my knowledge, it has not been constructed explicitly up to know. This might have to do with the failure of an off-shell construction. In fact, one may be tempted to expect that the multiplet, with the above-mentioned field content, exists off-shell since the number of bosonic and fermionic degrees of freedom balance off-shell. However, as shown in section 2, such an off-shell realization of the N=2 supersymmetry algebra (possibly with a central charge, and modulo gauge transformations) is not compatible with the free action for the minimal field content.
Of course, this does not exclude that an off-shell formulation with additional (auxiliary) fields may exist but this question will not be addressed here. We shall thus work with the minimal field content and derive in section 2 the N=2 supersymmetry transformations for the free action. The N=2 supersymmetry algebra is realized on-shell modulo gauge transformations of the 2-form gauge potentials, without a central charge. The gauge transformations that occur in the algebra depend explicitly and at most linearly on the spacetime coordinates.<sup>1</sup><sup>1</sup>1The occurrence of gauge transformations involving explicitly the spacetime coordinates can be understood from a duality relation between the double tensor multiplet and the hypermultiplet. This will be explained in more detail and generality in a separate work. In terms of N=1 multiplets, the double tensor multiplet consists of two linear multiplets.
Having derived the free double tensor multiplet, we then study its interactions. Apart from self-interactions, the interactions with N=2 vector multiplets and hypermultiplets are analysed because of their relevance in the string theory context.<sup>2</sup><sup>2</sup>2The interactions with the N=2 supergravity multiplet are not discussed here. We impose that the action of the interacting theory be Poincaré invariant, gauge invariant and N=2 supersymmetric, but with possibly modified gauge and supersymmetry transformations as compared to the free theory. More precisely, we will study Poincaré invariant nontrivial continuous deformations of the free theory for one N=2 double tensor and an arbitrary number of vector multiplets and hypermultiplets. The analysis uses a systematic approach which is based on an expansion in the deformation parameters (coupling constants) and is briefly reviewed in section 3. The first order deformations must be nontrivial on-shell invariants of the free theory, i.e., field polynomials which are invariant on-shell (modulo total derivatives) under the gauge and supersymmetry transformations of the free theory.
These on-shell invariants can be determined at each mass dimension separately (assigning the standard dimensions to the fields). In section 4 all such invariants with dimensions $`5`$ are determined. It turns out that there are no such invariants with dimensions $`4`$; in particular, there are thus no nontrivial power counting renormalizable couplings of one double tensor multiplet to vector or hypermultiplets at all. All nontrivial couplings with dimension 5 are trilinear in the fields. There are three different types of such couplings:
Type A: Self-couplings of the double tensor multiplet. They contain Freedman-Townsend interaction vertices .
Type B: Couplings of the double tensor multiplet to vector multiplets. These couplings are linear in the fields of the double tensor multiplet and in the fields of different vector multiplets; hence, couplings of this type involve at least two vector multiplets. They contain interaction vertices of the Henneaux-Knaepen type .
Type C: Couplings of the double tensor multiplet to hypermultiplets. These couplings are linear in the fields of the double tensor multiplet and quadratic in the fields of one or more hypermultiplets.
In section 5 we study whether these dimension 5 interaction vertices can be extended to higher orders in the deformation parameter(s). To that end we first reformulate the free double tensor multiplet by introducing an appropriate set of auxiliary fields. Then we point out a remarkable property common to all three types of first order deformations and discuss its consequences for the structure of the supersymmetry deformations. Finally we treat explicitly three examples, one for each coupling type described above.
The first example is the simplest one and arises from a type C coupling between the double tensor multiplet and one hypermultiplet. We complete this coupling to all orders in the deformation parameters. In this case the gauge transformations do not get deformed. In contrast, the supersymmetry transformations get deformed, without changing the supersymmetry algebra (the commutator of two deformed supersymmetry transformations is an ordinary translation plus a gauge transformation on-shell). The auxiliary fields mentioned above allow one to give the deformed action and supersymmetry transformations in a compact polynomial form. Upon elimination of the auxiliary fields, the deformed action and the supersymmetry transformations become non-polynomial in the deformation parameters and in the scalar fields of the hypermultiplet, but remain local (in fact, each term in the action is at most quadratic in derivatives of fields and the supersymmetry transformations of the elementary fields contain at most one derivative).
The second example arises from a coupling of type B between the double tensor multiplet and two vector multiplets. It is somewhat reminiscent of the N=2 supersymmetric gauge theories with the vector-tensor multiplet constructed in where the central charge of that multiplet was gauged (even though there is no central charge in the present case). In this case both the supersymmetry transformations and the gauge transformations get nontrivially deformed. Again, the supersymmetry algebra does not change: the commutator of two deformed supersymmetry transformations is a translation plus a deformed gauge transformation on-shell. As in the first example, the complete deformations of the action, gauge and supersymmetry transformations are given in a compact polynomial form using the auxiliary fields. Upon elimination of the auxiliary fields, the deformations of the action and symmetry transformations become non-polynomial in the deformation parameters and in the scalar and vector fields of the vector multiplets, but remain local.
As a third example we discuss the self-interactions of type A. Their completion to all orders is more involved and is not fully accomplished. However, the first order deformations of the gauge and supersymmetry transformations and the first and second order deformation of the action are computed explicitly in the formulation with the auxiliary fields. The result implies already that, in the formulation without auxiliary fields, the action and symmetry transformations would be non-polynomial in the deformation parameters and in the scalar fields and the 2-form gauge potentials. Furthermore it strongly suggests that all higher order deformations exist as well and allows one to guess the structure of the resulting full action and symmetry transformations.
Of course it should be stressed that these three examples are relatively simple N=2 supersymmetric gauge theories involving the double tensor multiplet. There may be more complicated models of this type. In particular, instead of discussing the couplings of type A, B or C separately, one may study linear combinations of them and investigate whether such linear combinations can be completed to higher orders. In fact, N=1 supersymmetric models of this more complicated type have been constructed in and this suggests that analogous N=2 supersymmetric models exist as well. Section 6 comments on the use of the N=1 superfield construction in in the present context. One may go even further and study whether couplings of type A, B or C can be combined with the well-known couplings relating only vector multiplets and hypermultiplets. Of course, there are many more open questions, such as the classification of first order interaction terms with dimensions $`6`$, and the coupling of the double tensor multiplet to N=2 supergravity.
## 2 The free double tensor multiplet
#### General ansatz.
The starting point is the standard free action for the above-described field content,
$$d^4x(_\mu a^i^\mu a^iH_\mu ^iH^{\mu i}\mathrm{i}\chi \overline{\chi }\mathrm{i}\psi \overline{\psi })$$
(2.1)
where the $`a^i`$ are the two real scalar fields ($`i=1,2`$), $`\psi `$ and $`\chi `$ are the two 2-component Weyl fermions (their complex conjugates are denoted by $`\overline{\psi }`$ and $`\overline{\chi }`$), and $`H_\mu ^i`$ are the Hodge-duals of the field strengths of the real 2-form gauge potentials $`B_{\mu \nu }^i`$,
$$H^{\mu i}=\frac{1}{2}ϵ^{\mu \nu \rho \sigma }_\nu B_{\rho \sigma }^i.$$
(2.2)
Furthermore, here and throughout the paper, summation over repeated indices of any kind (whether up or down) is understood (indices $`i`$ are never lowered).<sup>3</sup><sup>3</sup>3The remaining conventions and notation are analogous to those in , except that the Minkowski metric $`\eta _{\mu \nu }=\mathrm{𝑑𝑖𝑎𝑔}(1,1,1,1)`$ is used.
To examine whether this free action is N=2 supersymmetric and to determine the supersymmetry transformations, we make the most general ansatz for linear N=2 supersymmetry transformations compatible with Poincaré covariance and with the dimensions of the fields. We write these transformations in the form
$$\delta {}_{\xi }{}^{(0)}=\xi ^{\alpha i}D{}_{\alpha }{}^{(0)}{}_{}{}^{i}+\overline{\xi }_{\dot{\alpha }}^i\overline{D}{}_{}{}^{(0)}{}_{}{}^{\dot{\alpha }i}.$$
(2.3)
Here $`\xi _\alpha ^i`$ are two constant anticommuting Weyl spinors (they are the parameters of the supersymmetry transformations), $`D{}_{\alpha }{}^{(0)}^i`$ are the generators of the corresponding supersymmetry transformations and $`\overline{\xi }_{\dot{\alpha }}^i`$ and $`\overline{D}{}_{\dot{\alpha }}{}^{(0)}^i`$ are the complex conjugates of $`\xi _\alpha ^i`$ and $`D{}_{\alpha }{}^{(0)}^i`$ respectively. The ansatz takes then the form
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}a_{}^{j}`$ $`=`$ $`\frac{1}{2}(M^{ij}\chi _\alpha +N^{ij}\psi _\alpha )`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}B_{\mu \nu }^{j}`$ $`=`$ $`R^{ij}(\sigma _{\mu \nu }\chi )_\alpha +S^{ij}(\sigma _{\mu \nu }\psi )_\alpha `$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\chi _{\beta }^{}`$ $`=`$ $`0`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\psi _{\beta }^{}`$ $`=`$ $`0`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\chi }_{\dot{\alpha }}^{}`$ $`=`$ $`X^{ij}_{\alpha \dot{\alpha }}a^j+Y^{ij}H_{\alpha \dot{\alpha }}^j+Z^{ij}\sigma _{\alpha \dot{\alpha }}^\mu ^\nu B_{\nu \mu }^j`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\psi }_{\dot{\alpha }}^{}`$ $`=`$ $`\widehat{X}^{ij}_{\alpha \dot{\alpha }}a^j+\widehat{Y}^{ij}H_{\alpha \dot{\alpha }}^j+\widehat{Z}^{ij}\sigma _{\alpha \dot{\alpha }}^\mu ^\nu B_{\nu \mu }^j`$ (2.4)
where the coefficients $`M^{ij}`$, …, $`\widehat{Z}^{ij}`$ are complex numbers which are to be determined from the following requirements: (i) the free Lagrangian must be invariant modulo a total derivative under the above transformations, for any choice of $`\xi _\alpha ^i`$; (ii) the commutators of these transformations must fulfill the N=2 supersymmetry algebra at least on-shell, modulo gauge transformations and possibly with some central charges. Clearly, if there is a set of coefficients $`M^{ij}`$, …, $`\widehat{Z}^{ij}`$ which fulfills these requirements, it cannot be unique because the Lagrangian is also invariant under separate $`SO(2)`$ transformations of the $`a^i`$ and $`B_{\mu \nu }^i`$ and under $`SU(2)`$ and $`U(1)`$ transformations of the fermions. Hence, there is some freedom in writing the supersymmetry transformations owing to these global symmetries.
(i) is equivalent to the requirement that all $`D{}_{\alpha }{}^{(0)}^i`$-transformations of the free Lagrangian be total derivatives (the $`\overline{D}{}_{\dot{\alpha }}{}^{(0)}^i`$-transformations do not give additional conditions since the free Lagrangian is real modulo a total derivative). It is straightforward to verify that this imposes precisely the following conditions:
$$\mathrm{i}X^{ij}=M^{ij},\mathrm{i}\widehat{X}^{ij}=N^{ij},Y^{ij}=R^{ij},\widehat{Y}^{ij}=S^{ij},Z^{ij}=0=\widehat{Z}^{ij}.$$
(2.5)
Using this, the ansatz reduces to
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}a_{}^{j}`$ $`=`$ $`\frac{1}{2}(M^{ij}\chi _\alpha +N^{ij}\psi _\alpha )`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}B_{\mu \nu }^{j}`$ $`=`$ $`R^{ij}(\sigma _{\mu \nu }\chi )_\alpha +S^{ij}(\sigma _{\mu \nu }\psi )_\alpha `$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\chi _{\beta }^{}`$ $`=`$ $`0`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\psi _{\beta }^{}`$ $`=`$ $`0`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\chi }_{\dot{\alpha }}^{}`$ $`=`$ $`\mathrm{i}M^{ij}_{\alpha \dot{\alpha }}a^jR^{ij}H_{\alpha \dot{\alpha }}^j`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\psi }_{\dot{\alpha }}^{}`$ $`=`$ $`\mathrm{i}N^{ij}_{\alpha \dot{\alpha }}a^jS^{ij}H_{\alpha \dot{\alpha }}^j.`$ (2.6)
The remaining coefficients are subject to requirement (ii).
#### Absence of an off-shell representation.
There are no coefficients $`M^{ij}`$, $`N^{ij}`$, $`R^{ij}`$, $`S^{ij}`$ such that the transformations (2.6) give an off-shell representation of the standard N=2 supersymmetry algebra (modulo gauge transformations and possibly with a central charge). Such an off-shell representation would require in particular $`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},\overline{D}{}_{\dot{\alpha }}{}^{(0)}{}_{}{}^{j}\}=\mathrm{i}\delta ^{ij}_{\alpha \dot{\alpha }}`$ on the fermions. However, this leads to inconsistent equations for the coefficients $`M^{ij}`$, $`N^{ij}`$, $`R^{ij}`$, $`S^{ij}`$. Indeed, one finds
$`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},\overline{D}{}_{\dot{\alpha }}{}^{(0)}{}_{}{}^{j}\}\chi _\beta =\mathrm{i}\delta ^{ij}_{\alpha \dot{\alpha }}\chi _\beta ,\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},\overline{D}{}_{\dot{\alpha }}{}^{(0)}{}_{}{}^{j}\}\psi _\beta =\mathrm{i}\delta ^{ij}_{\alpha \dot{\alpha }}\psi _\beta `$
$`MM^{}=NN^{}=RR^{}=SS^{}=1,MN^{}=RS^{}=0`$
where $`M`$, $`N`$, $`R`$, $`S`$ are the matrices with entries $`M^{ij}`$, $`N^{ij}`$, $`R^{ij}`$, $`S^{ij}`$ respectively.
#### Determination of the supersymmetry transformations.
Coefficients $`M^{ij}`$, $`N^{ij}`$, $`R^{ij}`$, $`S^{ij}`$ which yield an on-shell representation of the N=2 supersymmetry algebra can be found by dualizing two scalar fields of an N=2 hypermultiplet. A hypermultiplet contains two complex scalar fields $`\phi ^i`$ ($`i=1,2`$) and two Weyl fermions (see section 4 for details). One may decompose the scalar fields into real and imaginary parts, $`\phi ^i=a^i+\mathrm{i}b^i`$, and then “dualize” the imaginary parts according to $`_\mu b^iϵ^{ji}H_\mu ^j`$ (the use of $`ϵ^{ji}`$ is a pure convention; other choices are available owing to the above-mentioned freedom in writing the supersymmetry transformations). The transformations of the $`B_{\mu \nu }^i`$ are chosen such that the transformations of $`_\mu b^i`$ and $`ϵ^{ji}H_\mu ^j`$ coincide on-shell, i.e., when the free field equations for the fermions are used. This dualization procedure yields automatically transformations that fulfill on-shell the N=2 supersymmetry algebra modulo gauge transformations. Furthermore the resulting transformations are indeed of the form (2.6). Hence, they are the sought N=2 supersymmetry transformations for the double tensor multiplet. One finds
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}a_{}^{j}`$ $`=`$ $`\frac{1}{2}(\delta ^{ij}\chi _\alpha +ϵ^{ij}\psi _\alpha )`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}B_{\mu \nu }^{j}`$ $`=`$ $`ϵ^{ij}(\sigma _{\mu \nu }\chi )_\alpha +\delta ^{ij}(\sigma _{\mu \nu }\psi )_\alpha `$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\chi _{\beta }^{}`$ $`=`$ $`0`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\psi _{\beta }^{}`$ $`=`$ $`0`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\chi }_{\dot{\alpha }}^{}`$ $`=`$ $`\mathrm{i}_{\alpha \dot{\alpha }}a^iϵ^{ij}H_{\alpha \dot{\alpha }}^j`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\psi }_{\dot{\alpha }}^{}`$ $`=`$ $`\mathrm{i}ϵ^{ij}_{\alpha \dot{\alpha }}a^jH_{\alpha \dot{\alpha }}^i.`$ (2.7)
From these transformations one reads off that the double tensor multiplet consists of two N=1 linear multiplets. For instance, the N=1 linear multiplets with respect to $`D{}_{\alpha }{}^{(0)}^1`$ are $`(a^1,B_{\mu \nu }^2,\chi )`$ and $`(a^2,B_{\mu \nu }^1,\psi )`$.
#### N=2 supersymmetry algebra.
As remarked above, the transformations (2.7) yield an on-shell representation of the N=2 supersymmetry algebra because they can be obtained by dualizing two scalar fields of a hypermultiplet. One finds that the algebra is realized on-shell modulo gauge transformations of the $`B_{\mu \nu }^i`$, without a central charge. More precisely, the algebra is realized even off-shell on the scalar fields,
$`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},\overline{D}{}_{\dot{\alpha }}{}^{(0)}{}_{}{}^{j}\}a^k`$ $`=`$ $`\mathrm{i}\delta ^{ij}_{\alpha \dot{\alpha }}a^k`$
$`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},D{}_{\beta }{}^{(0)}{}_{}{}^{j}\}a^k`$ $`=`$ $`0`$ (2.8)
while it holds on-shell on the fermions,
$`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},\overline{D}{}_{\dot{\alpha }}{}^{(0)}{}_{}{}^{j}\}\chi _\beta `$ $`=`$ $`\mathrm{i}\delta ^{ij}_{\alpha \dot{\alpha }}\chi _\beta \mathrm{i}ϵ^{ij}ϵ_{\alpha \beta }_{\gamma \dot{\alpha }}\psi ^\gamma \mathrm{i}\delta ^{ij}_{\alpha \dot{\alpha }}\chi _\beta `$
$`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},D{}_{\beta }{}^{(0)}{}_{}{}^{j}\}\chi _\gamma `$ $`=`$ $`0`$
$`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},D{}_{\beta }{}^{(0)}{}_{}{}^{j}\}\overline{\chi }_{\dot{\alpha }}`$ $`=`$ $`\mathrm{i}ϵ^{ij}ϵ_{\alpha \beta }_{\gamma \dot{\alpha }}\psi ^\gamma 0`$
$`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},\overline{D}{}_{\dot{\alpha }}{}^{(0)}{}_{}{}^{j}\}\psi _\beta `$ $`=`$ $`\mathrm{i}\delta ^{ij}_{\alpha \dot{\alpha }}\psi _\beta +\mathrm{i}ϵ^{ij}ϵ_{\alpha \beta }_{\gamma \dot{\alpha }}\chi ^\gamma \mathrm{i}\delta ^{ij}_{\alpha \dot{\alpha }}\psi _\beta `$
$`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},D{}_{\beta }{}^{(0)}{}_{}{}^{j}\}\psi _\gamma `$ $`=`$ $`0`$
$`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},D{}_{\beta }{}^{(0)}{}_{}{}^{j}\}\overline{\psi }_{\dot{\alpha }}`$ $`=`$ $`\mathrm{i}ϵ^{ij}ϵ_{\alpha \beta }_{\gamma \dot{\alpha }}\chi ^\gamma 0`$ (2.9)
where $``$ denotes on-shell equality. Finally, on the 2-form gauge potentials one has
$`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},\overline{D}{}_{\dot{\alpha }}{}^{(0)}{}_{}{}^{j}\}B_{\mu \nu }^k`$ $`=`$ $`\mathrm{i}\delta ^{ij}_{\alpha \dot{\alpha }}B_{\mu \nu }^k+2_{[\mu }\mathrm{\Lambda }_{\nu ]\alpha \dot{\alpha }}^{ijk}2ϵ^{ij}ϵ^{kl}x_{\alpha \dot{\alpha }}_{[\mu }H_{\nu ]}^l`$
$``$ $`\mathrm{i}\delta ^{ij}_{\alpha \dot{\alpha }}B_{\mu \nu }^k+_\mu \mathrm{\Lambda }_{\nu \alpha \dot{\alpha }}^{ijk}_\nu \mathrm{\Lambda }_{\mu \alpha \dot{\alpha }}^{ijk}`$
$`\{D{}_{\alpha }{}^{(0)}{}_{}{}^{i},D{}_{\beta }{}^{(0)}{}_{}{}^{j}\}B_{\mu \nu }^k`$ $`=`$ $`0`$ (2.10)
where
$$\mathrm{\Lambda }_{\mu \nu }^{ijk}=\frac{\mathrm{i}}{2}\eta _{\mu \nu }(\delta ^{jk}ϵ^{il}+\delta ^{il}ϵ^{jk}+\delta ^{ik}ϵ^{jl}+\delta ^{jl}ϵ^{ik})a^l+ϵ^{ij}ϵ^{kl}x_\nu H_\mu ^l\mathrm{i}\delta ^{ij}B_{\mu \nu }^k.$$
(2.11)
Note that the terms with the $`\mathrm{\Lambda }_{\mu \nu }^{ijk}`$ in (2.10) are special gauge transformations. Hence, the N=2 supersymmetry algebra is indeed represented on the 2-form gauge potentials on-shell modulo gauge transformations with “gauge parameters” involving the $`\mathrm{\Lambda }_{\mu \nu }^{ijk}`$.
To summarize, in the notation (2.3) one has on all fields
$$[\delta {}_{\xi }{}^{(0)},\delta {}_{\xi ^{}}{}^{(0)}]\xi ^\mu _\mu +\delta _\mathrm{\Lambda }^{(0)}$$
(2.12)
where $`\xi ^\mu `$ is a constant vector involving the supersymmetry parameters,
$$\xi ^\mu =\mathrm{i}(\xi ^i\sigma ^\mu \overline{\xi }^i\xi ^i\sigma ^\mu \overline{\xi }^i),$$
(2.13)
and $`\delta _\mathrm{\Lambda }^{(0)}`$ is a special gauge transformation,
$$\delta {}_{\mathrm{\Lambda }}{}^{(0)}B_{\mu \nu }^{i}=_\mu \mathrm{\Lambda }_\nu ^i_\nu \mathrm{\Lambda }_\mu ^i,\mathrm{\Lambda }_\mu ^i=\mathrm{\Lambda }_{\mu \nu }^{jki}(\xi ^j\sigma ^\nu \overline{\xi }^k\xi ^j\sigma ^\nu \overline{\xi }^k).$$
(2.14)
Note that both these gauge transformations and the terms with equations of motion which appear in the commutator on $`B_{\mu \nu }^i`$ involve explicitly the spacetime coordinates $`x^\mu `$. It is easy to check that the sum of these terms actually does not depend explicitly on the $`x^\mu `$ (it cannot because there is no explicit $`x`$-dependence in the supersymmetry transformations): the relevant terms in the first anticommutator (2.10) combine to the $`x`$-independent term $`2ϵ^{ij}ϵ^{kl}H_{[\nu }^l\sigma _{\mu ]\alpha \dot{\alpha }}`$. However, this term is not to be interpreted as some kind of “central charge” in the supersymmetry algebra because it equals a gauge transformation on-shell. The point is that this gauge transformation depends explicitly on the $`x^\mu `$, owing to
$`H_{[\nu }^l\sigma _{\mu ]\alpha \dot{\alpha }}=_{[\mu }(H_{\nu ]}^lx_{\alpha \dot{\alpha }})x_{\alpha \dot{\alpha }}_{[\mu }H_{\nu ]}^l_{[\mu }(H_{\nu ]}^lx_{\alpha \dot{\alpha }}).`$
A more complete discussion of the rôle and origin of such terms in supersymmetry algebras will be given elsewhere.
## 3 Brief description of deformation theory
To study interactions involving the double tensor multiplet, we shall start from the free action for one double tensor and an arbitrary number of vector multiplets and hypermultiplets. We shall seek deformations of the free action which are invariant under the standard Poincaré transformations and under possibly deformed gauge and supersymmetry transformations. The requirement that these deformations be continuous means that the deformed Lagrangian $`L`$, the corresponding gauge transformations $`\delta _\epsilon `$ and supersymmetry transformations $`\delta _\xi `$ can be expanded in deformation parameters (coupling constants) according to
$`L`$ $`=`$ $`L{}_{}{}^{(0)}+L{}_{}{}^{(1)}+L{}_{}{}^{(2)}+\mathrm{}`$ (3.1)
$`\delta _\epsilon `$ $`=`$ $`\delta {}_{\epsilon }{}^{(0)}+\delta {}_{\epsilon }{}^{(1)}+\delta {}_{\epsilon }{}^{(2)}+\mathrm{}`$ (3.2)
$`\delta _\xi `$ $`=`$ $`\delta {}_{\xi }{}^{(0)}+\delta {}_{\xi }{}^{(1)}+\delta {}_{\xi }{}^{(2)}+\mathrm{}`$ (3.3)
$`\delta _\xi ^{(k)}`$ $`=`$ $`\xi ^{\alpha i}D{}_{\alpha }{}^{(k)}{}_{}{}^{i}+\overline{\xi }_{\dot{\alpha }}^i\overline{D}{}_{}{}^{(k)}^{\dot{\alpha }i}`$ (3.4)
where $`L^{(k)}`$, $`\delta _\epsilon ^{(k)}`$, $`\delta _\xi ^{(k)}`$, $`D{}_{\alpha }{}^{(k)}^i`$ and $`\overline{D}{}_{}{}^{(k)}^{\dot{\alpha }i}`$ have order $`k`$ in the deformation parameters, and $`L^{(0)}`$, $`\delta _\epsilon ^{(0)}`$ and $`\delta _\xi ^{(0)}`$ denote the free Lagrangian, its gauge symmetries and supersymmetry transformations respectively. $`\epsilon `$ and $`\xi `$ denote collectively the “parameters” of gauge and N=2 supersymmetry transformations respectively, i.e., the $`\epsilon `$’s are arbitrary fields whereas the $`\xi `$’s are constant anticommuting spinors.
The invariance requirements in the deformed theory are
$$\delta _\epsilon L0,\delta _\xi L0,$$
(3.5)
where $``$ is equality modulo total derivatives. The analysis can now be performed “perturbatively” by expanding these invariance requirements in the deformation parameters. At first order, this requires
$`\delta {}_{\epsilon }{}^{(0)}L{}_{}{}^{(1)}+\delta {}_{\epsilon }{}^{(1)}L{}_{}{}^{(0)}0`$ (3.6)
$`\delta {}_{\xi }{}^{(0)}L{}_{}{}^{(1)}+\delta {}_{\xi }{}^{(1)}L{}_{}{}^{(0)}0.`$ (3.7)
These equations can also be cast in the form
$`\delta {}_{\epsilon }{}^{(0)}L{}_{}{}^{(1)}+{\displaystyle \underset{\mathrm{\Phi }}{}}(\delta {}_{\epsilon }{}^{(1)}\mathrm{\Phi }){\displaystyle \frac{\delta L^{(0)}}{\delta \mathrm{\Phi }}}0`$ (3.8)
$`\delta {}_{\xi }{}^{(0)}L{}_{}{}^{(1)}+{\displaystyle \underset{\mathrm{\Phi }}{}}(\delta {}_{\xi }{}^{(1)}\mathrm{\Phi }){\displaystyle \frac{\delta L^{(0)}}{\delta \mathrm{\Phi }}}0`$ (3.9)
where the sum $`_\mathrm{\Phi }`$ runs over all fields and $`\delta L{}_{}{}^{(0)}/\delta \mathrm{\Phi }`$ is the Euler-Lagrange derivative of the free Lagrangian with respect to $`\mathrm{\Phi }`$. This shows that $`L^{(1)}`$ has to be invariant on-shell under the zeroth order transformations $`\delta _\epsilon ^{(0)}`$ and $`\delta _\xi ^{(0)}`$, where this on-shell invariance refers to the free field equations $`\delta L{}_{}{}^{(0)}/\delta \mathrm{\Phi }=0`$. Furthermore, we are only interested in nontrivial deformations, i.e. in deformations that cannot be removed through mere field redefinitions. The free Lagrangian changes under infinitesimal field redefinitions $`\mathrm{\Delta }\mathrm{\Phi }`$ through terms $`_\mathrm{\Phi }(\mathrm{\Delta }\mathrm{\Phi })\delta L{}_{}{}^{(0)}/\delta \mathrm{\Phi }+_\mu M^\mu `$. These are terms which vanish on-shell in the free theory modulo total derivatives. Terms of this form are thus trivial and can be neglected without loss of generality.
Hence, the first step of the perturbative approach to the deformation problem is the determination of nontrivial on-shell invariants of the free theory. The first order deformation of the free Lagrangian is a linear combination of these on-shell invariants. The corresponding first order deformations of the gauge and supersymmetry transformations are the coefficient functions of the Euler-Lagrange derivatives $`\delta L{}_{}{}^{(0)}/\delta \mathrm{\Phi }`$ which appear in (3.8) and (3.9).
At second order, (3.5) gives
$`\delta {}_{\epsilon }{}^{(0)}L{}_{}{}^{(2)}+\delta {}_{\epsilon }{}^{(1)}L{}_{}{}^{(1)}+{\displaystyle \underset{\mathrm{\Phi }}{}}(\delta {}_{\epsilon }{}^{(2)}\mathrm{\Phi }){\displaystyle \frac{\delta L^{(0)}}{\delta \mathrm{\Phi }}}0`$ (3.10)
$`\delta {}_{\xi }{}^{(0)}L{}_{}{}^{(2)}+\delta {}_{\xi }{}^{(1)}L{}_{}{}^{(1)}+{\displaystyle \underset{\mathrm{\Phi }}{}}(\delta {}_{\xi }{}^{(2)}\mathrm{\Phi }){\displaystyle \frac{\delta L^{(0)}}{\delta \mathrm{\Phi }}}0.`$ (3.11)
These equations require that $`\delta {}_{\epsilon }{}^{(1)}L^{(1)}`$ and $`\delta {}_{\xi }{}^{(1)}L^{(1)}`$ be in the image of $`\delta _\epsilon ^{(0)}`$ and $`\delta _\xi ^{(0)}`$ respectively, at least on-shell (with respect to the free theory) and modulo total derivatives. This can impose relations between the coefficients of the on-shell invariants in $`L^{(1)}`$ (it can even set some of these coefficients to zero). Additional relations between these coefficients can be imposed by the equations arising from (3.5) at even higher orders of the deformation problem.
Such relations between the coefficients in $`L^{(1)}`$ have a cohomological characterization. In fact, the whole deformation theory sketched above can be usefully reformulated as a cohomological problem in the framework of an extended BRST formalism. The relevant cohomology is that of an extended BRST differential $`s^{(0)}`$ which encodes the global supersymmetry transformations, the gauge transformations and the equations of motion of the free theory. This cohomological formulation of the deformation theory is described in and extends the deformation theory developed in . In the cohomological approach, the classification of the on-shell invariants of the free theory amounts to compute the cohomology of the extended BRST differential of the free theory in the space of local functionals with ghost number 0. The deformation problem at orders $`2`$ is controlled by the same cohomology, but now at ghost number 1: the cohomology classes at ghost number 1 give the possible obstructions to the existence of a deformation at orders $`2`$. This rôle of the cohomology at ghost number 1 is similar to the characterization of candidate anomalies through the BRST cohomology in the quantum field theoretical context.
## 4 First order deformations of dimension $`5`$
### 4.1 Free action
The input for the deformation theory is the free action for one double tensor multiplet and a set of vector multiplets and hypermultiplets.<sup>4</sup><sup>4</sup>4The generalization to the case with more than one double tensor multiplet is left to the reader. The fields of a vector multiplet are a real gauge field $`A_\mu `$, a complex scalar field $`\varphi `$ and two Weyl fermions $`\lambda _\alpha ^i`$. The complex conjugates of $`\varphi `$ and $`\lambda _\alpha ^i`$ are denoted by $`\overline{\varphi }`$ and $`\overline{\lambda }_{\dot{\alpha }}^i`$ respectively (note: as before, complex conjugation does not lower the index $`i`$). The fields of a hypermultiplet are two complex scalar fields $`\phi ^i`$ and two Weyl fermions $`\rho _\alpha `$ and $`\eta _\alpha `$. Their complex conjugates are denoted by $`\overline{\phi }^i`$, $`\overline{\rho }_{\dot{\alpha }}`$ and $`\overline{\eta }_{\dot{\alpha }}`$. The free Lagrangian is
$`L^{(0)}`$ $`=`$ $`_\mu a^i^\mu a^iH_\mu ^iH^{\mu i}\mathrm{i}\chi \overline{\chi }\mathrm{i}\psi \overline{\psi }`$ (4.1)
$`\frac{1}{4}F_{\mu \nu }^AF^{\mu \nu A}+\frac{1}{2}_\mu \varphi ^A^\mu \overline{\varphi }^A2\mathrm{i}\lambda ^{iA}\overline{\lambda }^{iA}`$
$`+_\mu \phi ^{ia}^\mu \overline{\phi }^{ia}\mathrm{i}\rho ^a\overline{\rho }^a\mathrm{i}\eta ^a\overline{\eta }^a`$
where $`A`$ and $`a`$ label the vector multiplets and hypermultiplets respectively and $`F_{\mu \nu }^A`$ is the field strength of $`A_\mu ^A`$,
$$F_{\mu \nu }^A=_\mu A_\nu ^A_\nu A_\mu ^A.$$
Using again the notation (2.3), the N=2 supersymmetry transformations of the vector multiplets and hypermultiplets read
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}A_{\mu }^{A}`$ $`=`$ $`ϵ^{ij}(\sigma _\mu \overline{\lambda }^{jA})_\alpha `$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\varphi _{}^{A}`$ $`=`$ $`2\lambda _\alpha ^{iA}`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\varphi }_{}^{A}`$ $`=`$ $`0`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\lambda _{\beta }^{jA}`$ $`=`$ $`\frac{\mathrm{i}}{2}ϵ^{ij}\sigma _{\alpha \beta }^{\mu \nu }F_{\mu \nu }^A`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\lambda }_{\dot{\alpha }}^{jA}`$ $`=`$ $`\frac{\mathrm{i}}{2}\delta ^{ij}_{\alpha \dot{\alpha }}\overline{\varphi }^A`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\phi _{}^{ja}`$ $`=`$ $`ϵ^{ij}\rho _\alpha ^a`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\phi }_{}^{ja}`$ $`=`$ $`\delta ^{ij}\eta _\alpha ^a`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\rho _{\beta }^{a}`$ $`=`$ $`0`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\eta _{\beta }^{a}`$ $`=`$ $`0`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\rho }_{\dot{\alpha }}^{a}`$ $`=`$ $`\mathrm{i}ϵ^{ij}_{\alpha \dot{\alpha }}\overline{\phi }^{ja}`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\eta }_{\dot{\alpha }}^{a}`$ $`=`$ $`\mathrm{i}_{\alpha \dot{\alpha }}\phi ^{ia}`$ (4.2)
The gauge symmetries act nontrivially only on $`A_\mu ^A`$ and $`B_{\mu \nu }^i`$ according to
$`\delta _\epsilon {}_{}{}^{(0)}A_{\mu }^{A}`$ $`=`$ $`_\mu \epsilon ^A`$ (4.3)
$`\delta _\epsilon {}_{}{}^{(0)}B_{\mu \nu }^{i}`$ $`=`$ $`_\mu \epsilon _\nu ^i_\nu \epsilon _\mu ^i.`$ (4.4)
### 4.2 General remarks and strategy
As described in the previous section, the first step within the deformation approach is the determination of the nontrivial on-shell invariants of the free theory. The classification of these on-shell invariants can be carried out separately in subspaces of field polynomials with definite dimension and degree in the fields of the various multiplets.
To specify these subspaces, we assign dimension 1 to all bosons (scalar fields, vector fields, 2-form gauge potentials) and to the spacetime derivatives $`_\mu `$, dimension 3/2 to all fermions, and dimension 0 to the gauge parameters $`\epsilon ^A`$ and $`\epsilon _\mu ^i`$. Then the supersymmetry transformations $`D{}_{\alpha }{}^{(0)}^i`$ have dimension 1/2 and the gauge transformations $`\delta _\epsilon ^{(0)}`$ have dimension 0. Furthermore the $`D{}_{\alpha }{}^{(0)}^i`$ are linear in the fields and do not mix fields of different supermultiplets. One can therefore classify the on-shell invariants separately in subspaces of field polynomials characterized by $`(d,N_{TT},N_V,N_H)`$ where $`d`$ is the dimension, and $`N_{TT}`$, $`N_V`$ and $`N_H`$ are the degrees in the fields of the double tensor multiplet, vector multiplets and hyper multiplets respectively.
To specify the field polynomials in a given subspace, it is helpful to denote by $`N_b`$, $`N_f`$ and $`N_{}`$ the degree in the bosons, fermions and spacetime derivatives respectively. A field polynomial with a definite degree $`N_\mathrm{\Phi }`$ in all fields and a definite dimension $`d`$ fulfills thus
$$N_b+N_f=N_\mathrm{\Phi },N_b+N_{}+\frac{3}{2}N_f=d.$$
(4.5)
This yields in particular
$$N_{}+\frac{1}{2}N_f=dN_\mathrm{\Phi }.$$
(4.6)
A field polynomial characterized by $`(d,N_{TT},N_V,N_H)`$ has $`N_\mathrm{\Phi }=N_{TT}+N_V+N_H`$ and thus can contain only terms with $`(N_{},N_f)=(dN_\mathrm{\Phi },0)`$, $`(dN_\mathrm{\Phi }1,2)`$, …, $`(0,2d2N_\mathrm{\Phi })`$.
Note that $`N_\mathrm{\Phi }`$ ranges from 1 to $`d`$ (for given value of $`d`$). It is easy to verify that the values $`N_\mathrm{\Phi }=1`$ and $`N_\mathrm{\Phi }=d`$ do not give nontrivial on-shell invariants because we impose also Poincaré invariance. Indeed, in both cases the only Poincaré invariant field polynomials which are nontrivial and on-shell gauge invariant modulo a total derivative are polynomials in the undifferentiated scalar fields (there are gauge invariants with $`N_\mathrm{\Phi }=1`$, such as each $`A_\mu ^A`$, but they are not Poincaré invariant); no such polynomial is on-shell supersymmetric modulo a total derivative. In particular, there are no on-shell invariants with $`d=2`$. In the following we shall discuss the remaining cases with $`d=3,4,5`$ and $`N_{TT}1`$. The discussion covers both true interaction terms and terms quadratic in the fields.
The computations have been done in two steps: 1. Determination of the most general nontrivial real Poincaré invariant field polynomial $`P_{(d,N_{TT},N_V,N_H)}`$ which satisfies $`\delta {}_{\epsilon }{}^{(0)}P_{(d,N_{TT},N_V,N_H)}^{}0`$ where $``$ denotes on-shell equality in the free theory modulo a total derivative; 2. Imposing $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}P_{(d,N_{TT},N_V,N_H)}^{}0`$ which yields then the first order deformations $`L_{(d,N_{TT},N_V,N_H)}^{(1)}`$ for each case $`(d,N_{TT},N_V,N_H)`$ separately. Of course, $`P_{(d,N_{TT},N_V,N_H)}`$ and $`L_{(d,N_{TT},N_V,N_H)}^{(1)}`$ are determined only modulo trivial terms which do not matter. We will give “minimal expressions” for these polynomials containing as many strictly gauge invariant terms as possible, and as few trivial terms as possible.
### 4.3 d=3
The polynomials to be discussed are those with $`(d,N_{TT},N_V,N_H)=(3,2,0,0)`$, $`(3,1,0,1)`$ and $`(3,1,1,0)`$ respectively. The only Poincaré invariant terms $`(d,N_{TT},N_V,N_H)=(3,2,0,0)`$ are bilinears in the fermions of the double tensor multiplet, $`P_{(3,2,0,0)}=k_1\overline{\chi }\overline{\chi }+k_2\overline{\psi }\overline{\psi }+k_3\overline{\chi }\overline{\psi }+\mathrm{c}.\mathrm{c}.`$ where $`k_1,k_2,k_3`$. It is straightforward to verify that no such term is on-shell supersymmetric modulo a total derivative (unless $`k_1=k_2=k_3=0`$) because the $`H_\mu ^i`$ occur in the transformation of $`\overline{\psi }`$ and $`\overline{\chi }`$.
Similarly, the only Poincaré invariant terms with $`(d,N_{TT},N_V,N_H)=(3,1,0,1)`$ are linear combinations of bilinears in fermions, with one fermion of the double tensor multiplet and one fermion of a hypermultiplet respectively. Again, no nonvanishing linear combination of this type is on-shell supersymmetric modulo a total derivative owing to the presence of the $`H_\mu ^i`$ in the transformations of $`\overline{\psi }`$ and $`\overline{\chi }`$.
The case $`(d,N_{TT},N_V,N_H)=(3,1,1,0)`$ is more involved because now there are both bilinears in the fermions and terms with bosons which are Poincaré invariant and on-shell gauge invariant modulo a total derivative,
$$P_{(3,1,1,0)}=k_1^iA_\mu ^\mu a^i+k_2^iA_\mu H^{\mu i}+k_3^i\lambda ^i\psi +k_4^i\lambda ^i\chi +\overline{k}_3^i\overline{\lambda }^i\overline{\psi }+\overline{k}_4^i\overline{\lambda }^i\overline{\chi }$$
where $`k_1^i,k_2^i`$, $`k_3^i,k_4^i`$ (since the supersymmetry transformations do not mix the fields of different vector multiplets, the discussion can be made for each vector multiplet separately and we have dropped the index $`A`$). Modulo trivial terms, $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}P_{(3,1,1,0)}^{}`$ contains terms proportional to $`F_{\mu \nu }\sigma ^{\mu \nu }\psi `$, $`F_{\mu \nu }\sigma ^{\mu \nu }\chi `$ and $`H_{\alpha \dot{\alpha }}^j\overline{\lambda }^{\dot{\alpha }k}`$. $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}P_{(3,1,1,0)}^{}0`$ requires that the coefficients of these terms vanish. This imposes
$`ϵ^{ij}k_1^j\mathrm{i}k_2^i+\mathrm{i}ϵ^{ij}k_3^j=0`$
$`k_1^i\mathrm{i}ϵ^{ij}k_2^j+\mathrm{i}ϵ^{ij}k_4^j=0`$
$`ϵ^{ik}k_2^j\delta ^{ij}\overline{k}_3^kϵ^{ij}\overline{k}_4^k=0`$
which yield
$$k_1^i=k_2^i=k_3^i=k_4^i=0.$$
Remark. Recall that the cases $`N_{TT}=0`$ are not discussed here. In fact, there are nontrivial on-shell invariants with $`d=3`$ and $`N_{TT}=0`$. These are the mass terms for the fermions of the hypermultiplets which have $`(d,N_{TT},N_V,N_H)=(3,0,0,2)`$. As the above discussion shows, these mass terms have no counterparts with $`N_{TT}>0`$ owing to the presence of the 2-form gauge potentials in the double tensor multiplet (more precisely: the presence of $`H_\mu ^i`$ in the supersymmetry transformations of $`\overline{\psi }`$ and $`\overline{\chi }`$).
### 4.4 d=4
We need to discuss the cases $`N_\mathrm{\Phi }=2`$ and $`N_\mathrm{\Phi }=3`$. The cases $`N_\mathrm{\Phi }=2`$ are easy: there are simply no nontrivial terms which are quadratic in the fields, Poincaré invariant and on-shell gauge invariant modulo a total derivative. Indeed, the candidate terms $`(d,N_{TT},N_V,N_H)=(4,2,0,0)`$ are linear combinations of the terms $`H_\mu ^iH^{\mu j}`$, $`H_\mu ^i^\mu a^j`$, $`_\mu a^i^\mu a^j`$, $`\psi \overline{\psi }`$, $`\chi \overline{\chi }`$, $`\chi \overline{\psi }`$ (modulo trivial ones); candidate terms $`(d,N_{TT},N_V,N_H)=(4,1,1,0)`$ or $`(4,1,0,1)`$ are linear combinations of terms such as $`_\mu a^i^\mu \varphi ^A`$, $`H_\mu ^i^\mu \phi ^a`$, $`\lambda ^{iA}\overline{\psi }`$ etc.; all these terms vanish on-shell (in the free theory) modulo a total derivative. Note that the free Lagrangian itself is of this type and clearly vanishes on-shell modulo a total derivative.
The various terms with $`N_\mathrm{\Phi }=3`$ and $`N_{TT}1`$ are $`(d,N_{TT},N_V,N_H)=(4,3,0,0)`$, $`(4,2,1,0)`$, $`(4,2,0,1)`$, $`(4,1,2,0)`$, $`(4,1,1,1)`$, $`(4,1,0,2)`$. Owing to (4.6), the field polynomials in these subsectors contain only terms with $`(N_{},N_f)=(1,0)`$ or $`(0,2)`$.
Consider first the case $`(d,N_{TT},N_V,N_H)=(4,3,0,0)`$. Poincaré invariance excludes all terms with $`(N_{},N_f)=(1,0)`$. The most general Poincaré invariant field polynomial with $`(N_{},N_f)=(0,2)`$ is $`k_1^ia^i\overline{\psi }\overline{\psi }+k_2^ia^i\overline{\psi }\overline{\chi }+k_3^ia^i\overline{\chi }\overline{\chi }+k_4^iB_{\mu \nu }^i\overline{\psi }\overline{\sigma }^{\mu \nu }\overline{\chi }+\mathrm{c}.\mathrm{c}.`$ The requirement that it be gauge invariant on-shell modulo a total derivative yields $`k_4^i=0`$, i.e., $`P_{(4,3,0,0)}=k_1^ia^i\overline{\psi }\overline{\psi }+k_2^ia^i\overline{\psi }\overline{\chi }+k_3^ia^i\overline{\chi }\overline{\chi }+\mathrm{c}.\mathrm{c}.`$. $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}P_{(4,3,0,0)}^{}0`$ imposes $`k_1^i=k_2^i=k_3^i=0`$.
An analogous discussion shows that the other cases with $`N_V=0`$ \[$`(d,N_{TT},N_V,N_H)=(4,2,0,1)`$ and $`(4,1,0,2)`$\] do not yield nontrivial on-shell invariants.
The remaining cases $`(d,N_{TT},N_V,N_H)=(4,2,1,0)`$, $`(4,1,2,0)`$ and $`(4,1,1,1)`$ are more involved because now there are terms with $`(N_{},N_f)=(1,0)`$ and a new type of terms with $`(N_{},N_f)=(0,2)`$ which are Poincaré invariant and gauge invariant on-shell modulo total derivatives. These terms are of the type $`A_\mu ^Aj_A^\mu `$ where $`j_A^\mu `$ are Noether currents of the free theory which have dimension 3 and are bilinear in the fields.
In the case $`(d,N_{TT},N_V,N_H)=(4,2,1,0)`$ we can drop the index $`A`$ labelling the vector multiplets owing to $`N_V=1`$ (each vector multiplet can be treated separately because the supersymmetry transformations do not mix the fields of different vector multiplets). The general form of $`j^\mu `$ in $`A_\mu j^\mu `$ is in this case
$$j^\mu =k_1ϵ^{ij}a^i^\mu a^j+k_2\psi \sigma ^\mu \overline{\psi }+k_3\chi \sigma ^\mu \overline{\chi }+k_4\psi \sigma ^\mu \overline{\chi }+\overline{k}_4\chi \sigma ^\mu \overline{\psi }$$
where $`k_1,k_2,k_3`$, $`k_4`$ (this $`j^\mu `$ is thus a linear combination of five different real Noether currents). In addition, there are terms of the type met already above, involving one scalar field and two fermions (and no derivative),
$$\varphi (k_5\overline{\psi }\overline{\psi }+k_6\overline{\chi }\overline{\psi }+k_7\overline{\chi }\overline{\chi })+\overline{\varphi }(k_8\overline{\psi }\overline{\psi }+k_9\overline{\chi }\overline{\psi }+k_{10}\overline{\chi }\overline{\chi })+a^i\overline{\lambda }^j(k_{11}^{ij}\overline{\psi }+k_{12}^{ij}\overline{\chi })+\mathrm{c}.\mathrm{c}.$$
with $`k_5,\mathrm{},k_{12}^{ij}`$. It is easy to verify that all coefficients $`k_5,\mathrm{},k_{12}^{ij}`$ must vanish. This comes again from the fact that the $`D{}_{\alpha }{}^{(0)}^i`$-transformations of $`\overline{\chi }`$ and $`\overline{\psi }`$ contain the $`H_\mu ^i`$; as a consequence, $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}(k_5\varphi \overline{\psi }\overline{\psi }+\mathrm{})`$ contains in particular terms with one scalar field and one $`H`$, i.e. terms $`\varphi H\overline{\psi }`$, $`\overline{\varphi }H\overline{\psi }`$, $`\varphi H\overline{\chi }`$, $`\overline{\varphi }H\overline{\chi }`$, $`aH\overline{\lambda }`$. Such terms do not occur in $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}(A_\mu j^\mu )`$. This enforces $`k_5=\mathrm{}=k_{12}^{ij}=0`$. Finally, $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}(A_\mu j^\mu )0`$ imposes $`k_1=k_2=k_3=k_4=0`$. This simply follows from the fact that $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}(A_\mu j^\mu )`$ involves terms with $`\overline{\lambda }`$’s owing to $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}A_{\mu }^{}=ϵ^{ij}(\sigma _\mu \overline{\lambda }^j)_\alpha `$; evidently $`ϵ^{ij}(\sigma _\mu \overline{\lambda }^j)_\alpha j^\mu 0`$ imposes $`k_1=k_2=k_3=k_4=0`$.
The remaining cases $`(d,N_{TT},N_V,N_H)=(4,1,2,0)`$ and $`(4,1,1,1)`$ are similar. Again there terms of the form $`A_\mu ^Aj_A^\mu `$ with $`j`$’s of the form $`ss+f\overline{f}`$ and terms $`s\overline{f}\overline{f}+\mathrm{c}.\mathrm{c}.`$ where $`s`$ and $`f`$ stand for scalar fields and fermions respectively. The coefficients of all terms $`s\overline{f}\overline{f}`$ containing $`\overline{\psi }`$ or $`\overline{\chi }`$ must vanish because of the presence of $`H`$ in the supersymmetry transformations of $`\overline{\psi }`$ or $`\overline{\chi }`$. The remaining terms $`s\overline{f}\overline{f}+\mathrm{c}.\mathrm{c}.`$ have necessarily $`s=a^1`$ or $`s=a^2`$ and do not contain fermions of the double tensor multiplet (due to $`N_{TT}=1`$); their coefficients must vanish because their supersymmetry transformations contain 3-fermion-terms $`(D{}_{\alpha }{}^{(0)}{}_{}{}^{i}a_{}^{j})ff`$ which do not occur in $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}(A_\mu j^\mu )`$. It is then easy to verify that the terms $`A_\mu ^Aj_A^\mu `$ must vanish as well.
Remark. Again, the case $`N_{TT}=0`$ yields nontrivial on-shell invariants: $`(d,N_{TT},N_V,N_H)=(4,0,3,0)`$ gives the N=2 supersymmetric extension of the cubic vertices between the vector gauge fields, $`(d,N_{TT},N_V,N_H)=(4,0,1,2)`$ gives couplings between vector and hypermultiplets of the form $`A_\mu ^Aj_A^\mu +\mathrm{}`$ . These on-shell invariants yield of course the deformations of the free Lagrangian for vector multiplets and hypermultiplets to the standard $`N=2`$-supersymmetric abelian or nonabelian gauge theories.
### 4.5 d=5
We shall first discuss the polynomials with $`N_\mathrm{\Phi }=3`$ as they yield nontrivial first order deformations. The cases to be discussed are $`(d,N_{TT},N_V,N_H)=(5,3,0,0)`$, $`(5,1,2,0)`$, $`(5,1,0,2)`$, $`(5,2,1,0)`$, $`(5,2,0,1)`$, $`(5,1,1,1)`$. The first three cases yield the interaction vertices of type A, B and C mentioned in the introduction. It is helpful to observe that, for $`d=5`$ and $`N_\mathrm{\Phi }=3`$, there are only terms with $`(N_{},N_f)=(2,0)`$ and $`(1,2)`$, owing to (4.6). Furthermore, there are no nontrivial $`(N_{},N_f)=(2,0)`$-terms involving three scalar fields and no nontrivial $`(N_{},N_f)=(1,2)`$-terms involving a scalar field. Indeed, every Poincaré invariant $`(N_{},N_f)=(2,0)`$-term with three scalar fields $`s_1,s_2,s_3`$ can be brought to the form $`s_1_\mu s_2^\mu s_3`$ by adding trivial terms. However, $`s_1_\mu s_2^\mu s_3`$ itself is trivial because in the free theory it is on-shell equal to $`\frac{1}{2}_\mu (s_1s_2^\mu s_3s_3s_2^\mu s_1+s_1s_3^\mu s_2)`$. Every Poincaré invariant $`(N_{},N_f)=(1,2)`$-term with a scalar field $`s`$ is modulo a total derivative a linear combination of terms $`sf\overline{f}`$ and $`s(f)\overline{f}`$ which vanish on-shell in the free theory. This makes it relatively easy to determine $`P_{(d,N_{TT},N_V,N_H)}`$ in the various cases. I note that in order to compute $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}P_{(d,N_{TT},N_V,N_H)}^{}`$, it is often useful to employ
$$D{}_{\alpha }{}^{(0)}{}_{}{}^{i}H_{\mu }^{j}\frac{\mathrm{i}}{2}(ϵ^{ij}_\mu \chi _\alpha +\delta ^{ij}_\mu \psi _\alpha )$$
where $``$ is again on-shell equality in the free theory.
#### First order interactions of type A.
We start with the case $`(d,N_{TT},N_V,N_H)=(5,3,0,0)`$ and treat it in some detail to illustrate the calculations. In this case one finds (modulo trivial terms)
$`P_{(5,3,0,0)}`$ $`=`$ $`\frac{1}{2}k_1^kϵ^{ij}H_\mu ^iH_\nu ^jB_{\rho \sigma }^kϵ^{\mu \nu \rho \sigma }+k_2^{ijk}H_\mu ^iH^{\mu j}a^k+k_3^kϵ^{ji}a^i^\mu a^jH_\mu ^k`$
$`+(k_4^i\chi \sigma ^\mu \overline{\chi }+k_5^i\psi \sigma ^\mu \overline{\psi }+k_6^i\psi \sigma ^\mu \overline{\chi }+\overline{k}_6^i\chi \sigma ^\mu \overline{\psi })H_\mu ^i`$
where $`k_1^i,k_2^{ijk},k_3^i,k_4^i,k_5^i`$, $`k_6^i`$ and (without loss of generality) $`k_2^{ijk}=k_2^{jik}`$. One now computes $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}P_{(5,3,0,0)}^{}`$. Up to trivial terms the result is a linear combination of terms of the form $`\psi HH`$, $`\chi HH`$, $`\psi Ha`$, $`\chi Ha`$. There are two types of $`\psi HH`$-terms: terms $`H_\mu ^jH_\nu ^k\sigma ^{\mu \nu }\psi `$ which are antisymmetric in $`jk`$ (one has $`H_\mu ^jH_\nu ^k\sigma ^{\mu \nu }\psi =\frac{1}{2}ϵ^{jk}ϵ^{lm}H_\mu ^lH_\nu ^m\sigma ^{\mu \nu }\psi `$) and terms $`H_\mu ^jH^{\mu k}\psi `$ which are symmetric in $`jk`$. Similarly there are two types of $`\chi HH`$-terms. Vanishing of the coefficients of $`H_\mu ^jH_\nu ^k\sigma ^{\mu \nu }\psi `$ and $`H_\mu ^jH_\nu ^k\sigma ^{\mu \nu }\chi `$ imposes
$$\mathrm{i}k_1^i+ϵ^{ij}k_5^jk_6^i=0,\mathrm{i}k_1^i+ϵ^{ij}k_4^j+\overline{k}_6^i=0.$$
Vanishing of the coefficients of $`H_\mu ^jH^{\mu k}\psi `$ and $`H_\mu ^jH^{\mu k}\chi `$ imposes
$`\mathrm{i}ϵ^{ij}k_1^k+\frac{1}{2}ϵ^{il}k_2^{jkl}+\delta ^{ij}k_5^k+ϵ^{ij}k_6^k+(jk)=0`$
$`\mathrm{i}\delta ^{ij}k_1^k+\frac{1}{2}k_2^{jki}+ϵ^{ij}k_4^k+\delta ^{ij}\overline{k}_6^k+(jk)=0.`$
Vanishing of the coefficients of $`\psi H_\mu ^k^\mu a^j`$ and $`\chi H_\mu ^k^\mu a^j`$ imposes:
$`\mathrm{i}k_2^{ikj}\delta ^{ij}k_3^k2\mathrm{i}ϵ^{ij}k_5^k2\mathrm{i}\delta ^{ij}k_6^k=0`$
$`\mathrm{i}ϵ^{il}k_2^{lkj}+ϵ^{ij}k_3^k2\mathrm{i}\delta ^{ij}k_4^k2\mathrm{i}ϵ^{ij}\overline{k}_6^k=0.`$
These equations give
$$k_2^{ijk}=k_4^i=k_5^i=\mathrm{Re}k_6^i=0,k_3^i=2k_1^i,\mathrm{Im}k_6^i=k_1^i.$$
Choosing $`k_1^i`$ as deformation parameters, the resulting nontrivial first order deformation is thus
$`L{}_{(5,3,0,0)}{}^{(1)}=\frac{1}{2}k_1^kϵ^{ij}H_\mu ^iH_\nu ^jB_{\rho \sigma }^kϵ^{\mu \nu \rho \sigma }2k_1^iH_\mu ^iϵ^{jk}a^j^\mu a^k+\mathrm{i}k_1^iH_\mu ^i(\psi \sigma ^\mu \overline{\chi }\chi \sigma ^\mu \overline{\psi }),`$
$`k_1^i.`$ (4.7)
#### First order interactions of type B.
The case $`(d,N_{TT},N_V,N_H)=(5,1,2,0)`$ is more complex. One has
$`P_{(5,1,2,0)}`$ $`=`$ $`H_\mu ^iA_\nu ^A(k_1^{iAB}F^{\mu \nu B}+k_2^{iAB}F_{\rho \sigma }^Bϵ^{\mu \nu \rho \sigma })+a^iF_{\mu \nu }^A(k_3^{iAB}F^{\mu \nu B}+k_4^{iAB}F_{\rho \sigma }^Bϵ^{\mu \nu \rho \sigma })`$
$`+[H_\mu ^i\varphi ^A^\mu (k_5^{iAB}\overline{\varphi }^B+k_6^{iAB}\varphi ^B)+k_7^{ijAkB}H_\mu ^i\lambda ^{jA}\sigma ^\mu \overline{\lambda }^{kB}`$
$`+F_{\mu \nu }^A\lambda ^{iB}\sigma ^{\mu \nu }(k_8^{AiB}\chi +k_9^{AiB}\psi )+\mathrm{c}.\mathrm{c}.]`$
where $`k_1^{iAB},\mathrm{},k_4^{iAB}`$, $`k_5^{iAB},\mathrm{},k_9^{AiB}`$, and (without loss of generality) $`k_2^{iAB}=k_2^{iBA}`$, $`k_3^{iAB}=k_3^{iBA}`$, $`k_4^{iAB}=k_4^{iBA}`$, $`k_5^{iAB}=\overline{k}_5^{iBA}`$, $`k_6^{iAB}=k_6^{iBA}`$, $`k_7^{ijAkB}=\overline{k}_7^{ikBjA}`$ (e.g., $`k_6^{iAB}=k_6^{iBA}`$ can be imposed owing to $`H_\mu ^i\varphi ^{(A}^\mu \varphi ^{B)}=\frac{1}{2}H_\mu ^i^\mu (\varphi ^A\varphi ^B)0`$). $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}P_{(5,1,2,0)}^{}0`$ imposes
$`k_2^{iAB}=k_3^{iAB}=k_4^{iAB}=\mathrm{Im}k_5^{iAB}=k_6^{iAB}=k_8^{AiB}=k_9^{AiB}=\mathrm{Re}k_7^{ijAkB}=0,`$
$`k_1^{iAB}=k_1^{iBA}=2\mathrm{R}\mathrm{e}k_5^{iAB},\mathrm{Im}k_7^{ijAkB}=k_1^{iAB}\delta ^{jk}.`$
This gives
$`L{}_{(5,1,2,0)}{}^{(1)}=k_1^{iAB}H_\mu ^i(A_\nu ^AF^{\mu \nu B}\frac{1}{2}\varphi ^A^\mu \overline{\varphi }^B\frac{1}{2}\overline{\varphi }^A^\mu \varphi ^B2\mathrm{i}\lambda ^{jA}\sigma ^\mu \overline{\lambda }^{jB}),`$
$`k_1^{iAB}=k_1^{iBA}.`$ (4.8)
#### First order interactions of type C.
The case $`(d,N_{TT},N_V,N_H)=(5,1,0,2)`$ is quite simple. One has
$`P_{(5,1,0,2)}`$ $`=`$ $`H_\mu ^i(k_1^{ijakb}\phi ^{ja}^\mu \phi ^{kb}+k_2^{ijakb}\phi ^{ja}^\mu \overline{\phi }^{kb}`$
$`+k_3^{iab}\rho ^a\sigma ^\mu \overline{\rho }^b+k_4^{iab}\eta ^a\sigma ^\mu \overline{\eta }^b+k_5^{iab}\rho ^a\sigma ^\mu \overline{\eta }^b)+\mathrm{c}.\mathrm{c}.`$
where $`k_1^{ijakb},\mathrm{},k_5^{iab}`$, and, without loss of generality, $`k_1^{ijakb}=k_1^{ikbja}`$, $`k_2^{ijakb}=\overline{k}_2^{ikbja}`$, $`k_3^{iab}=\overline{k}_3^{iba}`$, $`k_4^{iab}=\overline{k}_4^{iba}`$. $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}P_{(5,1,0,2)}^{}0`$ imposes in addition $`k_3^{iab}=\overline{k}_4^{iab}`$, $`k_2^{ijakb}=2\mathrm{i}\delta ^{jk}k_3^{iab}`$ and $`k_1^{ijakb}=\mathrm{i}ϵ^{jk}k_5^{iab}`$. This yields
$`L{}_{(5,1,0,2)}{}^{(1)}=k_3^{iab}H_\mu ^i(2\mathrm{i}\phi ^{ja}^\mu \overline{\phi }^{jb}+\rho ^a\sigma ^\mu \overline{\rho }^b\eta ^b\sigma ^\mu \overline{\eta }^a)`$
$`+k_5^{iab}H_\mu ^i(\mathrm{i}ϵ^{jk}\phi ^{ja}^\mu \phi ^{kb}+\rho ^a\sigma ^\mu \overline{\eta }^b)+\mathrm{c}.\mathrm{c}.,`$
$`k_3^{iab}=\overline{k}_3^{iba},k_5^{iab}=k_5^{iba}.`$ (4.9)
#### Remaining cases.
The remaining cases do not give nontrivial first order deformations and will therefore be discussed only briefly. In the case $`(d,N_{TT},N_V,N_H)=(5,2,1,0)`$ one has
$`P_{(5,2,1,0)}=\varphi H_\mu ^i(k_1^{ij}H^{\mu j}+k_2^{ij}^\mu a^j)+k_3F_{\mu \nu }\chi \sigma ^{\mu \nu }\psi +H_\mu ^i\lambda ^j\sigma ^\mu (k_4^{ij}\overline{\psi }+k_5^{ij}\overline{\chi })+\mathrm{c}.\mathrm{c}.`$
where $`k_1^{ij},\mathrm{},k_5^{ij}`$. $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}P_{(5,2,1,0)}^{}0`$ gives
$$k_1^{ij}=\mathrm{}=k_5^{ij}=0.$$
The case $`(d,N_{TT},N_V,N_H)=(5,1,1,1)`$ is similar to the case $`(5,2,1,0)`$, owing to the duality relation between the double tensor multiplet and a hypermultiplet. As a consequence, it does not give nontrivial first order deformations.
In the case $`(d,N_{TT},N_V,N_H)=(5,2,0,1)`$ one has
$`P_{(5,2,0,1)}`$ $`=`$ $`H_\mu ^i(k_1^{ijk}H^{\mu j}\phi ^k+k_2^{ijk}\phi ^j^\mu a^k+k_3^i\chi \sigma ^\mu \overline{\rho }`$
$`+k_4^i\psi \sigma ^\mu \overline{\rho }+k_5^i\chi \sigma ^\mu \overline{\eta }+k_6^i\psi \sigma ^\mu \overline{\eta })+\mathrm{c}.\mathrm{c}.`$
where $`k_1^{ijk},\mathrm{},k_6^i`$. Again, $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}P_{(5,2,0,1)}^{}0`$ imposes
$$k_1^{ijk}=\mathrm{}=k_6^i=0.$$
The remaining cases with $`d=5`$ are those with $`N_\mathrm{\Phi }=2`$ and $`N_\mathrm{\Phi }=4`$. The field polynomials with $`N_\mathrm{\Phi }=2`$ contain terms with $`(N_{},N_f)=(3,0)`$ and $`(2,2)`$. It is easy to see that all Poincaré invariant terms of these types which are gauge invariant on-shell modulo total derivatives are trivial (an example is $`F^{\mu \nu }_\mu H_\nu ^i`$). Field polynomials with $`N_\mathrm{\Phi }=4`$ contain only terms with $`(N_{},N_f)=(1,0)`$ and $`(0,2)`$; there are no such Poincaré invariant polynomials with $`N_{TT}1`$ which are gauge invariant and N=2 supersymmetric modulo trivial terms.
## 5 New N=2 supersymmetric gauge theories
### 5.1 Reformulation of the free double tensor multiplet
The study and construction of the deformations at higher orders is considerably simplified by switching to an alternative (equivalent) formulation of the free double tensor multiplet. The free action (2.1) for the double tensor multiplet is replaced by
$$S{}_{TT}{}^{(0)}=d^4x(_\mu a^i^\mu a^i+h_\mu ^ih^{\mu i}+2h_\mu ^iH^{\mu i}\mathrm{i}\chi \overline{\chi }\mathrm{i}\psi \overline{\psi })$$
(5.1)
where the $`h_\mu ^i`$ are auxiliary vector fields. Eliminating them by their algebraic equations of motion reproduces (2.1). Since (2.1) and (5.1) agree except for the term $`(h_\mu ^i+H_\mu ^i)(h^{\mu i}+H^{\mu i})`$, (5.1) is invariant under the supersymmetry transformations (2.7) supplemented by $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}h_{\mu }^{j}=D{}_{\alpha }{}^{(0)}{}_{}{}^{i}H_{\mu }^{j}`$. Furthermore it is convenient to substitute $`h`$’s for $`H`$’s in the supersymmetry transformations of the fermions. This is achieved by adding suitable on-shell trivial symmetries to the transformations of the fermions and the $`h`$’s. The resulting new supersymmetry transformations $`\stackrel{~}{D}{}_{\alpha }{}^{(0)}^i`$ are
$`\stackrel{~}{D}{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\chi }_{\dot{\alpha }}^{}`$ $`=`$ $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\chi }_{\dot{\alpha }}^{}+{\displaystyle \frac{1}{2}}\sigma _{\alpha \dot{\alpha }}^\mu ϵ^{ij}{\displaystyle \frac{\delta L^{(0)}}{\delta h^{\mu j}}}`$
$`\stackrel{~}{D}{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\psi }_{\dot{\alpha }}^{}`$ $`=`$ $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\psi }_{\dot{\alpha }}^{}+{\displaystyle \frac{1}{2}}\sigma _{\alpha \dot{\alpha }}^\mu \delta ^{ij}{\displaystyle \frac{\delta L^{(0)}}{\delta h^{\mu j}}}`$
$`\stackrel{~}{D}{}_{\alpha }{}^{(0)}{}_{}{}^{i}h_{}^{\mu j}`$ $`=`$ $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}h_{}^{\mu j}{\displaystyle \frac{1}{2}}\sigma _{\alpha \dot{\alpha }}^\mu \left[ϵ^{ij}{\displaystyle \frac{\delta L^{(0)}}{\delta \overline{\chi }_{\dot{\alpha }}}}+\delta ^{ij}{\displaystyle \frac{\delta L^{(0)}}{\delta \overline{\psi }_{\dot{\alpha }}}}\right]`$
where $`\delta L{}_{}{}^{(0)}/\delta \mathrm{\Phi }`$ is the Euler-Lagrange derivative of the free Lagrangian in (5.1) with respect to $`\mathrm{\Phi }`$. The new transformations differ from the supersymmetry transformations (2.7) only through terms that vanish on-shell. Hence, they fulfill the N=2 supersymmetry algebra (on-shell and modulo gauge transformations). We shall from now on drop the tilde-symbol and denote the new supersymmetry transformations again by $`D{}_{\alpha }{}^{(0)}^i`$,
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}a_{}^{j}`$ $`=`$ $`\frac{1}{2}(\delta ^{ij}\chi _\alpha +ϵ^{ij}\psi _\alpha )`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}B_{\mu \nu }^{j}`$ $`=`$ $`ϵ^{ij}(\sigma _{\mu \nu }\chi )_\alpha +\delta ^{ij}(\sigma _{\mu \nu }\psi )_\alpha `$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}h_{\mu }^{j}`$ $`=`$ $`\frac{\mathrm{i}}{2}_\mu (ϵ^{ij}\chi _\alpha +\delta ^{ij}\psi _\alpha )`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\chi _{\beta }^{}`$ $`=`$ $`0`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\psi _{\beta }^{}`$ $`=`$ $`0`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\chi }_{\dot{\alpha }}^{}`$ $`=`$ $`\mathrm{i}_{\alpha \dot{\alpha }}a^i+ϵ^{ij}h_{\alpha \dot{\alpha }}^j`$
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\overline{\psi }_{\dot{\alpha }}^{}`$ $`=`$ $`\mathrm{i}ϵ^{ij}_{\alpha \dot{\alpha }}a^j+h_{\alpha \dot{\alpha }}^i.`$ (5.2)
The gauge transformations $`\delta _\epsilon ^{(0)}`$ do not change; the auxiliary fields $`h_\mu ^i`$ are invariant under these transformations. The free action and supersymmetry transformations of the vector multiplets and hypermultiplets are not changed.
The computations involve the following steps:
1. The auxiliary fields $`h_\mu ^i`$ substitute for $`(H_\mu ^i)`$ in the first order deformations (4.7), (4.8) and (4.9). This is possible because $`h_\mu ^i`$ and $`H_\mu ^i`$ coincide on-shell in the free theory.
2. The second step is the determination of the corresponding first order deformations $`\delta _\epsilon ^{(1)}`$ and $`D{}_{\alpha }{}^{(1)}^i`$ of the gauge and supersymmetry transformations. That amounts to making the coefficients of the Euler-Lagrange derivatives in (3.7) explicit for the respective first order deformations.
3. One computes $`\delta {}_{\epsilon }{}^{(1)}L^{(1)}`$ and $`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}L^{(1)}`$ and seeks an $`L^{(2)}`$ such that (3.10) and (3.11) hold. If necessary, one then proceeds analogously to higher orders.
### 5.2 Structure of the supersymmetry deformations
To carry out these computations and to understand the structure of the resulting supersymmetry deformations, the following observation is useful. The first order deformations (4.7), (4.8) and (4.9) have a remarkable property in common: in the fomulation with the auxiliary fields $`h_\mu ^i`$, all of them (and therefore all linear combinations of them too) are of the form
$$L{}_{}{}^{(1)}=h_\mu ^ij^{\mu i}$$
(5.3)
where $`j^{\mu i}`$ are Noether currents of the free theory,
$$_\mu j^{\mu i}=\underset{\mathrm{\Phi }}{}(\mathrm{\Delta }^i\mathrm{\Phi })\frac{\delta L^{(0)}}{\delta \mathrm{\Phi }}.$$
(5.4)
Here $`\mathrm{\Delta }^i`$ generates the global symmetry of the free Lagrangian corresponding by Noether’s first theorem to $`j^{\mu i}`$. Owing to the supersymmetry transformation of $`h_\mu ^i`$ in (5.2), one has
$$D{}_{\alpha }{}^{(0)}{}_{}{}^{i}L{}_{}{}^{(1)}=\frac{\mathrm{i}}{2}_\mu (ϵ^{ij}\chi _\alpha +\delta ^{ij}\psi _\alpha )j^{\mu j}+h_\mu ^jD{}_{\alpha }{}^{(0)}{}_{}{}^{i}j_{}^{\mu j}.$$
(5.5)
Owing to (5.4), the first term on the right hand side is a linear combination of the Euler-Lagrange derivatives of $`L^{(0)}`$ modulo a total derivative,
$$\frac{\mathrm{i}}{2}_\mu (ϵ^{ij}\chi _\alpha +\delta ^{ij}\psi _\alpha )j^{\mu j}\underset{\mathrm{\Phi }}{}\frac{\mathrm{i}}{2}(ϵ^{ij}\chi _\alpha +\delta ^{ij}\psi _\alpha )(\mathrm{\Delta }^j\mathrm{\Phi })\frac{\delta L^{(0)}}{\delta \mathrm{\Phi }}.$$
(5.6)
Hence, in order that a term of the form (5.3) gives a supersymmetric first order deformation, the second term on the right hand side of (5.5) must also be a linear combination of the Euler-Lagrange derivatives of $`L^{(0)}`$ (modulo a total derivative),
$$h_\mu ^jD{}_{\alpha }{}^{(0)}{}_{}{}^{i}j_{}^{\mu j}\underset{\mathrm{\Phi }}{}(\mathrm{\Delta }_\alpha ^i\mathrm{\Phi })\frac{\delta L^{(0)}}{\delta \mathrm{\Phi }},$$
(5.7)
for some transformations $`\mathrm{\Delta }_\alpha ^i`$. Equations (5.5) through (5.7) yield then first order deformations of the supersymmetry transformations of the following form:
$$D{}_{\alpha }{}^{(1)}{}_{}{}^{i}\mathrm{\Phi }=\mathrm{\Delta }_\alpha ^i\mathrm{\Phi }+\frac{\mathrm{i}}{2}(ϵ^{ij}\chi _\alpha +\delta ^{ij}\psi _\alpha )\mathrm{\Delta }^j\mathrm{\Phi }.$$
(5.8)
Note that (5.6) holds for any Noether current $`j^{\mu i}`$, in contrast to (5.7). The currents which appear in (4.7), (4.8) and (4.9) have thus the special property to fulfill (5.7).
Even though the supersymmetry transformations do get nontrivially deformed, one would not expect that the supersymmetry algebra gets deformed (otherwise the deformed supersymmetry algebra would contain global symmetries generated by transformations which are at least quadratic in the fields – a very unlikely possibility). Indeed one finds in all examples to be discussed in the following that the supersymmetry algebra in the deformed model is the standard one, i.e., it has the form
$$[\delta _\xi ,\delta _\xi ^{}]\xi ^\mu _\mu +\delta _{\mathrm{gauge}}$$
(5.9)
where $``$ is on-shell-equality in the deformed model (note: in the previous sections $``$ denoted on-shell-equality in the free model!), $`\delta _\xi `$ are the deformed supersymmetry transformations, $`\xi ^\mu _\mu `$ is an ordinary translation with parameter $`\xi ^\mu `$ as in (2.13), and $`\delta _{\mathrm{gauge}}`$ is a deformed gauge transformation.
### 5.3 First example (type C)
The simplest examples of the first order deformations (4.9) are those involving only one hypermultiplet. In these examples we can thus drop the index $`a`$ distinguishing different hypermultiplets. Furthermore we choose $`k_3^i=0`$ and $`k_5^i=\mathrm{i}g^i`$ with real $`g^i`$. In the formulation with the auxiliary fields $`h_\mu ^i`$, (4.9) then becomes
$$L{}_{}{}^{(1)}=g^ih_\mu ^i(ϵ^{jk}\phi ^j^\mu \phi ^k\mathrm{i}\rho \sigma ^\mu \overline{\eta }+\mathrm{c}.\mathrm{c}.),g^i.$$
#### Sketch of the computation.
$`L^{(1)}`$ is indeed of the form (5.3), with $`j^{\mu i}=g^ij^\mu `$ and $`j^\mu `$ the term in parenthesis. We have $`\delta {}_{\epsilon }{}^{(0)}L{}_{}{}^{(1)}=0`$, i.e., the gauge transformations do not get deformed at first order. In order to determine the first order deformations of the supersymmetry transformations one computes $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}L^{(1)}`$. The result is
$`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}L{}_{}{}^{(1)}\mathrm{\Gamma }_\alpha ^i[ϵ^{jk}\phi ^j{\displaystyle \frac{\delta L^{(0)}}{\delta \overline{\phi }^k}}+ϵ^{jk}\overline{\phi }^j{\displaystyle \frac{\delta L^{(0)}}{\delta \phi ^k}}+\eta _\beta {\displaystyle \frac{\delta L^{(0)}}{\delta \rho _\beta }}\rho _\beta {\displaystyle \frac{\delta L^{(0)}}{\delta \eta _\beta }}+\overline{\eta }_{\dot{\alpha }}{\displaystyle \frac{\delta L^{(0)}}{\delta \overline{\rho }_{\dot{\alpha }}}}\overline{\rho }_{\dot{\alpha }}{\displaystyle \frac{\delta L^{(0)}}{\delta \overline{\eta }_{\dot{\alpha }}}}]`$
$`+\mathrm{i}g^jh_{\alpha \dot{\alpha }}^j\left[\phi ^i{\displaystyle \frac{\delta L^{(0)}}{\delta \overline{\rho }_{\dot{\alpha }}}}ϵ^{ik}\overline{\phi }^k{\displaystyle \frac{\delta L^{(0)}}{\delta \overline{\eta }_{\dot{\alpha }}}}\right]+\mathrm{i}g^k(\phi ^i\sigma _{\mu \nu }\rho ϵ^{ij}\overline{\phi }^j\sigma _{\mu \nu }\eta )_\alpha {\displaystyle \frac{\delta L^{(0)}}{\delta B_{\mu \nu }^k}}`$
where
$$\mathrm{\Gamma }_\alpha ^i=\frac{\mathrm{i}}{2}g^j(ϵ^{ij}\chi _\alpha +\delta ^{ij}\psi _\alpha ).$$
(5.10)
Owing to $`j^{\mu i}=g^ij^\mu `$, we have $`\mathrm{\Delta }^i=g^i\mathrm{\Delta }`$, and (5.8) reads
$$D{}_{\alpha }{}^{(1)}{}_{}{}^{i}\mathrm{\Phi }=\mathrm{\Delta }_\alpha ^i\mathrm{\Phi }+\mathrm{\Gamma }_\alpha ^i\mathrm{\Delta }\mathrm{\Phi }$$
where $`\mathrm{\Delta }_\alpha ^i`$ acts nontrivially only on $`\overline{\rho }`$, $`\overline{\eta }`$ and the $`B`$’s,
$`\mathrm{\Delta }_\alpha ^i\overline{\rho }_{\dot{\alpha }}=\mathrm{i}g^kh_{\alpha \dot{\alpha }}^k\phi ^i`$
$`\mathrm{\Delta }_\alpha ^i\overline{\eta }_{\dot{\alpha }}=\mathrm{i}g^kh_{\alpha \dot{\alpha }}^kϵ^{ij}\overline{\phi }^j`$
$`\mathrm{\Delta }_\alpha ^iB_{\mu \nu }^j=\mathrm{i}g^j(ϵ^{ik}\overline{\phi }^k\sigma _{\mu \nu }\eta \phi ^i\sigma _{\mu \nu }\rho )_\alpha `$
$`\mathrm{\Delta }_\alpha ^i(\text{other fields})=0,`$
and $`\mathrm{\Delta }`$ rotates the fields of the hypermultiplet,
$`\mathrm{\Delta }\phi ^i=ϵ^{ij}\overline{\phi }^j,\mathrm{\Delta }\overline{\phi }^i=ϵ^{ij}\phi ^j`$
$`\mathrm{\Delta }\rho _\alpha =\eta _\alpha ,\mathrm{\Delta }\eta _\alpha =\rho _\alpha ,\mathrm{\Delta }\overline{\rho }_{\dot{\alpha }}=\overline{\eta }_{\dot{\alpha }},\mathrm{\Delta }\overline{\eta }_{\dot{\alpha }}=\overline{\rho }_{\dot{\alpha }}`$
$`\mathrm{\Delta }(\text{other fields})=0.`$ (5.11)
Next one computes $`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}L^{(1)}`$. One easily verifies $`\mathrm{\Delta }L{}_{}{}^{(1)}=0`$. Using in addition $`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}h_{\mu }^{j}=0`$ and $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}(g^jh_\mu ^j)=_\mu \mathrm{\Gamma }_\alpha ^i`$, it is straightforward to verify that one gets
$$D{}_{\alpha }{}^{(1)}{}_{}{}^{i}L{}_{}{}^{(1)}=D{}_{\alpha }{}^{(0)}{}_{}{}^{i}(g^jh^{\mu j}g^kh_\mu ^k\phi ^l\overline{\phi }^l).$$
#### Result.
This completes the construction of the deformation. Indeed, since the previous equation does not involve the free field equations, we can set $`D{}_{\alpha }{}^{(2)}{}_{}{}^{i}=0`$ and the term in parenthesis on the right hand side can be taken as $`L^{(2)}`$: we have $`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}L{}_{}{}^{(2)}=0`$ and $`\delta {}_{\epsilon }{}^{(0)}L{}_{}{}^{(2)}=0`$. Hence, one gets a deformed Lagrangian $`L=L{}_{}{}^{(0)}+L{}_{}{}^{(1)}+L^{(2)}`$ which is invariant under the gauge transformations $`\delta _\epsilon ^{(0)}`$ and the supersymmetry transformations $`D_\alpha ^i=D{}_{\alpha }{}^{(0)}{}_{}{}^{i}+D{}_{\alpha }{}^{(1)}^i`$. The result can be written more compactly in terms of an auxiliary covariant derivative
$$\widehat{D}_\mu =_\mu g^ih_\mu ^i\mathrm{\Delta }$$
(5.12)
with $`\mathrm{\Delta }`$ as in (5.11). The deformed Lagrangian for the double tensor multiplet and the hyper multiplet reads then
$$L=L{}_{TT}{}^{(0)}+\widehat{D}_\mu \phi ^i\widehat{D}^\mu \overline{\phi }^i\mathrm{i}\rho \sigma ^\mu \widehat{D}_\mu \overline{\rho }\mathrm{i}\eta \sigma ^\mu \widehat{D}_\mu \overline{\eta }$$
(5.13)
with $`L_{TT}^{(0)}`$ as in (5.1). The deformed supersymmetry transformations are
$`D_\alpha ^i\phi ^j`$ $`=`$ $`ϵ^{ij}\rho _\alpha +\mathrm{\Gamma }_\alpha ^iϵ^{jk}\overline{\phi }^k`$
$`D_\alpha ^i\overline{\phi }^j`$ $`=`$ $`\delta ^{ij}\eta _\alpha +\mathrm{\Gamma }_\alpha ^iϵ^{jk}\phi ^k`$
$`D_\alpha ^i\rho _\beta `$ $`=`$ $`\mathrm{\Gamma }_\alpha ^i\eta _\beta `$
$`D_\alpha ^i\eta _\beta `$ $`=`$ $`\mathrm{\Gamma }_\alpha ^i\rho _\beta `$
$`D_\alpha ^i\overline{\rho }_{\dot{\alpha }}`$ $`=`$ $`\mathrm{i}ϵ^{ij}\widehat{D}_{\alpha \dot{\alpha }}\overline{\phi }^j\mathrm{\Gamma }_\alpha ^i\overline{\eta }_{\dot{\alpha }}`$
$`D_\alpha ^i\overline{\eta }_{\dot{\alpha }}`$ $`=`$ $`\mathrm{i}\widehat{D}_{\alpha \dot{\alpha }}\phi ^i+\mathrm{\Gamma }_\alpha ^i\overline{\rho }_{\dot{\alpha }}`$
$`D_\alpha ^iB_{\mu \nu }^j`$ $`=`$ $`(ϵ^{ij}\sigma _{\mu \nu }\chi +\delta ^{ij}\sigma _{\mu \nu }\psi +\mathrm{i}g^jϵ^{ik}\overline{\phi }^k\sigma _{\mu \nu }\eta \mathrm{i}g^j\phi ^i\sigma _{\mu \nu }\rho )_\alpha `$
$`D_\alpha ^i`$ $`=`$ $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\text{on the other fields.}`$ (5.14)
It can now be explicitly verified that (5.9) holds in the deformed model.
One may finally eliminate the auxiliary fields $`h_\mu ^i`$ by their algebraic equations of motion. That amounts to the identification (both in the action and symmetry transformations)
$$h_\mu ^i\frac{1}{2}K^{ij}(H_\mu ^j+g^jϵ^{kl}\phi ^k_\mu \phi ^l\mathrm{i}g^j\rho \sigma _\mu \overline{\eta }+\mathrm{c}.\mathrm{c}.),K^{ij}=\delta ^{ij}\frac{g^ig^j\phi ^k\overline{\phi }^k}{1+g^mg^m\phi ^n\overline{\phi }^n}.$$
(5.15)
### 5.4 Second example (type B)
The simplest first order deformation (4.8) arises when only two vector multiplets are involved. So, in the following we take $`A=1,2`$. This gives $`k^{iAB}=g^iϵ^{AB}`$ in (4.8) and
$$L{}_{}{}^{(1)}=g^ih_\mu ^iϵ^{AB}(A_\nu ^AF^{\mu \nu B}\frac{1}{2}\varphi ^A^\mu \overline{\varphi }^B\frac{1}{2}\overline{\varphi }^A^\mu \varphi ^B2\mathrm{i}\lambda ^{jA}\sigma ^\mu \overline{\lambda }^{jB}),g^i.$$
#### Sketch of the computation.
Again $`L^{(1)}`$ has the form (5.3), with $`j^{\mu i}=g^ij^\mu `$. This time $`\delta {}_{\epsilon }{}^{(0)}L^{(1)}`$ does not vanish (even modulo a total derivative), i.e., the gauge transformations are deformed. One has
$`\delta {}_{\epsilon }{}^{(0)}L{}_{}{}^{(1)}\frac{1}{4}g^iϵ^{AB}\epsilon ^Aϵ_{\mu \nu \rho \sigma }F^{\rho \sigma B}{\displaystyle \frac{\delta L^{(0)}}{\delta B_{\mu \nu }^i}}+g^ih_\mu ^iϵ^{AB}\epsilon ^A{\displaystyle \frac{\delta L^{(0)}}{\delta A_\mu ^B}}.`$
The first order deformation of the gauge transformations is therefore
$`\delta {}_{\epsilon }{}^{(1)}A_{\mu }^{A}`$ $`=`$ $`g^ih_\mu ^iϵ^{AB}\epsilon ^B`$
$`\delta {}_{\epsilon }{}^{(1)}B_{\mu \nu }^{i}`$ $`=`$ $`\frac{1}{4}g^i\epsilon ^Aϵ^{AB}ϵ_{\mu \nu \rho \sigma }F^{\rho \sigma B},\delta {}_{\epsilon }{}^{(1)}(\text{other fields})=0`$
Next one computes $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}L^{(1)}`$. The result is again of the form
$$D{}_{\alpha }{}^{(0)}{}_{}{}^{i}L{}_{}{}^{(1)}\underset{\mathrm{\Phi }}{}(D{}_{\alpha }{}^{(1)}{}_{}{}^{i}\mathrm{\Phi })\frac{\delta L^{(0)}}{\delta \mathrm{\Phi }},D{}_{\alpha }{}^{(1)}{}_{}{}^{i}\mathrm{\Phi }=\mathrm{\Delta }_\alpha ^i\mathrm{\Phi }+\mathrm{\Gamma }_\alpha ^i\mathrm{\Delta }\mathrm{\Phi }$$
with $`\mathrm{\Gamma }_\alpha ^i`$ as in (5.10), but now $`\mathrm{\Delta }_\alpha ^i`$ is given by
$`\mathrm{\Delta }_\alpha ^i\lambda _\beta ^{jA}=\mathrm{i}ϵ^{ij}ϵ^{AB}g^kh_\mu ^kA_\nu ^B\sigma ^{\mu \nu }_{\alpha \beta }`$
$`\mathrm{\Delta }_\alpha ^i\overline{\lambda }_{\dot{\alpha }}^{jA}=\frac{\mathrm{i}}{2}\delta ^{ij}ϵ^{AB}g^kh_{\alpha \dot{\alpha }}^k\overline{\varphi }^B`$
$`\mathrm{\Delta }_\alpha ^iB_{\mu \nu }^j=\mathrm{i}g^jϵ^{AB}(\overline{\varphi }^A\sigma _{\mu \nu }\lambda ^{iB}+ϵ^{ik}A_{[\mu }^A\sigma _{\nu ]}\overline{\lambda }^{kB})_\alpha `$
$`\mathrm{\Delta }_\alpha ^i(\text{other fields})=0`$
and $`\mathrm{\Delta }`$ rotates the fields of the vector multiplets,
$$\mathrm{\Delta }X^A=ϵ^{AB}X^B\text{for}X^A\{A_\mu ^A,\varphi ^A,\overline{\varphi }^A,\lambda _\alpha ^{Ai},\overline{\lambda }_{\dot{\alpha }}^{iA}\},\mathrm{\Delta }(\text{other fields})=0.$$
(5.16)
To determine the second order deformation, one must compute both $`\delta {}_{\epsilon }{}^{(1)}L^{(1)}`$ and $`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}L^{(1)}`$. As in the first example, $`L^{(1)}`$ is $`\mathrm{\Delta }`$-invariant which makes it to easy to compute $`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}L^{(1)}`$. The result is
$`\delta {}_{\epsilon }{}^{(1)}L^{(1)}`$ $`=`$ $`\delta {}_{\epsilon }{}^{(0)}L{}_{}{}^{(2)}+\frac{1}{2}g^ig^jh^{\rho j}A^{\sigma A}\epsilon ^Aϵ_{\mu \nu \rho \sigma }{\displaystyle \frac{\delta L^{(0)}}{\delta B_{\mu \nu }^i}}`$
$`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}L^{(1)}`$ $`=`$ $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}L^{(2)}`$
where
$$L{}_{}{}^{(2)}=\frac{1}{2}(g^ih^{\mu i}g^jh^{\nu j}A_\mu ^AA_\nu ^Ag^ih^{\mu i}g^jh_\mu ^jA^{\nu A}A_\nu ^A+g^ih^{\mu i}g^jh_\mu ^j\varphi ^A\overline{\varphi }^A).$$
The second order deformations of the gauge and supersymmetry transformations are thus
$$\delta {}_{\epsilon }{}^{(2)}B_{\mu \nu }^{i}=\frac{1}{2}g^i\epsilon ^Aϵ_{\mu \nu \rho \sigma }g^jh^{\rho j}A^{\sigma A},\delta {}_{\epsilon }{}^{(2)}(\text{other fields})=0,D{}_{\alpha }{}^{(2)}{}_{}{}^{i}=0.$$
#### Result.
This completes the construction of the deformation. Indeed, one has $`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}L{}_{}{}^{(2)}=0`$ and $`\delta {}_{\epsilon }{}^{(2)}L{}_{}{}^{(2)}=0`$. Hence, $`L=L{}_{}{}^{(0)}+L{}_{}{}^{(1)}+L^{(2)}`$ is invariant (modulo total derivatives) under $`D_\alpha ^i=D{}_{\alpha }{}^{(0)}{}_{}{}^{i}+D{}_{\alpha }{}^{(1)}^i`$ and $`\delta _\epsilon =\delta {}_{\epsilon }{}^{(0)}+\delta {}_{\epsilon }{}^{(1)}+\delta _\epsilon ^{(2)}`$. Again, the deformed Lagrangian and symmetry transformations can be written more compactly in terms of auxiliary covariant derivatives (5.12) where now $`\mathrm{\Delta }`$ is given by (5.16). The deformed Lagrangian for two vector multiplets and the double tensor multiplets reads then
$$L=L{}_{TT}{}^{(0)}\frac{1}{4}\widehat{F}_{\mu \nu }^A\widehat{F}^{\mu \nu A}+\frac{1}{2}\widehat{D}_\mu \varphi ^A\widehat{D}^\mu \overline{\varphi }^A2\mathrm{i}\lambda ^{iA}\sigma ^\mu \widehat{D}_\mu \overline{\lambda }^{iA}$$
(5.17)
with $`L_{TT}^{(0)}`$ as in (5.1) and
$$\widehat{F}_{\mu \nu }^A=\widehat{D}_\mu A_\nu ^A\widehat{D}_\nu A_\mu ^A=F_{\mu \nu }^A+g^ih_\mu ^iϵ^{AB}A_\nu ^Bg^ih_\nu ^iϵ^{AB}A_\mu ^B.$$
(5.18)
The deformed gauge transformations are
$`\delta _\epsilon A_\mu ^A=_\mu \epsilon ^A+g^ih_\mu ^iϵ^{AB}\epsilon ^B=:\widehat{D}_\mu \epsilon ^A`$
$`\delta _\epsilon B_{\mu \nu }^i=\frac{1}{4}g^i\epsilon ^Aϵ^{AB}ϵ_{\mu \nu \rho \sigma }\widehat{F}^{\rho \sigma B}+_\mu \epsilon _\nu ^i_\nu \epsilon _\mu ^i`$
$`\delta _\epsilon (\text{other fields})=0`$ (5.19)
and the deformed supersymmetry transformations are
$`D_\alpha {}_{}{}^{i}A_{\mu }^{A}`$ $`=`$ $`ϵ^{ij}(\sigma _\mu \overline{\lambda }^{jA})_\alpha ϵ^{AB}\mathrm{\Gamma }_\alpha ^iA_\mu ^B`$
$`D_\alpha {}_{}{}^{i}\varphi _{}^{A}`$ $`=`$ $`2\lambda _\alpha ^{iA}ϵ^{AB}\mathrm{\Gamma }_\alpha ^i\varphi ^B`$
$`D_\alpha {}_{}{}^{i}\overline{\varphi }_{}^{A}`$ $`=`$ $`ϵ^{AB}\mathrm{\Gamma }_\alpha ^i\overline{\varphi }^B`$
$`D_\alpha {}_{}{}^{i}\lambda _{\beta }^{jA}`$ $`=`$ $`\frac{\mathrm{i}}{2}ϵ^{ij}\sigma _{\alpha \beta }^{\mu \nu }\widehat{F}_{\mu \nu }^Aϵ^{AB}\mathrm{\Gamma }_\alpha ^i\lambda _\beta ^{jB}`$
$`D_\alpha {}_{}{}^{i}\overline{\lambda }_{\dot{\alpha }}^{jA}`$ $`=`$ $`\frac{\mathrm{i}}{2}\delta ^{ij}\widehat{D}_{\alpha \dot{\alpha }}\overline{\varphi }^Aϵ^{AB}\mathrm{\Gamma }_\alpha ^i\overline{\lambda }_{\dot{\alpha }}^{jB}`$
$`D_\alpha {}_{}{}^{i}B_{\mu \nu }^{j}`$ $`=`$ $`(ϵ^{ij}\sigma _{\mu \nu }\chi +\delta ^{ij}\sigma _{\mu \nu }\psi +\mathrm{i}g^jϵ^{AB}\overline{\varphi }^A\sigma _{\mu \nu }\lambda ^{iB}+\mathrm{i}g^jϵ^{AB}ϵ^{ik}A_{[\mu }^A\sigma _{\nu ]}\overline{\lambda }^{kB})_\alpha `$
$`D_\alpha ^i`$ $`=`$ $`D{}_{\alpha }{}^{(0)}{}_{}{}^{i}\text{on the other fields.}`$ (5.20)
Again, one may check that (5.9) holds. Elimination of the auxiliary fields $`h_\mu ^i`$ is more cumbersome than in the first model but it is clearly possible.
### 5.5 Third example (type A)
In the formulation with the auxiliary fields, (4.7) becomes
$$L{}_{}{}^{(1)}=\frac{1}{2}g^kϵ^{ij}h_\mu ^ih_\nu ^jB_{\rho \sigma }^kϵ^{\mu \nu \rho \sigma }+2g^ih_\mu ^iϵ^{jk}a^j^\mu a^k+\mathrm{i}g^ih_\mu ^i(\chi \sigma ^\mu \overline{\psi }\psi \sigma ^\mu \overline{\chi }),g^i$$
(5.21)
where we have renamed $`k_1^i`$ to $`g^i`$. This is again of the form (5.3), with
$$j^{\mu i}=g^ij^\mu ,j^\mu =ϵ^{kj}h_\nu ^jB_{\rho \sigma }^kϵ^{\mu \nu \rho \sigma }+2ϵ^{jk}a^j^\mu a^k+\mathrm{i}\chi \sigma ^\mu \overline{\psi }\mathrm{i}\psi \sigma ^\mu \overline{\chi }.$$
This time one gets
$$\delta {}_{\epsilon }{}^{(0)}L{}_{}{}^{(1)}\underset{\mathrm{\Phi }}{}(\delta {}_{\epsilon }{}^{(1)}\mathrm{\Phi })\frac{\delta L^{(0)}}{\delta \mathrm{\Phi }},D{}_{\alpha }{}^{(0)}{}_{}{}^{i}L{}_{}{}^{(1)}\underset{\mathrm{\Phi }}{}(D{}_{\alpha }{}^{(1)}{}_{}{}^{i}\mathrm{\Phi })\frac{\delta L^{(0)}}{\delta \mathrm{\Phi }}$$
where
$`\delta {}_{\epsilon }{}^{(1)}B_{\mu \nu }^{i}`$ $`=`$ $`ϵ^{ij}g^k(h_\mu ^j\epsilon _\nu ^kh_\nu ^j\epsilon _\mu ^k),\delta {}_{\epsilon }{}^{(1)}(\text{other fields})=0`$ (5.22)
$`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}a_{}^{j}`$ $`=`$ $`\mathrm{\Gamma }_\alpha ^iϵ^{jk}a^k`$
$`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}B_{\mu \nu }^{j}`$ $`=`$ $`\mathrm{i}[g^j\sigma _{\mu \nu }(ϵ^{ik}\chi \delta ^{ik}\psi )a^k+\frac{1}{2}g^k(ϵ^{ij}\psi \delta ^{ij}\chi )B_{\mu \nu }^k]_\alpha `$
$`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}h_{\mu }^{j}`$ $`=`$ $`\frac{\mathrm{i}}{2}g^j(\delta ^{ik}\chi ϵ^{ik}\psi )h_\mu ^k`$
$`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}\chi _{\beta }^{}`$ $`=`$ $`\mathrm{\Gamma }_\alpha ^i\psi _\beta `$
$`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}\psi _{\beta }^{}`$ $`=`$ $`\mathrm{\Gamma }_\alpha ^i\chi _\beta `$
$`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}\overline{\chi }_{\dot{\alpha }}^{}`$ $`=`$ $`\mathrm{i}g^jh_{\alpha \dot{\alpha }}^jϵ^{ik}a^k+\mathrm{\Gamma }_\alpha ^i\overline{\psi }_{\dot{\alpha }}`$
$`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}\overline{\psi }_{\dot{\alpha }}^{}`$ $`=`$ $`\mathrm{i}g^jh_{\alpha \dot{\alpha }}^ja^i\mathrm{\Gamma }_\alpha ^i\overline{\chi }_{\dot{\alpha }}`$ (5.23)
with $`\mathrm{\Gamma }_\alpha ^i`$ as in (5.10). In contrast to the first two examples, $`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}h_{\mu }^{i}`$ does not vanish. This makes this example more complicated. To determine $`L^{(2)}`$, one must compute $`\delta {}_{\epsilon }{}^{(1)}L^{(1)}`$ and $`D{}_{\alpha }{}^{(1)}{}_{}{}^{i}L^{(1)}`$. The results are
$`\delta {}_{\epsilon }{}^{(1)}L{}_{}{}^{(1)}0,D{}_{\alpha }{}^{(1)}{}_{}{}^{i}L{}_{}{}^{(1)}D{}_{\alpha }{}^{(0)}{}_{}{}^{i}L^{(2)}`$ (5.24)
where
$`L^{(2)}`$ $`=`$ $`g^ig^jh^{\mu i}h_\mu ^ja^ka^kg^ig^ia^jh^{\mu j}a^kh_\mu ^k+\frac{1}{3}g^ig^iϵ^{jk}a^j_\mu a^kϵ^{lm}a^l^\mu a^m`$ (5.25)
$`g^ig^ia^jh_\mu ^j(\psi \sigma ^\mu \overline{\chi }+\chi \sigma ^\mu \overline{\psi })+\frac{\mathrm{i}}{2}g^ig^iϵ^{jk}a^j_\mu a^k(\chi \sigma ^\mu \overline{\psi }\psi \sigma ^\mu \overline{\chi })`$
$`\frac{1}{4}g^ig^i(\psi \psi \overline{\chi }\overline{\chi }+\chi \chi \overline{\psi }\overline{\psi }).`$
Note that $`L^{(2)}`$ is invariant under $`\delta _\epsilon ^{(0)}`$. Hence the two equations (5.24) are compatible. (5.24) shows also that there is no second order deformation of the gauge transformations. However, the second order deformation of the supersymmetry transformations does not vanish in this case because the free field equations are used in the second equation (5.24) (recall that $``$ stands for on-shell equality in the free theory modulo total derivatives). To go on along the above lines, one must determine $`D{}_{\alpha }{}^{(2)}^i`$, compute $`D{}_{\alpha }{}^{(2)}{}_{}{}^{i}L{}_{}{}^{(1)}+D{}_{\alpha }{}^{(1)}{}_{}{}^{i}L^{(2)}`$ etc.
Of course, it is at this stage not completely clear whether the deformation exists at all orders $`3`$. However, there are good reasons to assume that it does exist. A simple inductive argument shows that all $`L^{(r)}`$ which one would get by continuing the above procedure have dimension 4 when one assigns dimension $`1`$ to the coupling constants $`g^i`$. Hence, $`L^{(r)}`$ would be a linear combination of field monomials $`M^{(r)}`$ of dimension $`r+4`$, with coefficients of order $`r`$ in the $`g^i`$. Furthermore $`M^{(r)}`$ would have degree $`r+2`$ in the fields. Note that $`L^{(2)}`$ involves only the fields $`a^i,h_\mu ^i,\psi ,\chi ,\overline{\psi },\overline{\chi }`$ (but not the $`B_{\mu \nu }^i`$). Assume now that $`L^{(3)}`$, $`L^{(4)}`$, …can be chosen such that they involve also only these fields. The field monomials $`M^{(r)}`$, $`r2`$ would then fulfill
$$r2:N_h+N_a+N_f=r+2,N_{}+2N_h+N_a+\frac{3}{2}N_f=r+4$$
where $`N_h`$, $`N_a`$, $`N_f`$ are the degrees in the $`h_\mu ^i`$, $`a^i`$ and the fermions respectively, and $`N_{}`$ is the number of derivatives. This implies
$$r2:N_{}+N_h+\frac{1}{2}N_f=2.$$
Hence, each $`M^{(k)}`$ could only have $`(N_{},N_h,N_f,N_a)=(2,0,0,k+2)`$, $`(1,1,0,k+1)`$, $`(1,0,2,k)`$, $`(0,2,0,k)`$, $`(0,1,2,k1)`$ or $`(0,0,4,k2)`$. Modulo trivial terms, this would yield
$`r2:L^{(r)}`$ $`=`$ $`A{}_{}{}^{(r)}{}_{}{}^{ij}(g,a)_\mu a^i^\mu a^j+B{}_{}{}^{(r)}{}_{}{}^{ij}(g,a)h_\mu ^i^\mu a^j+C{}_{}{}^{(r)}{}_{}{}^{ij}(g,a)h_\mu ^ih^{\mu j}`$ (5.26)
$`+D{}_{}{}^{(r)}{}_{}{}^{ijk}(g,a)f^i\sigma ^\mu \overline{f}^j_\mu a^k+E{}_{}{}^{(r)}{}_{}{}^{ijk}(g,a)f^i\sigma ^\mu \overline{f}^jh_\mu ^k`$
$`+F{}_{}{}^{(r)}{}_{}{}^{ijkl}(g,a)f^if^j\overline{f}^k\overline{f}^l`$
where $`\{f^i\}=\{\psi ,\chi \}`$ and $`A{}_{}{}^{(r)}{}_{}{}^{ij}(g,a),\mathrm{},F{}_{}{}^{(r)}{}_{}{}^{ijkl}(g,a)`$ are polynomials in the $`g^i`$ and the undifferentiated $`a^i`$. In particular the $`h_\mu ^i`$ could be eliminated algebraically also in the deformed theory and the complete deformation of the gauge transformations would be given by $`\delta _\epsilon =\delta {}_{\epsilon }{}^{(0)}+\delta _\epsilon ^{(1)}`$. The supersymmetry transformations may of course receive higher order contributions $`D{}_{\alpha }{}^{(r)}^i`$. One can check that the supersymmetry algebra takes to first order again the standard form (5.9).
Remark. A sufficient condition for the existence of $`L^{(r)}`$ as in (5.26) to all orders can be formulated in cohomological terms. As remarked at the end of section 3, the deformation problem can be reformulated as a cohomological problem for an extended BRST differential $`s^{(0)}`$ which encodes the zeroth order gauge and supersymmetry transformations. It is then easy to show that the existence of $`L^{(r)}`$ as in (5.26) is controlled by the on-shell cohomology of $`s^{(0)}`$ (modulo total derivatives) in a space of Poincaré invariant local field polynomials in the $`a^i,h_\mu ^i,\psi ,\chi ,\overline{\psi },\overline{\chi }`$ and their derivatives which depend in addition linearly on the supersymmetry ghosts (as it is the cohomology at ghost number 1 which enters here; the ghosts of the gauge transformations do not come into play here because $`\delta _\epsilon ^{(0)}`$ and $`\delta _\epsilon ^{(1)}`$ act nontrivially only on the $`B_{\mu \nu }^i`$). Vanishing of that cohomology would guarantee the existence of $`L^{(r)}`$ as in (5.26) for all $`r`$. It is in fact reasonable to believe that this cohomology indeed vanishes because its counterpart for N=1 supersymmetric models with linear multiplets vanishes (this can be shown as in , owing to the fact that free linear multiplets have “QDS-structure” on-shell).
## 6 Comment on an N=1 superfield construction
It has been already mentioned that the free N=2 double tensor multiplet consists of two free N=1 linear multiplets. Similarly, a free N=2 vector multiplet consists of one free N=1 vector multiplet and one free N=1 chiral multiplet, while a free N=2 hypermultiplet consists of two N=1 chiral multiplets. It is well-known that all these N=1 multiplets have off-shell superfield descriptions involving auxiliary fields. These superfields might also be used to construct N=2 supersymmetric interactions involving the double tensor multiplet. A promising step into that direction is the construction in . It provides N=1 supersymmetric interactions of the sought type between N=1 linear multiplets and N=1 vector multiplets, and it can be extended so as to include N=1 chiral multiplets. Some of the resulting models will even be N=2 supersymmetric – an example has been recently given in . However, it is far from obvious how one can sieve out the models with a second supersymmetry systematically.
#### Acknowledgements.
The author acknowledges discussions with Sergei Kuzenko and Ulrich Theis and was supported by a DFG habilitation grant.
|
warning/0005/cond-mat0005336.html
|
ar5iv
|
text
|
# Specific heat of the ideal gas obeying the generalized exclusion statistics
## Abstract
We calculate the specific heat of the ideal gas obeying the generalized exclusion statistics (GES) in the continuum model and the tight binding model numerically. In the continuum model of 3-d space, the specific heat increases with statistical parameter at low temperature whereas it decreases with statistical parameter at high temperature. We find that the critical temperature normalized by $`\mu _f`$ (Fermi energy) is 0.290. The specific heat of 2-d space was known to be independent of $`g`$ in the continuum model, but it varies with $`g`$ drastically in the tight-binding model. From its unique behavior, identification of GES particles will be possible from the specific heat.
Keywords: Generalized exclusion statistics, Specific heat, Ideal gas, Tight-binding model.
PACS numbers: 05.30.-d, 05.70.-a, 05.90.+m
1 Introduction
The many-particle wave function is symmetric under an exchange of two identical particles for Boson and is antisymmetric for Fermion. The dimension of the Hilbert space is determined by counting the wave functions constrained by the symmetry. Intermediate statistics are under the intensive study recently. In one and two-dimensional space, arbitrary phases can appear in many-particle wave functions under an exchange of two identical particles, which are called anyon . In contrast with the anyon, Haldane proposed generalized statistics ten years ago without specific reference to spatial dimensions . It is called the generalized exclusion statistics (GES). In his proposal, the dimension of the Hilbert space $`d`$, and the particle number $`N`$, are connected by
$`\mathrm{\Delta }d=g\mathrm{\Delta }N`$ (1)
where $`\mathrm{\Delta }d`$ is the change of the available single-particle states number and $`\mathrm{\Delta }N`$ is the change of the particle number for identical particle system. $`g`$ is a statistical parameter. We call particles obeying the GES as $`g`$-on. The state-counting of many particle states obeying the GES is proposed later by Wu, which corresponds to Boson with $`g=0`$ and Fermion with $`g=1`$ . This counting rule seems to be exact in the limit of a large number of states , and the distribution function of the ideal $`g`$-on gas has been derived.
Because the statistics is the fundamental property of nature, it is important to know if there are particles obeying generalized statistics. For this purpose, physical properties of the ideal $`g`$-on gas should be fully understood. Thermodynamic properties of the ideal $`g`$-on gas are those properties and have been studied by several people. The mean occupation number was derived in the context of the one-dimensional solvable model and the lowest Landau level anyon model . Wu derived the entropy and free energy and so on , and Nayak et al. showed the duality . The Sommerfeld expansion was applied . Other thermal properties were studied in . In particular, the specific heat of the ideal $`g`$-on gas, which is the most fundamental observable quantity which may reflect the statistics, has been studied analytically in a simple model where the single particle energy is proportional to $`p^2`$. The $`g`$-ons which have more general energy dispersions have not been studied so far. In the present work, we find the specific heat of the ideal $`g`$-on gas in several spatial dimensions of wide energy band system and narrow energy band system based on numerical calculations.
First, we study the wide energy band system where the single particle energy is proportional to $`𝐩^2`$. In three-dimensional space, we find that the specific heat increases with statistical parameter at low temperature region whereas it decreases with statistical parameter at high temperature region. The critical temperature normalized by $`\mu _f`$ (Fermi energy) is found to be 0.290, at which the specific heat is independent of $`g`$. In two-dimensional space, the specific heats are independent of the statistical parameter. The analytical proof of the $`g`$-independence of the specific heat was given in . The origin of the $`g`$-independence is due to the constant density of state (DOS) for ideal gas in two-dimensional space.
Next, we study the ideal $`g`$-on gas on lattice space model, which has the narrow energy band, in two-dimensional space. The DOS is not constant in this model. This case would be realized in the fractional quantum Hall state (FQHS). The quasiparticle appearing in FQHS is known to be an anyonic soliton with a finite spatial extension. Two quasiparticles cannot overlap each other and have a minimum distance. So it would be reasonable to associate these quasiparticles with the particles on a lattice and to describe with the tight-binding model. It was shown, in fact, that anyons in the lowest Landau level behave as $`g`$-ons with a varying DOS . Thermodynamic properties of these gas have not been studied. Thus we study the specific heat of ideal $`g`$-on gas in the tight-binding model. By numerical calculations, we find that the specific heat of this model is totally different from the previous cases and varies with a statistical parameter of the GES.
The present paper is organized in the following way. In section 2, we review the GES and give the analytical expression of the distribution functions. In section 3 numerical calculations of the specific heats of the ideal $`g`$-on gas are given in three and two-dimensional space. In section 4, the specific heat of the tight-binding model of $`g`$-on gas is calculated numerically. Summary and discussion are given in section 5. In Appendix A, we give some useful relations. In Appendix B, we give the exact solution for the $`g`$-on distribution function of $`g=\frac{1}{N}`$ where $`N`$ is a natural number. Improvement of the Sommerfeld expansion and the duality of the coefficient are given in Appendix C.
2 Distribution function of $`g`$-on
For Boson and Fermion, the number of quantum states of $`N`$ identical particles occupying $`G`$ states is given by
$`W_b`$ $`=`$ $`{\displaystyle \frac{(G+N1)!}{N!(G1)!}},`$ (2)
$`W_f`$ $`=`$ $`{\displaystyle \frac{G!}{N!(GN)!}}`$ (3)
respectively. Wu proposed the interpolation formula implying the GES as
$`W_g={\displaystyle \frac{[G+(N1)(1g)]!}{N![GgN(1g)]!}}`$ (4)
with $`g=0`$ corresponding to Boson and $`g=1`$ Fermion. This is the state-counting rule for $`g`$-on.
Using Eq. (4), the distribution function reads
$`f(ϵ)`$ $`=`$ $`{\displaystyle \frac{1}{y(e^{\beta (ϵ\mu )})+g}}`$ (5)
where the function $`y`$ satisfies the functional equation
$`y(\zeta )^g(\mathrm{\hspace{0.25em}1}+y(\zeta ))^{1g}=\zeta ,\zeta =e^{\beta (ϵ\mu )}.`$ (6)
From Eq. (6) $`y(\zeta )=\zeta `$ for $`g=1`$, and $`y(\zeta )=\zeta 1`$ for $`g=0`$, respectively. We can easily obtain other exact solutions of Eq. (6) for $`g=2,3,4,\frac{1}{2},\frac{1}{3},\frac{1}{4}`$ by elementary calculation. Moreover, the other solutions for $`g=\frac{1}{N}`$ where $`N`$ is an arbitrary natural number ($`N>1`$) are given by
$`\overline{y}`$ $`=`$ $`\left(1{\displaystyle \frac{1}{N}}\right)\left[{}_{N1}{}^{}F_{N2}^{}({\scriptscriptstyle \frac{1}{N}},{\scriptscriptstyle \frac{1}{N}},{\scriptscriptstyle \frac{2}{N}},\mathrm{}{\scriptscriptstyle \frac{N2}{N}};{\scriptscriptstyle \frac{1}{N1}},{\scriptscriptstyle \frac{2}{N1}},\mathrm{}{\scriptscriptstyle \frac{N2}{N1}};{\scriptscriptstyle \frac{N^Ne^{N(xp)}}{(N1)^{N1}}})1\right]`$ (7)
where $`x\frac{ϵ}{t}`$, $`p\frac{\mu }{t}`$ and $`{}_{l}{}^{}F_{n}^{}(\alpha _1,\mathrm{},\alpha _l,\beta _1,\mathrm{},\beta _n;z)`$ is the hypergeometric function. The temperature $`t`$ is normalized by $`\mu _f`$ ($`\beta \mu _f\frac{1}{t}`$). The derivation of $`\overline{y}`$ is given in Appendix B.
The positivity of $`\zeta `$ leads to the generalized exclusion principle
$`f(ϵ){\displaystyle \frac{1}{g}}.`$ (8)
At zero temperature, we get $`\zeta =0`$ and $`y(\zeta )=0`$ for $`ϵ\mu _g0`$, and $`\zeta =\mathrm{}`$ and $`y(\zeta )=\mathrm{}`$ for $`ϵ\mu _g0`$, where $`\mu _g`$ is a pseudo-Fermi energy for arbitrary $`g`$ cases. Then the value of the average occupation number at zero temperature is
$`f(ϵ)=\{\begin{array}{cc}\frac{1}{g}\hfill & (ϵ\mu _g0),\hfill \\ 0\hfill & (ϵ\mu _g0).\hfill \end{array}`$ (11)
Eq. (11) means that one particle occupies $`\frac{1}{g}`$ states at zero temperature.
3 The specific heat of the ideal $`g`$-on gas in 3-d and 2-d space
We numerically compute the specific heat of the ideal $`g`$-on gas in 3-$`d`$ and 2-$`d`$ space in this section. We assume that the particle of any GES has the same mass and the spectrum, $`\frac{p^2}{2m}`$, and neglect the spin degree of freedom. In $`d`$-dimensional space, the DOS for the system with particle number, $`N`$, is $`D(ϵ)=q(d)ϵ^{\frac{d}{2}1}`$ where $`q(d)=\frac{Nd}{2}\mu _f^{\frac{d}{2}}`$ and $`\mu _f`$ is Fermi energy given by Eq. (21) in Appendix A. The chemical potential $`\mu (T)`$ is determined by the particle number $`N=_0^{\mathrm{}}D(ϵ)f(\beta (ϵ\mu ))𝑑ϵ`$. Using the $`\mu (T)`$, we can calculate the average energy $`E`$ and the specific heat $`C`$ from next relations $`E=_0^{\mathrm{}}D(ϵ)ϵf(\beta (ϵ\mu ))𝑑ϵ`$ and $`C=k_B\beta ^2\frac{dE}{d\beta }`$ where $`\beta =\frac{1}{k_BT}`$. Converting the integral variables $`ϵ`$ to $`y`$ for $`N`$ and $`E`$ gives the next relations
$`{\displaystyle \frac{2}{d}}=t^{\frac{d}{2}}{\displaystyle _a^{\mathrm{}}}𝑑y{\displaystyle \frac{\left\{\mathrm{log}(\frac{y}{a})^g(\frac{y+1}{a+1})^{1g}\right\}^{\frac{d}{2}1}}{y(y+1)}}`$ (12)
and
$`{\displaystyle \frac{E}{N\mu _f}}={\displaystyle \frac{d}{2}}t^{\frac{d}{2}+1}{\displaystyle _a^{\mathrm{}}}𝑑y{\displaystyle \frac{\left\{\mathrm{log}(\frac{y}{a})^g(\frac{y+1}{a+1})^{1g}\right\}^{\frac{d}{2}}}{y(y+1)}}.`$ (13)
$`a`$ corresponds to the value of $`y`$ for $`ϵ=0`$, which satisfies $`\beta \mu =g\mathrm{log}a+(1g)\mathrm{log}(a+1)`$. For arbitrary $`g`$-on except Boson, we can solve Eq. (12) with respect to $`a`$ at fixed temperature $`t`$ numerically and obtain the average energy from Eq. (13).
$`d=3`$
For three-dimensional space, the average energy $`E(t)`$ is shown in Fig. 1. At low temperature, the average energy increases with $`g`$, which reflects the generalized exclusion principle. At high temperature, all curves have the same slope and they go to the classical limit.
The specific heat versus temperature is shown in Fig. 2. The specific heat changes continuously and has one fixed point at $`t=0.290`$. The entropy increases with $`g`$ below this temperature, but decreases with $`g`$ above this temperature. This reflects the generalized exclusion principle. For $`g=0`$ case (Boson), infinite particle numbers in the ground state cause the Bose-Einstein condensation. For $`g0`$ case, no condensation occurs at low temperature as mentioned in . The behavior of the specific heat is different for different statistics, so in principle we are able to identify the GES from the specific heat.
Some arguments , in which the specific heat of the ideal $`g`$-on gas is expressed as a power series of the temperature (Sommerfeld expansion for Fermion), suggest that the coefficient of the first power of the temperature increases with $`g`$ in 3-d space (we improve the Sommerfeld expansion and show the duality of the coefficient in Appendix C). Our calculation is not based on the Sommerfeld expansion, and it is possible to check the validity of the Sommerfeld expansion. The deviation between the linear part in the temperature and numerical result is given in Fig. 3. At low temperature, the deviation is small, so the Sommerfeld expansion is good. However, the deviation decreases with $`g`$. For the Boson’s case below the Boson’s critical temperature it behaves as $`t^{3/2}`$. So it may be natural to see that the specific heat for very small $`g`$ differs from the linear temperature dependence. Hence the Sommerfeld expansion is not good for small $`g`$.
$`d=2`$
For two-dimensional space, the average energy $`E(t)`$ is plotted in Fig. 4. At zero temperature, $`E`$ reflects the generalized exclusion principle and depends on $`g`$. At high temperature it approaches to the classical limit. The specific heat is shown in Fig. 5. Figure 5 shows that the specific heat is independent of the statistical parameter $`g`$ at arbitrary temperature. The analytical proof of the $`g`$-independence of the specific heat has been given in , using the $`ϵ`$-independence of the DOS ($`D(ϵ)=\frac{N}{\mu _f}`$ for 2-$`d`$ space).
4 Specific heat of the tight-binding model of the ideal $`g`$-on gas in 2-d space
The quasiparticle of Laughlin’s theory of the FQHS has a finite energy width and is approximately described by the lattice model. Girvin et al. calculated the energy spectrum of the quasiparticle of the Laughlin’s theory numerically, based on the single mode approximation analogous to the Feynman’s theory of superfluid helium. Their result implies that the energy spectrum of the excited state in FQHS seems to be explained by the tight-binding model. Laughlin’s theory describes the FQHS ($`\nu `$ means the filling factor of the FQHS) of $`\nu =\frac{1}{3}`$ beautifully. In this case, the quasiparticle has $`\frac{e}{3}`$ charge , and satisfies fractional statistics of $`\frac{1}{3}`$. Hence there may be three quasiparticles in one available single-particle state; that is $`g=\frac{1}{3}`$ statistics. It would be reasonable to express the quasiparticle in the Laughlin’s theory by the tight-binding of $`g`$-on gas. We study the $`g`$-on in the tight-binding model, consequently.
In the tight-binding model, the energy spectrum is $`ϵ=c\left(\mathrm{cos}\left(k_xb\right)+\mathrm{cos}\left(k_yb\right)\right)`$ and the DOS of this model is given by
$`D(ϵ)={\displaystyle \frac{M}{c\pi ^2}}K\left({\displaystyle \frac{1}{2}}\sqrt{4\left({\displaystyle \frac{ϵ}{c}}\right)^2}\right)`$ (14)
where $`c`$ is the hopping constant and $`b`$ is the lattice spacing, and $`M(\frac{S}{b^2})`$ is the number of the lattice. $`K(x)`$ is the complete elliptic integral of the first kind
$`K(x)`$ $`=`$ $`{\displaystyle _0^{\frac{\pi }{2}}}𝑑\varphi {\displaystyle \frac{1}{\sqrt{1x^2\mathrm{sin}^2\varphi }}}.`$ (15)
The particle number and the average energy of this model is given by
$`N`$ $`=`$ $`{\displaystyle _{2c}^{2c}}𝑑ϵD(ϵ)f(ϵ)={\displaystyle \frac{M}{c\pi ^2}}{\displaystyle _{2c}^{2c}}𝑑ϵK\left(\sqrt{1\left({\displaystyle \frac{1}{2}}{\displaystyle \frac{ϵ}{c}}\right)^2}\right){\displaystyle \frac{1}{y(\zeta )+g}},`$ (16)
$`E`$ $`=`$ $`{\displaystyle _{2c}^{2c}}𝑑ϵD(ϵ)ϵf(ϵ)={\displaystyle \frac{M}{c\pi ^2}}{\displaystyle _{2c}^{2c}}𝑑ϵK\left(\sqrt{1\left({\displaystyle \frac{1}{2}}{\displaystyle \frac{ϵ}{c}}\right)^2}\right){\displaystyle \frac{ϵ}{y(\zeta )+g}}.`$ (17)
It is very difficult to calculate the specific heat for arbitrary value of $`g`$. Hence we select special values of $`g`$, $`g=1,\frac{1}{2},\frac{1}{3},\frac{1}{4}`$. We fix $`\frac{N}{M}=\frac{1}{2}`$, which corresponds to the half-filling state for Fermion and the chemical potential of Fermion is zero. We numerically calculate the temperature dependence of the average energy by using $`\mu (t)`$ and obtain the specific heat.
The result is given in Fig. 8. We cannot obtain the specific heat of $`g=\frac{1}{4},0`$ at low temperature range now owing to technical problem of numerical calculation. In the high temperature limit, the one-particle energy goes to zero. In this limit, the distribution of the low energy particle number is very small, because many particles have much of energy. The integral decreases at high temperature, since the integral region is restricted between $`2c`$ and $`2c`$. As a result, the specific heat vanishes in the low temperature limit. On the other hand, the specific heat becomes maximum at one temperature value of order $`1`$; that is, when the temperature goes to the order of the hopping constant (the temperature is normalized by the hopping coupling $`c`$), the fluctuation of energy is very large. From the DOS in Fig. 7, we read that the most of the contribution comes from the particle having the energy of the order of hopping constant. This explains why $`C(t)`$ has a maximum qualitatively. $`C(t)`$ becomes maximum at $`t=0.45,0.56,0.61`$ for $`g=1,\frac{1}{2},\frac{1}{3}`$. The maximum values are 0.49, 0.71, 0.77 (Fig. 9). The difference of the maximum values reflect each statistics. See Fig. 6.
From our results, $`g=\frac{1}{3}`$ case, which will be realized in the $`\nu =\frac{1}{3}`$ FQHS, is different from Fermion case. Therefore, we will be able to observe the exotic statistics $`g=\frac{1}{3}`$ by measuring the specific heat of the $`\nu =\frac{1}{3}`$ FQHS at the temperature for the order of the hopping constant. However, it would be difficult to observe the specific heat of 2-d electron system, since the thermal effect of the orthogonal directions to the plane affects the plane’s thermal phenomena, in which the FQHS is realized.
5 Summary and Discussion
The continuum model in three-dimensional space shows that the specific heat depends on statistics. Therefore, we will be able to observe the signature of exotic statistics by measuring the specific heat. Moreover, we find that there is a critical temperature $`t=0.290`$ by numerical calculation. Above this temperature the specific heat increases with the statistical parameter but below this temperature the specific heat decreases.
As was shown before, the continuum model in two-dimensional space shows that the specific heat does not depend on statistics. The $`g`$-independence is caused by the constant DOS. In the tight-binding model, where the DOS is not constant, the specific heat of the ideal $`g`$-on gas depends on statistics even in two-dimensional space. Hence we are able to distinguish the GES by measuring the value of the specific heat especially from the peak of the specific heat. The temperature and the value of the specific heat in which the specific heat becomes maximum is obtained for several $`g`$ values. It would be exciting to identify the exotic statistics.
T.-H. A. is grateful to K. Ishikawa for suggesting these subjects and careful reading of this manuscript and also thank N. Maeda and J. Goryo for inspiring and critical discussions. It is a pleasure to thank H. Suzuki for very useful discussions.
Appendix A: Some useful relations
From Eq. (6) $`y`$ satisfies
$`y(\zeta )^g(\mathrm{\hspace{0.33em}1}+y(\zeta ))^{1g}=\zeta `$ (18)
where $`\zeta =e^{\beta (ϵ\mu )}`$. Taking logarithm of it gives
$`\beta (ϵ\mu )=g\mathrm{log}y+(1g)\mathrm{log}(y+1).`$ (19)
We convert the integral variable $`ϵ`$ into $`y`$. From Eq. (19) we obtain
$`dϵ={\displaystyle \frac{1}{\beta }}{\displaystyle \frac{y+g}{y(y+1)}}dy.`$ (20)
The density of state $`D(ϵ)`$ of the ideal gas in $`d`$-dimensional space is given by $`D(ϵ)=\frac{(2m\pi )^{\frac{d}{2}}}{\mathrm{\Gamma }(\frac{d}{2})}\left(\frac{L}{h}\right)^dϵ^{\frac{d}{2}1}`$ where $`h`$ is the Planck constant. The Fermi energy is determined by $`N=_0^{\mu _f}𝑑ϵD(ϵ)`$ is
$`\mu _f=\left\{\mathrm{\Gamma }\left({\displaystyle \frac{d}{2}}+1\right)\rho \right\}^{\frac{2}{d}}{\displaystyle \frac{h^2}{2m\pi }},`$ (21)
where $`\rho =\frac{N}{V}`$ ($`V=L^d`$). Especially, in three and two-dimensional space, the Fermi energy is given by
$`\mu _f`$ $`=`$ $`\left({\displaystyle \frac{3\rho }{4\pi }}\right)^{\frac{2}{3}}{\displaystyle \frac{h^2}{2m}},`$ (22)
$`\mu _f`$ $`=`$ $`{\displaystyle \frac{\rho }{\pi }}{\displaystyle \frac{h^2}{2m}}.`$ (23)
Appendix B: Exact solution of the distribution function for $`g=\frac{1}{N}`$
We obtain the analytical solution of $`y`$ in this Appendix . In particular, in the case of $`g=\frac{1}{N}`$ where $`N`$ is a natural number (except $`N=1`$), $`y`$ is represented by the hypergeometric function.
In general, the $`i`$-th solution $`\overline{z_i}`$ of $`f(z)=0`$, where $`z`$ is a complex variables, is gained by
$`\overline{z_i}={\displaystyle \frac{1}{2\pi i}}{\displaystyle 𝑑z\frac{f^{}(z)}{f(z)}z}`$
where the contour of the integral is circle around $`z_i`$. From Eq. (6) in low temperature limit and $`ϵ\mu 0`$, we see that $`e^{\frac{ϵ\mu }{t}}0`$ and then $`ye^{\frac{ϵ\mu }{t}}`$. So the solution of $`y`$ in low temperature limit will behave as $`e^{\frac{ϵ\mu }{t}}`$ and go to zero. Using this formula and analytical continuation of $`y`$, the solution is given by
$`\overline{y}={\displaystyle \frac{1}{2\pi i}}{\displaystyle 𝑑y}{\displaystyle \frac{(y+1)^{\frac{1}{g}1}+(\frac{1}{g}1)y(y+1)^{\frac{1}{g}2}}{y(y+1)^{\frac{1}{g}1}e^{\frac{xp}{g}}}}y`$ (24)
where the contour of the integral is circle around $`0`$ and $`x\frac{ϵ}{t}`$, $`p\frac{\mu }{t}`$. In low temperature limit, Eq. (24) is expanded by $`\frac{y}{e^{xp}}`$. The residue gives the integral value. After short calculation, the solution $`\overline{y}`$ is given by
$`\overline{y}`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\mathrm{\Gamma }({\scriptscriptstyle \frac{n}{g}}+n+1)}{\mathrm{\Gamma }({\scriptscriptstyle \frac{n}{g}}+2)}}{\displaystyle \frac{(e^{\frac{xp}{g}})^n}{n!}}.`$ (25)
Because the singularity of the solution depends on $`g`$, we consider the case of $`g=\frac{1}{N}`$ where $`N`$ is natural number to avoid the singular expression of Eq. (25). By using the important relation of Gamma function $`\mathrm{\Gamma }(z)\mathrm{\Gamma }(1z)=\frac{\pi }{\mathrm{sin}\pi z}`$, the solution is represented by a nonsingular form. The procedure of the calculation is as follows;
$`\overline{y}`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\mathrm{\Gamma }({\scriptscriptstyle \frac{n}{g}}+n+1)}{\mathrm{\Gamma }({\scriptscriptstyle \frac{n}{g}}+2)}}{\displaystyle \frac{(e^{\frac{(xp)}{g}})^n}{n!}}`$
$`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\mathrm{\Gamma }(nN1)}{\mathrm{\Gamma }(nNn)}}{\displaystyle \frac{(e^{N(xp)})^n}{n!}}`$
$`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(e^{N(xp)})^n}{n!}}{\displaystyle \frac{\frac{N^{N\left(n\frac{1}{N}\right)}}{(2\pi )^{\frac{N1}{2}}\sqrt{N}}\mathrm{\Gamma }(n{\scriptscriptstyle \frac{1}{N}})\mathrm{\Gamma }(n)\mathrm{\Gamma }(n{\scriptscriptstyle \frac{1}{N}}+{\scriptscriptstyle \frac{2}{N}})\mathrm{}\mathrm{\Gamma }(n{\scriptscriptstyle \frac{1}{N}}+{\scriptscriptstyle \frac{N1}{N}})}{\frac{(N1)^{(N1)n}}{(2\pi )^{\frac{N2}{2}}\sqrt{N1}}\mathrm{\Gamma }(n)\mathrm{\Gamma }(n+{\scriptscriptstyle \frac{1}{N1}})\mathrm{\Gamma }(n+{\scriptscriptstyle \frac{2}{N1}})\mathrm{}\mathrm{\Gamma }(n+{\scriptscriptstyle \frac{N2}{N1}})}}`$
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }N}}\sqrt{{\displaystyle \frac{N1}{N}}}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{n!}}\left({\displaystyle \frac{N^Ne^{N(xp)}}{(N1)^{N1}}}\right)^n\underset{f(n,N)}{\underset{}{{\displaystyle \frac{\mathrm{\Gamma }(n{\scriptscriptstyle \frac{1}{N}})\mathrm{\Gamma }(n{\scriptscriptstyle \frac{1}{N}}+{\scriptscriptstyle \frac{2}{N}})\mathrm{}\mathrm{\Gamma }(n{\scriptscriptstyle \frac{1}{N}}+{\scriptscriptstyle \frac{N1}{N}})}{\mathrm{\Gamma }(n+{\scriptscriptstyle \frac{1}{N1}})\mathrm{\Gamma }(n+{\scriptscriptstyle \frac{2}{N1}})\mathrm{}\mathrm{\Gamma }(n+{\scriptscriptstyle \frac{N2}{N1}})}}}}`$
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }N}}\sqrt{{\displaystyle \frac{N1}{N}}}f(0,N)\left[{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{n!}}\left({\displaystyle \frac{N^Ne^{N(xp)}}{(N1)^{N1}}}\right)^n{\displaystyle \frac{f(n,N)}{f(0,N)}}1\right](N>1)`$
$`=`$ $`\left(1{\displaystyle \frac{1}{N}}\right)\left[{}_{N1}{}^{}F_{N2}^{}({\scriptscriptstyle \frac{1}{N}},{\scriptscriptstyle \frac{1}{N}},{\scriptscriptstyle \frac{2}{N}},\mathrm{}{\scriptscriptstyle \frac{N2}{N}};{\scriptscriptstyle \frac{1}{N1}},{\scriptscriptstyle \frac{2}{N1}},\mathrm{}{\scriptscriptstyle \frac{N2}{N1}};{\scriptscriptstyle \frac{N^Ne^{N(xp)}}{(N1)^{N1}}})1\right].`$
The solution of $`y`$ is
$`\overline{y}`$ $`=`$ $`\left(1{\displaystyle \frac{1}{N}}\right)\left[{}_{N1}{}^{}F_{N2}^{}({\scriptscriptstyle \frac{1}{N}},{\scriptscriptstyle \frac{1}{N}},{\scriptscriptstyle \frac{2}{N}},\mathrm{}{\scriptscriptstyle \frac{N2}{N}};{\scriptscriptstyle \frac{1}{N1}},{\scriptscriptstyle \frac{2}{N1}},\mathrm{}{\scriptscriptstyle \frac{N2}{N1}};{\scriptscriptstyle \frac{N^Ne^{N(xp)}}{(N1)^{N1}}})1\right]`$ (26)
where $`N>1`$. In Fermion case ($`N=1`$), we return to the first expression (25) and can get the Fermion distribution function. The hypergeometric function is represented by multiple integral, so by analytical continuation the present solution is valid in all temperature region. This solution perfectly corresponds to $`g=\frac{1}{2}(N=2)`$ and $`g=\frac{1}{3}(N=3)`$ which is obtained exactly in Eq. (6). We numerically calculate the low temperature behavior of the 3-d specific heat by using this solution.
Appendix C: Duality for the coefficient of the Sommerfeld expansion
We improve the Sommerfeld expansion and show the duality relation of the coefficient for $`g`$-on in this Appendix . The Sommerfeld expansion is the expansion in terms of small deviations from the step function
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{x^{s1}}{y+g}}𝑑x`$ $`=`$ $`{\displaystyle _0^p}x^{s1}\left({\displaystyle \frac{1}{y+g}}{\displaystyle \frac{1}{g}}\right)𝑑x+{\displaystyle _p^{\mathrm{}}}{\displaystyle \frac{x^{s1}}{y+g}}𝑑x+{\displaystyle _0^p}{\displaystyle \frac{x^{s1}}{g}}𝑑x.`$
We set
$`p=\overline{p}(1g)\mathrm{log}2`$
to express the upper and lower value of the integral of $`y`$ explicitly. Then the expansion is as follows;
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{x^{s1}}{y+g}}𝑑x`$ $`=`$ $`{\displaystyle _0^{\overline{p}}}x^{s1}\left({\displaystyle \frac{1}{y+g}}{\displaystyle \frac{1}{g}}\right)𝑑x+{\displaystyle _{\overline{p}}^{\mathrm{}}}{\displaystyle \frac{x^{s1}}{y+g}}𝑑x+{\displaystyle _0^{\overline{p}}}{\displaystyle \frac{x^{s1}}{g}}𝑑x`$
$`=`$ $`{\displaystyle _0^{\overline{p}}}x^{s1}{\displaystyle \frac{y}{g(y+g)}}𝑑x+{\displaystyle _{\overline{p}}^{\mathrm{}}}{\displaystyle \frac{x^{s1}}{y+g}}𝑑x+{\displaystyle \frac{1}{g}}{\displaystyle \frac{\overline{p}^s}{s}}`$
$`=`$ $`{\displaystyle _{\overline{p}}^0}(\overline{p}u)^{s1}{\displaystyle \frac{y}{g(y+g)}}𝑑u+{\displaystyle _0^{\mathrm{}}}(\overline{p}+v)^{s1}{\displaystyle \frac{1}{y+g}}𝑑x+{\displaystyle \frac{1}{g}}{\displaystyle \frac{\overline{p}^s}{s}}.`$
In the limit of $`t0`$ ($`\overline{p}\mathrm{}`$), we can approximate this integral as follows
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{x^{s1}}{y+g}}𝑑x`$ $``$ $`{\displaystyle _0^{\mathrm{}}}(\overline{p}u)^{s1}{\displaystyle \frac{y}{g(y+g)}}𝑑u+{\displaystyle _0^{\mathrm{}}}(\overline{p}+v)^{s1}{\displaystyle \frac{1}{y+g}}𝑑x+{\displaystyle \frac{1}{g}}{\displaystyle \frac{\overline{p}^s}{s}}`$
$`=`$ $`{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\mathrm{\Gamma }(s)}{\mathrm{\Gamma }(sn)\mathrm{\Gamma }(n+1)}}\left(1\right)^n{\displaystyle \frac{\overline{p}^{sn1}}{g}}{\displaystyle _0^{\mathrm{}}}u^n{\displaystyle \frac{y}{y+g}}𝑑u`$
$`+{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\mathrm{\Gamma }(s)}{\mathrm{\Gamma }(sn)\mathrm{\Gamma }(n+1)}}\overline{p}^{sn1}{\displaystyle _0^{\mathrm{}}}v^n{\displaystyle \frac{1}{y+g}}𝑑v+{\displaystyle \frac{1}{g}}{\displaystyle \frac{\overline{p}^s}{s}}`$
$`=`$ $`{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\mathrm{\Gamma }(s)}{\mathrm{\Gamma }(sn)\mathrm{\Gamma }(n+1)}}\overline{p}^{sn1}\left({\displaystyle \frac{(1)^n}{g}}B_n(g)+C_n(g)\right)+{\displaystyle \frac{1}{g}}{\displaystyle \frac{\overline{p}^s}{s}}`$
where $`x+\overline{p}u`$, $`x\overline{p}v`$ in third line, and $`B_n(g)`$ and $`C_n(g)`$ are given by converting $`u`$, $`v`$ into $`y`$
$`B_n(g)`$ $``$ $`{\displaystyle _0^{\mathrm{}}}u^n{\displaystyle \frac{y}{y+g}}𝑑u={\displaystyle _1^0}{\displaystyle \frac{dy}{y+1}}\left(g\mathrm{log}y(1g)\mathrm{log}(y+1)+(1g)\mathrm{log}2\right)^n,`$
$`C_n(g)`$ $``$ $`{\displaystyle _0^{\mathrm{}}}v^n{\displaystyle \frac{1}{y+g}}𝑑v={\displaystyle _1^{\mathrm{}}}{\displaystyle \frac{dy}{y(y+1)}}\left(g\mathrm{log}y+(1g)\mathrm{log}(y+1)(1g)\mathrm{log}2\right)^n.`$
In $`B_n(g)`$ and $`C_n(g)`$, we convert $`y`$ into $`z=\frac{y}{y+1}`$ and obtain
$`B_n(g)`$ $`=`$ $`{\displaystyle _{1/2}^0}{\displaystyle \frac{dz}{1z}}\left(g\mathrm{log}z+\mathrm{log}(1z)+(1g)\mathrm{log}2\right)^n,`$
$`C_n(g)`$ $`=`$ $`{\displaystyle _{1/2}^1}{\displaystyle \frac{dz}{z}}\left(g\mathrm{log}z\mathrm{log}(1z)(1g)\mathrm{log}2\right)^n.`$
Moreover, in $`C_n(g)`$ and $`B_n(g)`$, we convert $`z`$ into $`z=1w`$ and $`g`$ into $`\frac{1}{g}`$ respectively, then
$`C_n(g)`$ $`=`$ $`{\displaystyle _{1/2}^0}{\displaystyle \frac{dw}{1w}}\left(g\mathrm{log}(1w)\mathrm{log}w(1g)\mathrm{log}2\right)^n,`$
$`B_n(g)`$ $`=`$ $`{\displaystyle \frac{1}{g^n}}{\displaystyle _{1/2}^0}{\displaystyle \frac{dz}{1z}}\left(\mathrm{log}z+g\mathrm{log}(1z)(1g)\mathrm{log}2\right)^n.`$
So we get the duality relation
$`C_n(g)`$ $`=`$ $`g^nB_n({\scriptscriptstyle \frac{1}{g}}).`$ (27)
Finally, the Sommerfeld expansion is given by
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{x^{s1}}{y+g}}𝑑x={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\mathrm{\Gamma }(s)}{\mathrm{\Gamma }(sn)\mathrm{\Gamma }(n+1)}}\overline{p}^{sn1}\left({\displaystyle \frac{(1)^n}{g}}B_n(g)g^nB_n({\scriptscriptstyle \frac{1}{g}})\right)+{\displaystyle \frac{1}{g}}{\displaystyle \frac{\overline{p}^s}{s}}.`$ (28)
The Fermion case ($`g=1`$), the odd number of $`n`$ in the sum does not cancel.
|
warning/0005/cond-mat0005528.html
|
ar5iv
|
text
|
# Exact solution of stochastic directed sandpile model
\[
## Abstract
We introduce and analytically solve a directed sandpile model with stochastic toppling rules. The model clearly belongs to a different universality class from its counterpart with deterministic toppling rules, previously solved by Dhar and Ramaswamy. The critical exponents are $`D_{||}=7/4`$, $`\tau =10/7`$ in two dimensions and $`D_{||}=3/2`$, $`\tau =4/3`$ in one dimension. The upper critical dimension of the model is three, at which the exponents apart from logarithmic corrections reach their mean-field values $`D_{||}=2`$, $`\tau =3/2`$.
\]
Numerical and analytical studies of sandpile models of Self-Organized Criticality continue to be a subject of considerable research activity. In particular, a lot of effort has been recently invested in establishing the set of universality classes in these systems . The consensus seems to be that the universality class of a d-dimensional sandpile model is determined by answers to the following list of questions:
(i) Is it a critical slope or a critical height model? In other words, does a site topple when its local slope or its height exceeds a certain threshold value. Critical height models (e.g. Bak-Tang-Wiesenfeld (BTW) model ) were studied more extensively in the past and are in general better understood.
(ii) Is sand redistributed isotropically in a toppling event? According to this property sandpiles could be classified as isotropic or directed (anisotropic). The common knowledge is that it is a relevant parameter, i.e. an arbitrary small anisotropy of toppling rules usually drives the model to the directed universality class.
(iii) Finally, it is important if the sand is distributed deterministically or randomly in each individual toppling. In a model with deterministic toppling rules the configuration of the bulk of the sandpile remains unchanged if every single site on the lattice topples exactly once. This additional symmetry is usually important for the universality class of the model: e.g. the deterministic one-dimensional isotropic critical height sandpile (1D BTW) has only trivially distributed avalanches of fractal dimension 2, while its cousins with additional randomness in toppling rules, such as the Zaitsev model , Oslo model , Linear Interface Model , etc., seem to belong to the same universality class where avalanches have a non-integer fractal dimension $`D2.23`$ and a probability distribution with a clean power law exponent $`\tau 1.27`$.
Despite much numerical and analytical effort on the original BTW sandpile model (which is a deterministic/isotropic/critical-height model in the above classification), its critical exponents in two dimensions still remain controversial . The situation is somewhat better for directed models. Soon after the original BTW sandpile model, Dhar and Ramaswamy introduced and exactly solved in all dimensions its directed cousin – the Dhar-Ramaswamy (DR) model .
Both BTW and DR models have deterministic toppling rules. As far as stochastic models are concerned there is preciously little analytical results. Apart from an exact solution of a model, equivalent to the 1D stochastic directed sandpile , stochastic sandpiles were studied only numerically. In this paper we present an analytical study of the stochastic directed sandpile model in all dimensions. Stochastic directed sandpiles were brought to the attention of the community in two recent papers , reporting numerical studies of several variants of such models in two dimensions. During the preparation of this manuscript, there appeared a closely related preprint by Paczuski and Bassler in which an analytical study of the directed stochastic sandpile model was presented and similar results were obtained. In particular, using different analytical arguments they have arrived at the same set of exponents.
The microscopic rules of the stochastic directed sandpile model that we selected to study are closely related to those of the DR model . These rules are modified in the spirit of a stochastic isotropic sandpile model known as the Manna model . It is easier to define our rules in two dimensions, while generalization to higher dimensions is straightforward. A stable configuration of our model is specified by the integer height of the sandpile $`z(x_1,x_2)1`$ at each point of a 2D square lattice. The lattice has open boundary conditions along the diagonal coordinate, $`x_{||}=x_1+x_2`$, and periodic boundary conditions in the transversal direction $`x_{}=x_1x_2`$. The sand is added randomly at the line with $`x_{||}=0`$ and falls off the edge at $`x_{||}=L_{||}`$. The difference between our model and the DR model lies in toppling rules. In both cases, once the height at any given site exceeds one, this site becomes unstable and loses two grains of sand to its nearest neighbors in the direction of increasing $`x_{||}`$. However, while in the DR model each of these two neighbors gets exactly one grain of sand, in our stochastic variant the decision where to move any particular grain is done independently for each grain. In other words, with probability $`1/4`$ both grains end up on the left neighbor, with probability $`1/4`$ they go to the right neighbor, and only with probability $`1/2`$ will each neighbor get one grain as in the DR model. Obviously, on average each neighbor gets one grain, yet the additional stochastic element in these rules drives the model away from the universality class of the DR model . It is easy to see that unlike the deterministic rules of the DR model, the new stochastic rules allow for multiple topplings of some sites within one avalanche. Indeed, let us consider an example where the first active site of the avalanche toppled one grain of sand to each of its two neighbors. Let us further assume that these neighbors both had $`z=1`$ so that they both toppled and by chance distributed all four of the resulting grains of sand to the same site in the next layer. This site has received four grains of sand and is guaranteed to topple twice. The numerical simulations confirm the existence of multiple topplings in other variants of a stochastic directed sandpile model.
In order to get an analytical handle on the properties of our model we employ the same trick that was used by one of us to solve the 1D directed stochastic sandpile model . Due to the Abelian nature of the model, we can change the order in which topplings are performed without changing the outcome. It is convenient to do topplings layer by layer. This means that we topple any given site as many times as necessary to make it stable before moving on to the next unstable site, and we topple all unstable sites in one layer (a given $`x_{||}`$) before toppling any site in the next layer. Let us concentrate on a site with coordinates $`x_{||}`$ and $`x_{}`$ immediately after we have finished with topplings in the ($`x_{||}1`$)-th layer. Assume that two of its neighbors with coordinates $`x_{||}1`$ and $`x_{}\pm 1`$ have toppled, respectively, $`n_1=n(x_{||}1,x_{}1)`$ and $`n_2=n(x_{||}1,x_{}+1)`$ times. The average number of grains of sand that our selected site would receive from the previous layer is $`(n_1+n_2)`$. In the DR model there are no fluctuations around this average. Also, due to the absence of multiple topplings in this deterministic directed model, $`n_1`$ and $`n_2`$ can be only $`0`$ or $`1`$. Therefore, in the DR model a site can receive either $`n_1+n_2=2`$ grains, in which case it is guaranteed to topple exactly once, or $`n_1+n_2=1`$, in which case it can topple with probability $`1/2`$ (i.e. it topples if it had $`z=1`$ before the transfer and remains stable if it had $`z=0`$). From this one can show that in the DR model the set of sites which topple at each layer form an interval with no holes inside. The size of this interval as a function of the layer number $`x_{||}`$ performs an ordinary random walk.
In the stochastic model the relation between the number of topplings in two subsequent layers is more complicated. Let us concentrate on the behavior of the total number of topplings $`N(x_{||})=_x_{}n(x_{||},x_{})`$ in a given layer $`x_{||}`$. This number fixes the number of grains of sand transferred from the layer $`x_{||}`$ to the next layer as $`2N(x_{||})`$. It is easy to see that a site which has received an even number $`2k`$ of grains of sand from a previous layer will always topple exactly $`k`$ times and, therefore, will transfer the same $`2k`$ grains of sand to the layer directly below it. That means that as far as $`N(x_{||})`$ is concerned, such sites behave in a completely passive manner, i.e. they do not lead to a decrease or an increase of the total number of topplings $`N(x_{||})`$ from layer to layer. On the other hand, any site which received an odd number $`2k+1`$ of grains of sand from a previous layer has equal chance to topple $`k`$ times (if it had $`z=0`$ before the transfer) or $`k+1`$ times (if it had $`z=1`$). In the former case this site would decrease the grain flow $`2N(x_{||})`$ by one, while in the latter – increase by 1. Let us call any site which has received an odd number of grains of sand from the previous layer an active site. The equation, which is the central result of this work, relating the change in the total number of topplings from layer to layer to the number of active sites $`N_a(x_{||})`$ in a given layer is $`N(x_{||})=N(x_{||}1)+\frac{1}{2}_{a=1}^{N_a(x_{||})}\xi _a`$, where all $`\xi _a`$ are $`1`$ or $`+1`$ with equal probability and independent of each other. These random numbers correspond to whether each of the $`N_a(x_{||})`$ active sites had the height $`z=0`$ or $`z=1`$ before the avalanche started. It is straightforward to demonstrate that, as in the DR model, in the steady state of the directed stochastic model all possible stable configurations of $`z`$ are equally represented, and, therefore, there are no correlations between the heights at different sites, and each height is equally likely to have $`z=0`$ or $`z=1`$. It is more convenient to rewrite the above equation in a continuous notation, which works as long as $`N_a(x_{||})>>1`$ :
$$\frac{dN(x_{||})}{dx_{||}}=\frac{1}{2}\sqrt{N_a(x_{||})}\eta (x_{||}).$$
(1)
Here $`\eta (t)`$ is a standard Gaussian variable with zero mean and the standard deviation equal to unity. This equation describes an unbiased random walk $`N(x_{||})`$ vs $`x_{||}`$ with a variable step, given by $`\frac{1}{2}\sqrt{N_a(x_{||})}`$. A random walk (an avalanche) starts with $`N(0)=1`$ and ends at $`x_{||}`$ when $`N(x_{||})0`$ for the first time. Let us assume that both $`N(x_{||})`$ and $`N_a(x_{||})`$ in a surviving avalanche scale with $`x_{||}`$ with the exponents $`\alpha `$ and $`\alpha _a`$, respectively. Alternatively one can say that they are related to each other by $`N_aN^{\alpha _a/\alpha }`$. Plugging this relation into Eq. (1), after easy algebra we get the exponent relation
$$\alpha =\frac{1+\alpha _a}{2}.$$
(2)
In addition to this exponent relation, the mapping of the avalanche process onto a random walk immediately gives us the power law $`\tau _{||}`$ in the probability distribution of avalanche sizes. Indeed, an avalanche ends when the random walk described by Eq. (1) first enters the $`N(x_{||})0`$ semi-axis, and it is a standard result of the theory of stochastic processes that the distribution of first returns of a generalized random walk has an exponent $`\tau _{||}=1+\alpha `$. This agrees with the well-known exponent relation $`\tau _{||}=D_{||}`$ valid for a general directed sandpile model. Indeed, the fractal dimension $`D_{||}`$, defined by $`_{i=1}^{x_{||}}N(i)=Sx_{||}^{D_{||}}`$, is obviously related to $`\alpha `$ through $`D_{||}=1+\alpha `$.
The Eq. (1) applies equally well to the DR and the stochastic directed sandpile models. The difference between these two models lies only in the scaling of the number of active sites with $`x_{||}`$. As was explained above, in the 2D DR model the only two active sites lie at the edge of an interval of toppled sites. Indeed, only these sites get 1 grain of sand, while the rest get either 0 or 2. Therefore, in the 2D DR model $`N_a(x_{||})=2`$ is just a constant, $`\alpha _a=0`$, and the Eq. (1) describes an ordinary random walk, in which $`N(x_{||})x_{||}^\alpha =x_{||}^{1/2}`$. The introduction of a stochastic element in particle redistribution dramatically changes the number of active sites at any given layer of the avalanche. Indeed, when grains are distributed independently, any site which has at least one toppled neighbor in the previous layer is equally likely to receive an even or odd number of grains of sand, and therefore, it has a probability $`1/2`$ of becoming active. Thus, in the stochastic model the exponent $`\alpha _a`$ defines how the number of distinct sites that topple at least once, scales with the layer number $`x_{||}`$. The difference between exponents $`\alpha `$ and $`\alpha _a`$ comes solely from the existence of multiple topplings. These two exponents have to obey the inequality $`\alpha \alpha _a`$, and their difference $`\alpha \alpha _a`$ determines how the average number of topplings $`n_{\mathrm{top}}(x_{||})`$ at a given site in $`x_{||}`$-th layer scales with $`x_{||}`$: $`n_{\mathrm{top}}N/(2N_a)x_{||}^{\alpha \alpha _a}`$.
We proceed with an argument that in the 2D directed stochastic model $`\alpha _a=1/2`$, and, therefore, by the virtue of Eq. (2) $`\alpha =3/4`$. It is a straightforward task to determine the average number of topplings $`n(x_{||},x_{})`$ at a given site $`x_{||},x_{}`$, where the average is performed over the whole ensemble of avalanches, so that avalanches that die out before reaching this site, contribute 0 to the average. As was noted in , due to the conservation of sand and the stationarity of the sandpile $`n(x_{||},x_{})`$ has to satisfy the diffusion equation with a source:
$$\frac{n(x_{||},x_{})}{x_{||}}=\frac{1}{2}\frac{^2n(x_{||},x_{})}{x_{}^2}+\delta (x_{||})\delta (x_{}).$$
(3)
This average balance equation is also exact for our stochastic model, where it proves that, like in the DR model, sites that topple at least once are spread over the interval of length $`\mathrm{\Delta }x_{}x_{||}^{1/2}`$ in the $`x_{||}`$ layer. In the DR model these sites form a dense interval with no holes, and, therefore, their number is known to scale exactly as $`x_{||}^{1/2}`$. The situation is somewhat less obvious in the stochastic directed model, where the set of toppled sites can have holes. However, one can argue that these holes would mostly be concentrated near the boundaries of the avalanche in any given layer, while the majority of toppled sites in the center would form a compact interval. Indeed, as will be confirmed later, the 2D stochastic directed model is characterized by multiple topplings, where a site at a layer $`x_{||}`$ would typically topple $`n_{\mathrm{top}}(x_{||})x_{||}^{1/4}`$ times within one avalanche. Since any of the $`2n_{\mathrm{top}}`$ grains is equally likely to go to each of the two nearest neighbors in the next layer, the situation where one of these neighbors would receive less than two grains is exponentially unlikely for large $`n_{\mathrm{top}}`$. But once a site has received two or more grains of sand it is guaranteed to topple at least once. Therefore, both nearest neighbors down the slope from the site which toppled many times would most likely topple at least once. In other words, the creation of a new hole (a region free of topplings) is exponentially suppressed near sites which themselves toppled many times. This means that for a sufficiently large $`x_{||}`$ the majority of sites (especially those close to the center of the avalanche region) will belong to a hole-free region. Since the size of the interval covered by an avalanche scales as $`x_{||}^{1/2}`$, the number of active sites should also scale as $`N_ax_{||}^{1/2}`$. From $`\alpha _a=1/2`$ with the help of Eq. (1) one gets $`\alpha =3/4`$, $`\tau _{||}=D_{||}=7/4`$, and $`\tau =1+(\tau _{||}1)/D_{||}=10/7`$. These results are in a nice agreement both with previous numerical simulations of various versions of stochastic directed sandpile model in two dimensions and with our own simulations of the model. In Fig. 1 we present the results of our simulations for the effective exponents $`\alpha =d\mathrm{log}N/d\mathrm{log}x_{||}`$ and $`\alpha _a=d\mathrm{log}N_a/d\mathrm{log}x_{||}`$ as a function of $`x_{||}`$. The numerical exponent $`\alpha `$ nicely agrees with the analytical results. The exponent $`\alpha _a`$ is less clean due to the presence of holes near the boundary of the avalanche region. The exponent seems first to overshoot to a value of almost $`0.6`$ but then goes down so that in the end of the range of our simulations, $`x_{||}30000`$, it is consistent with our theoretical prediction $`\alpha _a=1/2`$.
Unlike its deterministic cousin, the stochastic directed sandpile model exhibits a non-trivial scaling even in one dimension. The 1D deterministic directed sandpile model is trivial in the sense that there is just one SOC configuration and the addition of a grain of sand always results in an avalanche of $`L_{||}`$ topplings in which this grain is transported and discarded at the open boundary of the system. A 1D stochastic directed sandpile model can have several variants of simple microscopic toppling rules. In one variant, which is essentially identical to the model studied by one of us in , once a height at a given site exceeds one, either one or two grains are transferred to the nearest neighbor down the slope. It is easy to see (for details see ) that this model is equivalent to a 1D random walk so that $`N_a=\mathrm{const}`$, while the typical number of topplings $`N`$ scales as a function of $`x_{||}=x`$ as $`N(x)x^{1/2}`$. The distribution of avalanche spatial length in this model has an exponent $`\tau _{||}=3/2`$, while that of avalanche volume – $`\tau =4/3`$. In another variant of the 1D stochastic toppling rules an unstable site always loses two grains of sand and each of these grains with equal probability goes to the nearest or next-nearest neighbors down the slope.
As in the DR model the upper critical dimension for the stochastic directed sandpile model is $`d_u=3`$. In this dimension the expected number of topplings at each site in a layer $`x_{||}`$ grows only logarithmically with $`x_{||}`$. Therefore, $`\alpha =\alpha _a`$ apart from the logarithmic corrections. From Eq. (1) in this case we get $`\alpha =(1+\alpha )/2`$, which has the solution $`\alpha =1`$, $`\tau _{||}=2`$, $`\tau =3/2`$. This is a standard set of mean-field exponents for any branching (avalanche) process in high enough dimension. In Fig. 2 we plot the numerical effective exponents in the 3D stochastic directed model. They agree well with the mean field values.
In conclusion, we have found an analytic solution of the stochastic directed sandpile model in any dimension. The main difference of this model from its deterministic counterpart – the Dhar-Ramaswamy model – lies in the fractal dimension of the set of active sites, i.e sites that can add or remove one grain from the overall flow of sand between two subsequent layers. Whereas in the 2D DR model in any layer there are only two sites at the edges of the interval of toppled sites which are active, in the 2D stochastic directed sandpile model each of approximately $`x_{||}^{1/2}`$ toppled sites in this interval has a 1/2 chance of being active in the above sense. This leads to an increase in the fractal dimension of an avalanche from $`D_{||}=3/2`$ to $`D_{||}=7/4`$ due to the appearance of multiple topplings. The difference between critical properties of stochastic and deterministic directed models disappears in high dimensions $`d3`$, where multiple topplings in a stochastic directed sandpile become prohibitively unlikely and all exponents acquire their mean-field values.
|
warning/0005/cond-mat0005464.html
|
ar5iv
|
text
|
# Finite-temperature ordering in two-dimensional magnets
## I Introduction
In recent years the increasing interest in the physics of low-dimensional magnets has led to a deep analysis of the thermodynamic behaviour of two-dimensional quantum antiferromagnets and, in particular, of the isotropic quantum Heisenberg antiferromagnet on the square lattice (QHAF). Such model has in fact been widely used to describe the magnetic properties of many quasi two-dimensional real compounds, from the $`S=1/2`$ cuprate La<sub>2</sub>CuO<sub>4</sub> to the $`S=5/2`$ compound Rb<sub>2</sub>MnF<sub>4</sub>, recently studied by Lee et al..
Different theoretical methods have been used to examine the rich reservoir of experimental data and the picture of the subject is now well focused, albeit with some shaded parts. In particular open questions still exist on the low-temperature region, where the spin correlation length becomes of the order of $`10^2`$ lattice spacings and the real magnets are seen to develop macroscopic areas of correlated spins.
The theoretical debate on the low-temperature regime has been mainly dedicated to the isotropic QHAF, but real compounds are not actually well described by such isotropic model when temperature is lowered: the experimental evidence of a finite-temperature transition, opposite to the Mermin-Wagner theorem assertion that such a transition cannot occur in the two-dimensional isotropic QHAF, suggests that three-dimensional correlations and possible anisotropy effects, as well as a combination of both, must be considered.
The magnetic structure of layered real compounds is such that the exchange integral $`J`$ for neighbouring spins belonging to the same plane is orders of magnitude larger than that for neighbouring spins on different planes , hereafter called $`J^{}`$; one would hence naively expect the magnetic properties to be those of an effective two-dimensional magnet down to temperatures of the order of $`J^{}`$, until the transition towards an ordered three-dimensional phase should take place. However, the experimentally observed transition occurs at a critical temperature of the order of $`J`$, signalling the transition itself to be driven by the intra-layer exchange interaction; details of such interaction, such as possible easy-axis or easy-plane anisotropies, are hence fundamental in the analysis of the critical behaviour.
Several works (see Ref. for a review) have shown that, besides the superexchange interaction, there exist further interaction mechanisms whose effects may be taken into account by inserting proper anisotropy terms in the magnetic Hamiltonian; in particular, the transition observed in K<sub>2</sub>NiF<sub>4</sub> ($`S=1`$), Rb<sub>2</sub>FeF<sub>4</sub> ($`S=2`$), K<sub>2</sub>MnF<sub>4</sub>, Rb<sub>2</sub>MnF<sub>4</sub> ($`S=5/2`$), and others, is seen to be possibly due to an easy-axis anisotropy. Such anisotropy has often been described in the literature via an external staggered magnetic field: This choice, despite allowing a qualitative description of the experimental data, lacks the fundamental property of describing a genuine phase transition, as the field explicitly breaks the symmetry and makes the model ordered at all temperatures.
In order to produce a second order transition known to be due to an easy-axis anisotropy, it is actually appropriate to insert such anisotropy in the form of an exchange one, thus preserving the symmetry under inversion along the easy-axis, whose spontaneous breaking manifests itself in the transition: subject of this paper is the study of the thermodynamic properties of the resulting model, hereafter called Easy-Axis Quantum Heisenberg Antiferromagnet (EA-QHAF).
In Sec.II we define the model and discuss its general properties, posing the problems we want to address. The method used is briefly described in Sec.III, where results for several thermodynamic quantities and different values of the anisotropy are shown and commented. In Sec.IV we compare our results with the available experimental data for the staggered magnetization and susceptibility, and for the correlation length, of the $`S=5/2`$ real compound Rb<sub>2</sub>MnF<sub>4</sub>, while conclusions are drawn in Sec.V.
## II Model and Method
The EA-QHAF on the square lattice is described by the Hamiltonian
$$\widehat{}=\frac{J}{2}\underset{𝐢,𝐝}{}\left[\mu \left(\widehat{S}_𝐢^x\widehat{S}_{𝐢+𝐝}^x+\widehat{S}_𝐢^y\widehat{S}_{𝐢+𝐝}^y\right)+\widehat{S}_𝐢^z\widehat{S}_{𝐢+𝐝}^z\right]$$
(1)
where $`𝐢=(i_1,i_2)`$ runs over the sites of a square lattice, $`𝐝`$ connects each site to its four nearest neighbours, $`J>0`$ is the antiferromagnetic exchange integral and $`\mu `$ is the anisotropy parameter ($`0\mu <1`$ for easy-axis models). The spin operators $`\widehat{S}_𝐢^\alpha `$ ($`\alpha =x,y,z`$) are such that $`|\widehat{𝐒}|^2=S(S+1)`$ and obey $`[\widehat{S}_𝐢^\alpha ,\widehat{S}_𝐣^\beta ]=i\epsilon _{\alpha \beta \gamma }\delta _{\mathrm{𝐢𝐣}}\widehat{S}_𝐢^\gamma `$.
When $`\mu =1`$ the model looses its easy-axis character and reduces to the isotropic QHAF.
The $`\mu =0`$ case will be hereafter called Ising limit, not to be confused with the genuine Ising model, reproduced by Eq.(1) with $`\mu =0`$ and $`S=1/2`$. Despite being a very particular case of Eq.(1), the two dimensional Ising model on the square lattice is a fundamental point of reference for the study of the thermodynamic properties of the EA-QHAF.
In particular, a renormalization-group analysis of the classical counterpart of the model Eq.(1) foresees the occurrence of an Ising-like transition at finite temperature for any value of $`\mu `$, no matter how near to the isotropic value $`\mu =1`$; this analysis is supported by several works based on classical Monte Carlo simulations.
In the quantum case, however, no information is given about the value of the critical temperature $`T_\mathrm{c}(\mu ,S)`$ as a function of the anisotropy and of the spin, save the fact that $`T_\mathrm{c}(0,1/2)=0.567`$ and $`T_\mathrm{c}(1,S)=0`$. We do not hence know whether or not the small anisotropy ($`|\mu 1|10^2`$) observed in real compounds can be responsible of transitions occurring at critical temperatures of the order of $`J`$, given also the fact that we expect quantum fluctuations to lower the critical temperature with respect to the classical case.
We have developed a quantitative analysis of several thermodynamic properties of the model, by means of the semiclassical method called pure-quantum self-consistent harmonic approximation (PQSCHA), already succesfully applied to many magnetic systems in one and two dimensions. The method reduces quantum expressions for statistical averages to effective classical-like ones, containing temperature and spin-dependent renormalization parameters. The thermodynamics of the effective model can then be studied by means of classical techniques, like Monte Carlo simulations.
The PQSCHA is known to be particularly suitable for anisotropic (easy-plane or easy-axis) models, but it also gives very good results in the isotropic QHAF so that we can confidently use it to investigate the possible crossover between a Heisenberg-like behaviour at high temperature and an Ising-like one near $`T_\mathrm{c}`$; such crossover is detected in the experimental data to such an extent that, well above the transition, real compounds can be satisfactorily described by the isotropic QHAF .
## III Results
The main output of the PQSCHA is the effective Hamiltonian appearing in all statistical averages, whose expression for the model described by Eq.(1) reads
$`_{\mathrm{eff}}`$ $`=`$ $`{\displaystyle \frac{1}{2}}J\stackrel{~}{S}^2{\displaystyle \underset{𝐢,𝐝}{}}\left[\theta _{}^4\mu \left(s_𝐢^xs_{𝐢+𝐝}^x+s_𝐢^ys_{𝐢+𝐝}^y\right)+\theta _{}^2\theta _{}^2s_𝐢^zs_{𝐢+𝐝}^z\right]`$ (2)
$`+`$ $`𝒢(t,\stackrel{~}{S}),`$ (3)
where $`𝐬=(s^x,s^y,s^z)`$ is a classical unit vector, $`\stackrel{~}{S}=S+1/2`$, and $`t=T/J\stackrel{~}{S}^2`$ is the reduced temperature hereafter used. The appearance of the minus sign in front of the effective Hamiltonian is due to the fact that in Eq.(3), as in all the classical-like expressions reported below, spins belonging to one of the two sublattices have been flipped, being this an innocuous operation at a classical level.
The renormalization parameters are
$$\theta _{}^2=1\frac{𝒟_{}}{2},\theta _{}^2=1\frac{𝒟_{}}{2},$$
(4)
where the coefficients
$`𝒟_{}`$ $`=`$ $`{\displaystyle \frac{1}{N\stackrel{~}{S}}}{\displaystyle \underset{𝐤}{}}{\displaystyle \frac{a_𝐤}{b_𝐤}}\left(1\mu \gamma _𝐤\right)_𝐤,`$ (5)
$`𝒟_{}`$ $`=`$ $`{\displaystyle \frac{1}{N\stackrel{~}{S}}}{\displaystyle \underset{𝐤}{}}{\displaystyle \frac{a_𝐤}{b_𝐤}}\left(1{\displaystyle \frac{\gamma _𝐤}{\mu }}\right)_𝐤`$ (6)
are self-consistently determined by solving Eqs.(4) and (6), with
$$a_𝐤^2=4\left(\theta _{}^2+\mu \theta _{}^2\gamma _𝐤\right),b_𝐤^2=4\left(\theta _{}^2\mu \theta _{}^2\gamma _𝐤\right),$$
(7)
and
$$\gamma _𝐤=\frac{1}{4}\underset{𝐝}{}e^{i𝐤𝐝},_𝐤=\mathrm{coth}f_𝐤\frac{1}{f_𝐤},f_𝐤=\frac{a_𝐤b_𝐤}{2\stackrel{~}{S}t},$$
k being the wave vector in the first Brillouin zone.
The temperature and spin dependent uniform term
$$𝒢(t,\stackrel{~}{S})=2J\stackrel{~}{S}^2(1\theta _{}^2\theta _{}^2)+J\stackrel{~}{S}^2t\underset{𝐤}{}\mathrm{ln}\left(\frac{\mathrm{sinh}f_𝐤}{\theta _{}^2f_𝐤}\right),$$
does not enter the expressions for the statistical averages, but contributes to the free energy and to the related thermodynamic quantities.
The renormalization coefficients $`𝒟_{}`$ and $`𝒟_{}`$ measure the pure-quantum fluctuations parallel and perpendicular to the easy-axis, respectively, of one spin with respect to its nearest neighbours, and vanish in both the high-temperature and the classical $`S\mathrm{}`$ limit. It is worthwhile noticing that the fluctuations along the easy-axis only enters the renormalization of the $`z`$-component of the exchange interaction.
Defining the effective exchange integral and effective anisotropy
$$J_{\mathrm{eff}}=J\theta _{}^2\theta _{}^2,\mu _{\mathrm{eff}}=\mu \frac{\theta _{}^2}{\theta _{}^2}$$
Eq.(3) can be written in the form
$`_{\mathrm{eff}}`$ $`=`$ $`{\displaystyle \frac{1}{2}}J_{\mathrm{eff}}\stackrel{~}{S}^2{\displaystyle \underset{𝐢,𝐝}{}}\left[\mu _{\mathrm{eff}}\left(s_𝐢^xs_{𝐢+𝐝}^x+s_𝐢^ys_{𝐢+𝐝}^y\right)+s_𝐢^zs_{𝐢+𝐝}^z\right]`$ (8)
$`+`$ $`𝒢(t,\stackrel{~}{S}),`$ (9)
to make evident that the PQSCHA leads to a classical-like effective model of the same form of the original quantum one, whose thermodynamic properties can be studied by classical numerical technique, properly taking into account the spin and temperature dependence of the renormalized parameters $`J_{\mathrm{eff}}`$ and $`\mu _{\mathrm{eff}}`$. As $`\theta _{}^2<\theta _{}^2<1`$, it is $`J_{\mathrm{eff}}<J`$ and $`1>\mu _{\mathrm{eff}}>\mu `$, so that quantum effects are seen to cause the weakening of both the exchange interaction and the easy-axis anisotropy; in the isotropic limit $`𝒟_{}=𝒟_{}`$ and $`\mu _{\mathrm{eff}}=\mu =1`$.
The PQSCHA expression for the statistical average of a quantum operator $`\widehat{O}`$ is
$$\widehat{O}=\frac{1}{𝒵_{\mathrm{eff}}}d^N𝐬O_{\mathrm{eff}}e^{\beta _{\mathrm{eff}}}$$
(10)
where $`\beta =T^1`$, $`𝒵_{\mathrm{eff}}=d^^N𝐬e^{\beta _{\mathrm{eff}}}`$; $`O_{\mathrm{eff}}`$ is determined by the same procedure leading to $`_{\mathrm{eff}}`$ and contains temperature and spin dependent renormalizations; after Eq.(10) most thermodynamic quantities may be written in a particularly suggestive form in terms of classical-like statistical averages defined by the effective Hamiltonian
$$\mathrm{}_{\mathrm{eff}}=\frac{1}{𝒵_{\mathrm{eff}}}d^N𝐬(\mathrm{})e^{\beta _{\mathrm{eff}}};$$
as far as the evaluation of $`\mathrm{}_{\mathrm{eff}}`$ is concerned, one must keep in mind that, because of the temperature dependence of the effective anisotropy appearing in $`_{\mathrm{eff}}`$, each point in temperature corresponds to a different effective model. This means that, if the classical Monte Carlo technique is used, as done in this work, the simulated model changes with temperature, so that, at variance with the isotropic case, no existing classical data can be used and a complete series of ad hoc simulations must be carried on. Nevertheless, the computational effort required is still that of a classical simulation, as the evaluation of the renormalized parameters is a matter of a few seconds on a standard PC.
The application of the PQSCHA to the EA-QHAF leads to the following results for the thermodynamic quantities we have considered:
* Internal energy
$`u\widehat{}/(NJ\stackrel{~}{S}^2)`$:
$$u=\frac{1}{NJ\stackrel{~}{S}^2}_{\mathrm{eff}}_{\mathrm{eff}}𝒢(t,\stackrel{~}{S})+(t,\stackrel{~}{S}),$$
(11)
where
$$=2(\theta _{}^2\theta _{}^2)\left[\theta _{}^2+\frac{2}{1\mu ^2}(\theta _{}^2\theta _{}^2)\right]$$
is a negative zero-point quantum correction term due to the anisotropy (it vanishes in the isotropic limit).
* Staggered magnetization
$`m_𝐢(1)^{i_1+i_2}\widehat{S}_𝐢^z/N\stackrel{~}{S}`$:
$$m=\theta _{}^2s_𝐢^z_{\mathrm{eff}}+(t,\stackrel{~}{S});$$
(12)
where the magnetization renormalization is seen to be due to an effective spin reduction ($`\theta _{}^2<1`$)) and to the appearance of the negative term $`=(\theta _{}^2\theta _{}^2)/(1\mu ^2)`$ which vanishes in the isotropic limit and is finite for $`\mu =0`$.
* Staggered correlation function above $`T_\mathrm{c}`$ $`G(r)(1)^{r_1+r_2}\widehat{𝐒}_𝐢\widehat{𝐒}_{𝐢+𝐫}/\stackrel{~}{S}^2`$ with $`𝐫=(r_1,r_2)`$ any vector of the square lattice and $`r=|𝐫|`$:
$$G(r)=\theta _𝐫^4𝐬_𝐢𝐬_{𝐢+𝐫}_{\mathrm{eff}},$$
(13)
where $`\theta _𝐫^2=1𝒟_𝐫/2`$ and
$$𝒟_𝐫=\frac{1}{N\stackrel{~}{S}}\underset{𝐤}{}\frac{a_𝐤}{b_𝐤}\left(1\mathrm{cos}(𝐤𝐫)\right)_𝐤$$
is a further site-dependent renormalization coefficient. For increasing $`r`$, the coefficient $`𝒟_𝐫`$ rapidly converges to a uniform term, so that the asymptotic ($`r\mathrm{}`$) behaviour of $`G(r)`$ is actually determined by that of the effective classical-like correlation function $`𝐬_𝐢𝐬_{𝐢+𝐫}_{\mathrm{eff}}`$.
* Staggered static susceptibility above $`T_\mathrm{c}`$ $`\chi _𝐫G(r)/3`$:
$$\chi =\frac{1}{3}\left[\frac{S(S+1)}{\stackrel{~}{S}^2}+\underset{𝐫0}{}G(r)\right]$$
(14)
* Correlation length above $`T_\mathrm{c}`$ We have determined the correlation length $`\xi `$ by fitting $`G(r)`$ with the expression proposed by Serena,Garcia and Levanyuk
$$G(r)\frac{1}{\xi ^{\frac{1}{4}}}\frac{e^{r/\xi }}{\left(r/\xi \right)^{\frac{1}{2}}+\left(r/\xi \right)^{\frac{1}{4}}}$$
(15)
which interpolates the two asymptotic behaviours $`r\mathrm{}`$ and $`r0`$ of the Ising model.
In what follows, we show our results as obtained combining the above PQSCHA expressions with the numerical output of the classical Monte Carlo simulations we have performed to evaluate the effective statistical averages $`\mathrm{}_{\mathrm{eff}}`$. At variance with the isotropic case, where results for different values of the spin are obtained by the same series of classical simulations, we now have to fix the value of the spin in order to determine, for a given value of $`\mu `$, the corresponding $`\mu _{\mathrm{eff}}`$ to be used in the simulation. We have hence concentrated ourselves on the case $`\mu =0.9942`$ and $`S=5/2`$, being these the anisotropy and spin values corresponding to the real compound Rb<sub>2</sub>MnF<sub>4</sub>; the more anisotropic case $`\mu =0.7`$ and $`S=5/2`$ has also been considered, because of its expectedly more marked Ising-like features.
In Fig.1 and 2 we show the internal energy and specific heat, versus temperature, for both values of $`\mu `$, compared with the isotropic case $`\mu =1`$. As the value of the critical temperature increases for larger anisotropy, but diminishes when smaller spin values are considered, we avoid the confusing direct comparison between our curves ($`\mu >0`$,$`S=5/2`$) and the correponding quantities for the Ising model ($`\mu =0`$,$`S=1/2`$), by showing the latter as Insets.
The Ising character of the $`\mu =0.7`$ model is evidenced by the pronounced peak in the specific heat, corresponding to a qualitatively different temperature dependence of the internal energy with respect to that of the isotropic model. Although such difference is almost not perceptible when the $`\mu =0.9942`$ curve is considered, a clear peak in the specific heat is still present, testifying to a persistence of the Ising-like behaviour even in this quasi-isotropic model.
The same conclusion is drawn when the staggered magnetization is considered: in Fig.3 we see that for $`\mu =0.9942`$ there exist a wide temperature range where the system is ordered, with a critical temperature $`t_\mathrm{c}`$ that, despite being lower than the one relative to the $`\mu =0.7`$ case, is still of the order of $`J`$. In the inset we show the magnetization curves normalized to their saturation values, as functions of $`t/t_\mathrm{c}`$, together with that of the Ising model: It is evident that an increased anisotropy causes a sharpening of the way the magnetization vanishes.
The critical temperatures hereafter used are $`t_\mathrm{c}=0.785`$ for $`\mu =0.7`$ and $`t_\mathrm{c}=0.575`$ for $`\mu =0.9942`$; these values have been determined by locating at best the correlation length divergence and consistently coincide with those emerging from the analysis of the critical behaviour of other quantities. For instance the critical temperatures determined by fitting the magnetization curves (see the dotted lines in Fig.3) are $`t_\mathrm{c}^{}=0.787`$ for $`\mu =0.7`$ and $`t_\mathrm{c}^{}=0.576`$ for $`\mu =0.9942`$.
It is to be noticed that even below $`t_\mathrm{c}`$, where the finite value of the magnetization would suggest a complete predominance of the Ising character, the system does actually display features which are distinctive of the isotropic model. In particular, the specific heat for both values of $`\mu `$ shows, after an exponential start typical of a gapped dispertion relation, a change in the curvature and an almost linear temperature dependence, due to the excitation of long-wavelength low-energy modes of the same type of those characterizing the isotropic model. It is just in the vicinity of $`t_\mathrm{c}`$ that a new change in the curvature announces the forthcoming transition.
In Fig.4 we show the correlation function $`G(r)`$ as a function of $`r`$ in the quasi-isotropic case $`\mu =0.9942`$ and for three different temperatures. The fit of our data with the Serena-Garcia-Levanyuk function Eq.(15) is very good in the whole temperature range examined, and hence leads to a clean evaluation of the correlation length.
In Fig.5 we show the correlation length and also report the curve for the isotropic model: we notice that the $`\mu =0.9942`$ curve lays on the isotropic one up to correlation lengths of the order of $`20`$ lattice spacings (i.e. $`t1.03t_\mathrm{c}`$), while for $`\mu =0.7`$ a deviation is evident already for $`\xi 2`$ (i.e. $`t1.3t_\mathrm{c}`$). This means that, in the former case, there is a temperature region where the model significantly behaves like the isotropic model, as far as the correlation length is concerned, and it is hence meaningful to introduce the idea of a crossover from a Heisenberg- to an Ising-like regime; on the contrary, when $`\mu =0.7`$ the Ising character of the model is manifest already when correlations over a few lattice spacings develop.
## IV Comparison with experimental data
The results shown in previous section qualitatively explain the mechanism possibly underlying the finite-temperature ordering experimentally observed in many quasi-two dimensional real magnets.
As for a more precise quantitative analysis, we have concentrated ourselves on the $`S=5/2`$ magnet Rb<sub>2</sub>MnF<sub>4</sub>: reason for this choice is the availability of recent neutron scattering data relative to such compound and the fact that, because of its cristallographic structure, Rb<sub>2</sub>MnF<sub>4</sub> is known to behave as a two-dimensional magnet both above and below the observed transition. This means that the critical behaviour is not contaminated by the onset of three-dimensional order and a clean characterization of the transition is possible, as well as a meaningful comparison with the experimental data for the magnetization below $`T_\mathrm{c}`$.
The model parameters $`J_\mathrm{s}=7.62\pm 0.09\mathrm{K}`$ and $`\mu _\mathrm{s}=0.9953`$ available in the literature for the compound Rb<sub>2</sub>MnF<sub>4</sub> are obtained by fitting the extrapolated $`T0`$ experimental data for the spin-wave frequencies with the expression
$$\omega _𝐤=4J_\mathrm{s}S\sqrt{\frac{1}{\mu _\mathrm{s}^2}\gamma _𝐤^2};$$
(16)
this means that $`J_\mathrm{s}`$ and $`\mu _\mathrm{s}`$ are renormalized by the zero-point quantum fluctuations and are not the bare values to be inserted in Eq.(1). These have hence been determined equating Eq.(16) with the zero-temperature dispersion relation relative to the EA-QHAF as given by the PQSCHA
$$\omega _𝐤=4J\stackrel{~}{S}\mu \theta _{}^2(0)\sqrt{\frac{\theta _{}^4(0)}{\mu ^2\theta _{}^4(0)}\gamma _𝐤^2}$$
(17)
where $`\theta _{}^2(0)`$ and $`\theta _{}^2(0)`$ are the renormalization parameters defined in Eq.(4) evaluated at $`t=0`$.
The resulting equation for $`\mu `$
$$\mu =\mu _s\frac{\theta _{}^2(0)}{\theta _{}^2(0)}$$
must be self-consistently solved, as both $`\theta _{}`$ and $`\theta _{}`$ depend on $`\mu `$, and gives $`\mu =0.9942`$.
Once $`\mu `$ is determined, the equation for the exchange integral
$$J=\frac{S}{\stackrel{~}{S}}\frac{J_s}{\mu \theta _{}^2(0)}$$
is straightforwardly solved and gives $`J=7.42\mathrm{K}`$;
In Fig.6 our results for the staggered magnetization are shown together with the experimental data from Ref. and the interpolating curve there proposed. Besides the overall agreement in the whole ordered phase, it is to be noticed that our prediction for the value of the critical temperature perfectly coincides with the one deriving from the experimental analysis, which gives $`T_\mathrm{c}=38.4\mathrm{K}`$ (i.e. $`t_\mathrm{c}=0.575`$); such a large value of $`T_\mathrm{c}`$ with respect to the exchange integral $`J=7.42\mathrm{K}`$, should not surprise, as the squared of the spin value has been actually extracted from the latter.
In order to obtain the best quantitative description of the EA-QHAF in the paramagnetic phase, and given the small anisotropy of Rb<sub>2</sub>MnF<sub>4</sub>, the lowest-temperature data for the staggered susceptibility and correlation length, shown in Figs.7 and 8, are determined by the PQSCHA version introduced in Ref. and there shown to be the most appropriate to study the isotropic model. The difference between such version and the original one described in Sec.II consists in the appearance of the renormalization parameters $`\kappa _,^2`$, instead of $`\theta _,^2`$, in Eqs.(7), with $`\kappa _,^2=\theta _,^2𝒟_,^{\mathrm{cl}}/2`$, $`𝒟_,^{\mathrm{cl}}`$ being the renormalization coefficients determined by the classical self-consistent harmonic approximation.
The agreement between our results and the experimental data is indeed noticeable, given also the fact that no best-fit procedure has been used. In addition, our results represent a clear improvement with respect to those coming from a mean-field treatment of the anisotropy, proposed by Keimer et al., which enables one to derive the correlation length of the anisotropic model directly from the data of the isotropic one, as done in Ref. starting from the PQSCHA results for the isotropic model itself. In particular, it is evident from Fig.8 that the mean-field approach, apart from accounting for the existence of the phase transition, leads to an overestimate of the critical temperature, while the model with exchange anisotropy gives a very accurate estimate of $`t_c`$.
## V Conclusions
In this work we have studied the easy-axis quantum Heisenberg antiferromagnet on the square lattice by means of the pure-quantum self-consistent harmonic approximation; expressions for several quantum statistical averages have been determined for the general model, with any value of the spin and of the anisotropy. The numerical work, consisting in classical Monte Carlo simulations on a properly renormalized model, have been concentrated on the $`S=5/2`$, $`\mu =0.7`$ and $`S=5/2`$, $`\mu =0.9942`$ cases, the latter corresponding to the real compound Rb<sub>2</sub>MnF<sub>4</sub>.
We have shown that a finite temperature transition is present in both cases and that such transition is cleary of an Ising type; the value of the corresponding critical temperature have been determined by the analysis of the temperature dependence of the correlation length and has been always found to perfectly agree with that extracted by the analysis of other thermodynamic quantities.
Despite the essential presence of the transition, the EA-QHAF does also display, both below and above $`t_\mathrm{c}`$, features which are typical of the isotropic model. In the ordered phase it is the specific heat behaviour that testify to the existence of long-wavelength low-energy isotropic-like excitations. On the other hand, when the critical region is abandoned in the paramagnetic phase, the anisotropy looses its fundamental role and a crossover towards the isotropic behaviour is observed, at least as far as the correlation length is concerned; such crossover, however, has a weaker meaning for larger anisotropy, being confined, already for $`\mu =0.7`$, to the high temperature region where $`\xi `$ is the order of the lattice spacing and differences between different models become irrelevant.
We have compared our theoretical results with the neutron scattering experimental data for the staggered magnetization, staggered susceptibility and correlation length of the real compound Rb<sub>2</sub>MnF<sub>4</sub> and found an excellent agreement both for the overall temperature behaviour and for the value of the critical temperature.
We can hence conclude that the experimentally observed finite temperature transition in Rb<sub>2</sub>MnF<sub>4</sub> is due to an easy-axis anisotropy in the intra-layer exchange interaction and that, despite the small value of the anisotropy, the compound shows an Ising-like critical behaviour.
## VI Acknowledgments
We acknowledge Prof. R. J. Birgeneau and Dr. Y. S. Lee for sending us preprints and experimental data.
This work has been partially supported by the COFIN98-MURST fund.
|
warning/0005/astro-ph0005586.html
|
ar5iv
|
text
|
# Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and GalaxiesBased in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555.
## 1. Introduction
It is well known that individual galaxies fall into a fairly clear–cut morphological classification scheme, as first proposed by Hubble (1936). Over the decades, Hubble’s scheme has been expanded upon and exploited for quantifying galactic evolution, galactic stellar populations and star formation efficiencies, the relative gas–phase component to stellar component in galaxies, and the effects of environment on galaxy formation. Similarly, one of the central motivations for studying intervening quasar absorption lines, especially those selected by the presence of metal lines, is that they also provide insights into galactic evolution, not of the stars and stellar dynamics, but of the chemical, ionization, and kinematic conditions of interstellar and halo gas.
Historically, absorption line systems have been classified in a taxonomic system by their gas cross sections. Categorized by increasing Hi column densities are the Ly$`\alpha `$ forest clouds, with $`N(\text{H}\text{i})10^{15}`$ cm<sup>-2</sup>, the sub–Lyman limit systems, having $`N(\text{H}\text{i})10^{16}`$ cm<sup>-2</sup>, the Lyman limit break systems, with $`N(\text{H}\text{i})10^{17.3}`$ cm<sup>-2</sup>, and the damped Ly$`\alpha `$ systems, having $`N(\text{H}\text{i})10^{20.3}`$ cm<sup>-2</sup>. Metal–line systems are either selected by the presence of strong Mgii $`\lambda \lambda 2796,2803`$ absorption or strong Civ $`\lambda \lambda 1548,1550`$ absorption.
What has not been explored, however, is the taxonomy of absorption line systems when the Hi, Mgii, and Civ absorption strengths and the gas kinematics are equally considered. Ultimately, categorizing the relationships between several different absorption properties may provide clues central to understanding the different physical natures of the various types of systems.
Furthermore, while our empirical knowledge of the absorption strengths, kinematics, and physical extent of galaxies selected by Mgii absorption has steadily progressed over the last decade (e.g. Bergeron & Boissé (1991); Lanzetta & Bowen (1990), 1992; Bergeron et al. (1992); Le Brun et al. (1993); Steidel et al. (1994); Churchill et al. (1996); Guillemin & Bergeron (1997)), little is directly known about Hi absorption and higher ionization absorption (esp. Siiv, Civ, Nv, and Ovi) in these galaxies.
Arguably, the Mgii–selected systems are ideally suited for a taxonomic study of absorption systems because:
(1) those with $`W_r(\text{Mg}\text{ii})0.3`$ Å are known to be directly associated with galaxies (Bergeron & Boissé (1991); Steidel, Dickinson, & Persson 1994; Churchill et al. (1996)) and/or sub–galactic metal–enriched environments (Yanny (1992); Yanny & York (1992)). HST imaging has revealed that these galaxies have a wide variety of line–of–sight orientations and “normal” morphologies (see Steidel et al. (1997); Steidel (1998)). Since magnesium is an $`\alpha `$–group element yielded by Type II supernovae, it is expected that the association with galaxies will hold to the highest redshifts.
(2) they arise in structures having a five decade range of Hi column densities, including sub–Lyman limit systems (Churchill et al. 1999a ; Churchill et al. 2000, hereafter Paper I), Lyman limit systems (e.g. Steidel & Sargent (1992); Paper I), and damped Ly$`\alpha `$ systems (e.g. Le Brun et al. (1997); Rao & Turnshek (1998); Boissé et al. (1998)) which means that a large range of galactic environments will be sampled.
(3) for $`z<2.2`$ their statistical properties and for $`z<1.4`$ their kinematic properties have been thoroughly documented (e.g. Lanzetta, Turnshek, & Wolfe (1987); Sargent, Boksenberg, & Steidel (1988); Steidel & Sargent (1992); Petitjean & Bergeron (1990); Steidel & Sargent (1992); Churchill (1997); Churchill et al. 1999a ) which means that the low redshift database is already in place and can be used for evolution studies when higher redshift data are obtained.
(4) they are seen to give rise to a range of Civ and other high ionization absorption in UV space–based spectra (Bergeron et al. (1994); Paper I), which means that the more general ionization conditions can be studied in detail (e.g. Churchill & Charlton (1999)).
In this paper, we study the combined available data on a sample of 45 Mgii absorption–selected systems at redshifts $`0.4`$$`1.4`$. We focus on the Mgii and Feii absorption strengths and kinematics, the Ly$`\alpha `$ and Civ absorption strengths (Paper I), and the ground–based derived galaxy luminosities, colors, and impact parameters. We perform a multivariate analysis (Babu & Feigelson (1996); Johnson & Wichern (1992)) on the absorption line data in order to objectively quantify systematic differences between and/or groupings of the high ionization and neutral hydrogen absorption properties of Mgii selected systems.
The paper is organized as follows: In § 2, we describe the data, the sample selection, and the data analysis. In § 3, we investigate the variation in the high and low ionization properties of intermediate redshift Mgii absorbers using multivariate analysis. We introduce a taxonomy that serves as an objective guide for classifying the variations in Mgii absorber properties. In § 4, we investigate the overall ionization and kinematic conditions. In § 5 we investigate the relationship between the absorption properties and host galaxy properties and in § 6 offer a speculative discussion on the relationships between the absorber classes and their possible evolution. In § 7 we summarize the main points of this work.
## 2. The Data: Sample Selection and Analysis
We targeted quasar absorption line systems at intermediate redshifts that have been discovered by the presence of a Mgii $`\lambda \lambda 2796,2803`$ doublet (e.g. Lanzetta, Turnshek, & Wolfe (1987); Sargent, Boksenberg, & Steidel (1988); Steidel & Sargent (1992)). We also include in our study the “weak” systems \[those with $`W_r(\text{Mg}\text{ii})<0.3`$ Å\], which are more numerous in their redshift path density (Churchill et al. 1999a ). A detailed account of the sample selection is given in Paper I, but we briefly outline the sample properties here.
The Mgii absorbers were selected from a high resolution ($`6`$ km s<sup>-1</sup>) survey (Churchill (1997)) using the HIRES spectrograph (Vogt et al. (1994)) on the Keck I telescope. For 45 of these systems, additional wavelength coverage in the ultraviolet was available in the HST/FOS archive and the database compiled by the HST QSO Absorption Line Key Project (Bahcall et al. (1993); Bahcall et al. (1996); Jannuzi et al. (1998)).
The resulting database of Mgii absorbing systems and their rest–frame equivalent widths are listed in Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555.. The first three columns are the quasar name, the absorber redshift, and the Mgii $`\lambda 2796`$ rest–frame equivalent width, $`W_r(\text{Mg}\text{ii})`$. The fourth and fifth columns list the Mgii kinematic spread, $`\omega _v`$ (see Equation 1), and the kinematic composition, i.e. the number of Voigt Profile components (see § 2.2). Columns six, seven, and eight list the Feii, Civ, and Ly$`\alpha `$ rest–frame equivalent widths. Column nine gives the various subsamples used for our analyses. Column ten lists the taxonomic “class” of each system (explained in § 3.2).
Full details of the data analysis are given in Paper I, including the continuum fitting, the line finding, the equivalent width measurements, the establishment of a redshift zero point for the FOS spectra, the procedure for line identifications, and the techniques employed to measure Lyman limit breaks, when present.
As shown in Figure 2 of Paper I, the Mgii equivalent width distribution is consistent with that of an unbiased sample of absorbers for $`W_r(\text{Mg}\text{ii})1.3`$ Å (Churchill et al. 1999a ). There are five systems with $`W_r(\text{Mg}\text{ii})>1.3`$ Å, which may appear as a bias toward strong systems. When this may be a concern for our analysis, we discuss this possibility.
With regard to detection sensitivity, the sample is 72% complete to a $`W_r(\text{Mg}\text{ii})`$ rest–frame detection threshold of $`0.02`$ Å and 93% complete to a $`0.03`$ Å threshold (Figure 3 of Paper I). In column nine of Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555., we have designated those systems having rest–frame detection thresholds greater than $`0.03`$ Å as Sample A. For these absorbers, unresolved Mgii absorption features with $`W_r(\text{Mg}\text{ii})0.03`$ Å cannot be detected. The other subsample designations, “CA” and IC”, are explained in §§ 3 and 4, respectively.
### 2.1. Galaxy Sample
For roughly 60 Mgii absorbers having $`W_r(\text{Mg}\text{ii})0.3`$ Å, Steidel et al. (1994) identified associated galaxies. Only 16 of the 45 Mgii systems presented here, of which 21 have $`W_r(\text{Mg}\text{ii})0.3`$ Å, have confirmed galaxy counterparts (not all fields studied here have been imaged).
The galaxy properties were obtained from broad–band $`g(4900/700)`$, $`\mathrm{}(6930/1500)`$, $`i(8000/1450)`$ and $`K`$–band images of the QSO fields and their redshifts were spectroscopically verified to be coincident (within $`100`$ km s<sup>-1</sup>) with those of the Mgii absorbers. PSF subtraction of the quasar was performed for all fields, enabling small impact parameter galaxies to be identified. Further details of the imaging and spectroscopic observations are described in Steidel et al. (1994).
The galaxy properties are presented in Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555.; the columns from left to right are the quasar field, the galaxy redshift, the absolute $`B`$ magnitude, $`M_B`$, the absolute $`K`$ magnitude, $`M_K`$, the de–reddened $`BK`$ color, the galaxy–quasar impact parameter in $`h^1`$kpc ($`H_0=100`$ km s<sup>-1</sup> Mpc<sup>-1</sup>, $`q_0=0.05`$), and a reference if previously published. In a few cases, morphology information is available from published HST images (Le Brun et al. (1997); Steidel et al. (1997); Steidel (1998)).
It is always possible that a galaxy is misidentified, that another galaxy or more than one galaxy is giving rise to the absorption. Possible selection effects due to misidentifications and incompleteness of galaxy redshifts in the individual quasar fields are discussed by Steidel et al. (1994) and examined in detail by Charlton & Churchill (1996).
### 2.2. Data Analysis: Kinematics
To measure the kinematic “spread” of the low ionization gas directly from the flux values, we use the second velocity moment of optical depth across the Mgii $`\lambda 2796`$ profile, defined as,
$$\omega _v=\left\{\frac{\tau _a(v_i)v_i^2\mathrm{\Delta }v_i}{\tau _a(v_i)\mathrm{\Delta }v_i}\right\}^{1/2},$$
(1)
where $`\tau _a(v_i)=\mathrm{ln}[I_c(v_i)/I(v_i)]`$ is the apparent optical depth measured directly from the flux values (Savage & Sembach (1991)) at velocity $`v_i`$, $`\mathrm{\Delta }v_i=(v_{i1}v_{i+1})/2`$, $`v_i=c(\lambda _i/\lambda _{obs}1)`$, and $`\lambda _{obs}=2796.352(1+z_{\mathrm{abs}})`$. The sums are performed only over velocity intervals in which absorption features have been detected at the $`5\sigma `$ level, thus eliminating terms consistent with noise (see Paper I for details on feature detection). The quantity measured by $`\omega _v`$ is similar to the standard deviation measured from a normal distribution. The uncertainty in $`\omega _v`$ is obtained from simple error propagation (Churchill (1997); also see Sembach & Savage (1992)). The velocity zero point of each system is set at the median wavelength of the apparent optical depth distribution of the Mgii $`\lambda 2796`$ profile.
The value of $`\omega _v`$ depends upon the velocity squared, and thus its value is sensitive to the presence of very weak Mgii absorption at large velocity. Therefore, $`\omega _v`$ is sensitive to the equivalent width detection threshold of the HIRES spectra. We are 93% complete to a $`5\sigma `$ rest–frame equivalent width detection threshold of $`0.03`$ Å for the full sample of 45 systems (see Figure 3 from Paper I). To enforce uniform evaluation of the kinematic spread for the full sample, we omitted features with $`W_r(\text{Mg}\text{ii})<0.03`$ Å from the computation of $`\omega _v`$. Six systems required this censorship step; they are footnoted in Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555.. The maximum change in $`\omega _v`$ from its uncensored value was a $`20`$% difference, except for one of the absorbers (PKS $`0454+039`$ at $`z=1.1532`$), which was changed by a $`40`$% difference.
To measure the kinematic “composition” of the low ionization gas, we use Voigt profile fitting. We use our own program, MINFIT (Churchill (1997)), which performs a $`\chi ^2`$ minimization using the NETLIB–slatec routine dnls1 (More (1978)). In fully saturated profiles, the number of clouds is often underestimated by a factor of a few (Churchill (1997)). A corollary is that the cloud column densities, $`b`$ parameters, and velocities are correlative with the number of clouds. We return to this issue in § 4.2.
## 3. Characterizing Mgii Absorber Properties
We have constructed a subsample of 30 systems, Sample CA (CA = Cluster Analysis), in order to examine variations and possible trends between the absorption properties listed in Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555.. All Sample CA members have measured $`W_r(\text{Ly}\alpha )`$, $`W_r(\text{Mg}\text{ii})`$, and $`\omega _v`$ and have measurements of or limits on Feii and Civ. Systems with upper limits on $`W_r(\text{Fe}\text{ii})`$ were included in the sample because these limits are already capable of demonstrating a paucity of Feii absorption (only seven Sample CA systems have Feii limits). Small $`W_r(\text{Mg}\text{ii})`$ systems with upper limits on Civ were excluded from the sample because the limits were not always stringent. However, multiple cloud systems with limits on $`W_r(\text{C}\text{iv})`$ and larger $`W_r(\text{Mg}\text{ii})`$ are included. The members of Sample CA are given in column nine of Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555..
In Figure 1, we present three dimensional plots of (a) $`W_r(\text{Mg}\text{ii})`$ vs. $`W_r(\text{Ly}\alpha )`$ vs. $`W_r(\text{C}\text{iv})`$, (b) $`W_r(\text{Mg}\text{ii})`$ vs. $`W_r(\text{Ly}\alpha )`$ vs. $`W_r(\text{Fe}\text{ii})`$, and (c) $`W_r(\text{Mg}\text{ii})`$ vs. $`\omega _v`$ vs. $`W_r(\text{C}\text{iv})`$ for Sample CA. Note that distribution of equivalent widths are not random; there is a clear trend for $`W_r(\text{Mg}\text{ii})`$ to increase with $`W_r(\text{Ly}\alpha )`$, for example. On the other hand, it is clear that there are significant spreads, or variations in the absorption strengths. In the case of $`W_r(\text{Fe}\text{ii})`$, we see a very large range of values (Figure 1$`b`$) that trace $`W_r(\text{Ly}\alpha )`$, and to a lesser extent $`W_r(\text{Mg}\text{ii})`$. $`W_r(\text{C}\text{iv})`$ exhibits a significant spread for a given $`W_r(\text{Mg}\text{ii})`$$`W_r(\text{Ly}\alpha )`$ locus (Figure 1$`a`$). Moreover, note the groupings of $`W_r(\text{Mg}\text{ii})`$, $`\omega _v`$, and $`W_r(\text{C}\text{iv})`$ (Figure 1$`c`$).
From an empirical point of view, the qualitative appearance of groupings in Figure 1 would suggest that Mgii–selected systems can be further categorized (quantitatively) by their absorption properties. However, our sample is small, having only 30 data points, and it is not clear whether the apparent groupings would statistically be present in other “realizations” of the data<sup>1</sup><sup>1</sup>1Monte Carlo realizations of our sample could shed light on this issue if, and only if, we had a priori knowledge of the distribution functions for all the properties being studied.. This concern is best addressed by using a multivariate analysis in which all available absorption properties are incorporated simultaneously (multidimensional version of Figure 1 with one dimension for each property).
When the Ly$`\alpha `$, Feii, and Civ and Mgii kinematics are simultanously considered, do we find various “classes” of Mgii absorbers? In other words, can we quantitatively describe both the variations and the trends by considering all the absorption line data presented in Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555.? To address these questions, we applied “Tree Clustering” and “$`K`$–means Clustering” analysis to sample CA.
### 3.1. A Multivariate Analysis
Multivariate clustering analysis algorithms are designed to organize data of many variables into catagories so that natural groupings of the data can be examined in a completely unbiased manner (i.e. no model is imposed upon the data). Full details of these techniques can be found elsewhere (e.g. Johnson & Wichern 1992; Babu & Feigelson 1996); below, we provide limited background material. We used the STATISTICA software package (www.statsoft.com).
#### 3.1.1 Tree Cluster Analysis
In tree cluster analysis, the data occupy a multi–dimensional space, one dimension for each measured absorption property. Clustering algorithms compute the “distances” between each pair of points (absorbers) and then amalgamate them into clusters. We note that tree cluster analysis is not subject to significance testing because the result itself is the most significant solution under the assumption of no a priori hypothesis regarding the data.
The amalgamation process begins with each absorber in a unique class by itself and then proceeds by relaxing the criterion of uniqueness in subsequent steps. With each step, the algorithm amalgamates larger and larger clusters of increasingly disimilar properties, until in the final step all absorbers are joined together in a single class. Graphically, clusters appear as distinct “branches” in a hierarchical tree, with similarities linked at nodes. We used Euclidean distances \[i.e. $`(_i[x_iy_i]^2)^{1/2}`$\] and Ward’s method (Ward 1963) for the amalgamation algorithm. Ward’s method minimizes the sum of squares of the distances between clusters. The combination of Euclidean distances and Ward’s amalgamation rule uses an analysis of variances approach for which $`N(0,1)`$ standardization of the data is optimal. With this standardization, each variable is baselined and scaled to have a zero mean and unity standard deviation. $`N(0,1)`$ standardization ensures that the distance between any pair of points in multidimensional space is not biased by a large dynamic range in one or more of the variables (dimensions).
#### 3.1.2 $`K`$–means Clustering
In $`K`$–means clustering analysis, we begin with the assumption of a set number of clusters and the algorithm finds the most significant, or distinct, clusters possible. Starting with $`K`$ random clusters, the algorithm moves points (absorbers) between clusters until both the variability within clusters is minimized and the variability between clusters is maximized. We find that $`K=5`$ is the highest number of clusters allowed such that all clusters are considered significant, based upon MANOVA tests (Johnson & Wichern (1992); also see Mukherjee et al. (1998)). For five clusters, the MANOVA probabilities, $`p`$, for accepting a result as valid, (i.e. representative of a unique population) are highly significant. We obtained $`p<10^5`$ for each cluster.
### 3.2. Results
In Figure 2, we present a dendrogram, the tree diagram showing the results of our cluster analysis. Along the bottom horizontal axis are the individual Mgii systems identified by quasar and absorber redshift. The vertical axis is the linkage distance, $`LD`$, the distance between clusters. The larger the value of $`LD`$ between two branches, the less related are the objects on each branch.
At $`LD16`$ there is a natural grouping into three clusters. These three clusters are predominantly distinguished by very strong Ly$`\alpha `$ \[right branch, labeled “DLA/Hi–rich”\], weaker Civ \[center branch, labeled “Civ–weak”\], and intermediate to stronger Civ \[left branch, labeled “Civ–strong”\] absorption.
Further insight into the physical differences in the absorbers can be obtained if we adopt a “less significant” linkage distance. At $`LD4`$ there is a secondary grouping into six clusters. The Civ–strong and Civ–weak clusters have each been separated by variations in Mgii strengths and kinematics and the DLA/Hi–Rich cluster has been separated by a spread in the Ly$`\alpha `$ strengths. However, we treat the DLA/Hi–Rich systems as a single cluster because the MANOVA significance test from a $`K`$–means clustering analysis indicates that the further splitting of the DLA/Hi–Rich cluster is not justified. On Figure 2, we have labeled each of these five clusters as “Classic”, “Double”, “Single/Weak”, “Civ deficient”, and “DLA/Hi–Rich”.
It is in a diagram of the $`K`$–cluster means, Figure 3, that the properties distinguishing each cluster become apparent. Across the horizontal axis of Figure 3 are the absorption properties; from left to right they are the Mgii, Ly$`\alpha `$, Civ, and Feii absorption “strengths”, and the “strength” of the kinematic spread. Recall that the data have been $`N(0,1)`$ standardized. Thus, for each given absorption property, the mean value for all systems is zero and the vertical axis is in units of standard deviations. Each cluster is represented by a different data point type: Classic (solid circle); Double (solid square); Single/Weak (open circle); Civ deficient (solid triangle); and DLA/Hi–Rich (solid pentagon). These classes are also connected by unique line types (i.e. dash–dot, solid, etc.); it is important to not only compare a given property across clusters, but to compare how all properties are segregated by cluster. We now briefly discuss some distinguishing features of each absorber class.
#### 3.2.1 Classic Systems
This class appears to have what might be thought of as “typical”, or non–extreme properties. As seen in Figure 3, this class is characterized by having Mgii, Ly$`\alpha `$, Feii, and Civ equivalent widths, and a kinematic spread within $`0.5\sigma `$ of the respective means for the overall sample.
We call these systems “Classic” because they can be thought to represent the most common type of Mgii absorber observable in earlier generation, low resolution surveys (e.g. Steidel & Sargent (1992); Lanzetta, Turnshek, & Wolfe (1987)). To the sensitivities of these surveys, it was found that virtually all Mgii absorption–selected systems also had Civ absorption.
#### 3.2.2 Civ “Deficient” Systems
The Civ–deficient systems have Mgii, Ly$`\alpha `$, and Feii properties identical to the Classics and constitute a similar fraction of the overall Mgii absorber population. Their distinguishing properties are significantly lower Civ absorption strengths.
#### 3.2.3 Double Systems
The Doubles are set apart from the Classics by having at least twice the Mgii and Civ absorption strengths and Mgii kinematics. Since the mean Mgii strength for Doubles is in the same regime as of the DLA/Hi–Rich systems, it is the Civ strengths and extreme Mgii kinematics that set this class apart from the others. Doubles also have Ly$`\alpha `$ and Feii absorption strengths that are systematically greater than those of the Classics.
#### 3.2.4 Single/Weak Systems
The Single/Weak class is defined foremost by very small Mgii strengths and kinematics. In fact, with the exception of two systems, all are single, narrow clouds (often unresolved in the HIRES spectra).
The Single/Weak systems are underrepresented in Sample CA, mostly because seven of the systems in our database have no measured Ly$`\alpha `$. If these seven systems were included with their Civ limits artificially treated as detections, the mean Civ for this class would drop to $`1.1`$ (this being an upper limit) in the $`K`$–means cluster diagram (Figure 3). Thus, there is a spread in the distribution of Civ strengths associated with Single/Weak systems. The true mean Feii is also probably lower than that shown on Figure 3 because the limits on Feii in the HIRES spectra scatter about the measured mean. Thus, higher sensitivity spectra would reduce the mean Feii for this class and could reveal a spread in Feii strengths. Only for the Single/Weak class, for which many of the properties were below our detection thresholds, are the above issues pertinent.
#### 3.2.5 DLA/Hi–Rich Systems
The DLA/Hi–Rich class has very strong Mgii, and extremely strong Ly$`\alpha `$ and Feii, relative to the overall Mgii absorber population. They have Civ strengths below that of the Classics and Mgii kinematics typical of the Classics and Civ–deficient absorbers. Their Mgi and Feii strengths are five to ten times greater than those of the typical Classic absorber. Unlike the elevated Ly$`\alpha `$ strengths, which are due to broad damping wings, the Mgi and Feii strengths are driven by the line–of–sight kinematic spreads of the gas for multiple, saturated components. For the intermediate ionization species, Siii, Alii, Cii and Siiii, the strengths are somewhat greater than those of the Classics, on average. The higher ionization species, Siiv, Nv and Ovi (to the extent the latter two have been measured), have strengths consistent with those of the Classics absorbers.
### 3.3. Robustness of the Clustering Analysis
Two concerns regarding the clustering analysis are: (1) possible biasing due to an slight, but apparent, overabundance of absorbers with $`W_r(\text{Mg}\text{ii})>1.3`$ Å, and (2) the exclusion of the “weak” systems for which only upper limits on the Civ equivalent widths were measured.
The systems with $`W_r(\text{Mg}\text{ii})>1.3`$ Å are all members of the DLA/Hi–Rich class. Of these, four are bonified DLAs with $`N(\text{H}\text{i})2\times 10^{20}`$ cm<sup>-2</sup>, whereas two have column densities below this classical threshold (i.e. they are Hi–Rich systems). In an unbiased survey for $`z<1.65`$ DLAs, using Mgii absorption as a selection method, Rao & Turnshek (1999) found that 14% of the Mgii systems with $`W_r(\text{Mg}\text{ii})>0.3`$ Å are DLAs. In our sample, 22 systems have $`W_r(\text{Mg}\text{ii})>0.3`$ Å; thus, 18% of our sample is comprised of DLAs. This is not inconsistent with the unbiased results of Rao & Turnshek.
The presence of a DLA/Hi–Rich class in our sample is not due to the slight bias (overabundance) of the larger $`W(\text{Mg}\text{ii})`$; it is mostly defined by large Hi and Feii equivalent widths. In fact, one DLA has $`W_r(\text{Mg}\text{ii})=0.9`$ Å ($`z=0.5764`$ toward Q $`0117+213`$), and was classified as DLA/Hi–Rich in spite of the fact that this Mgii equivalent width is well below $`1.3`$ Å. In other words, even if the Mgii equivalent widths of the DLA/Hi–Rich systems were quite small, the class would be unchanged in the cluster analysis due to the very strong Hi and Feii strengths. This is clearly shown in Figure 3.
To investigate whether membership in a given class (especially the Single/Weak class) may be sensitive to the inclusion of systems for which only an upper limit is available for $`W_r(\text{C}\text{iv})`$, we ran the cluster analysis including the four additional systems with information on Ly$`\alpha `$ (without regard to whether Civ was measured or an upper limit)A. In this analysis, the Civ upper limits were treated as measured values. The class memberships were unchanged for the original 30 systems. The added systems were classified as Single/Weak based upon their small $`W_r(\text{Mg}\text{ii})`$ and narrow kinematics. This was expected, given the already large spread in $`W_r(\text{C}\text{iv})`$ for the Single/Weak class.
We thus conclude that the clustering results are robust; they are not affected by biases nor are they sensitive to changes in the sample.
### 3.4. Absorption Strengths
In Figure 4, we present the rest–frame equivalent widths (taken from Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555. and from Tables 3 and 4 of Paper I) vs. $`W_r(\text{Mg}\text{ii})`$. We have also included $`W_r(\text{Ca}\text{ii})`$ vs. $`W_r(\text{Mg}\text{ii})`$. The panels are ordered by increasing ionization potential from the upper left to the lower right and the data point types are the same as in Figure 3.
For the non–DLA/Hi–Rich systems, note that the absorption strengths of the low ionization species Mgi, Feii, Siii, Alii, and Cii increase with increasing $`W_r(\text{Mg}\text{ii})`$, indicating these species arise in the phase giving rise to the Mgii absorption. Overall, the higher ionization species Siiv and Civ exhibit more scatter, as quantified by the dispersion per observed range.
We caution that the classification scheme introduced above should not be taken to suggest that Mgii absorbers group into discretized classes. Discretization is a byproduct of clustering analysis. In fact, the distribution functions of the equivalent widths plotted in Figure 4 are characterized by single modes and decreasing tails<sup>2</sup><sup>2</sup>2The exceptions are $`W_r(\text{Mg}\text{i})`$, $`W_r(\text{Fe}\text{ii})`$, and $`W_r(\text{Ly}\alpha )`$, which are bimodal due to the DLA/Hi–Rich class. However, it is not clear if this bimodality is due to small numbers and/or is due to the relatively large number of DLAs in our sample. As such, any single absorption property, viewed in this univariate fashion, is distributed continuously. However, from the perspective of a multivariate analysis, it is clear that the overall properties of Mgii absorbers group in well defined regions of a “multi–dimensional space”.
## 4. Inferring Ionization Conditions and Kinematics
To examine the kinematic and ionization conditions, we define a new subsample, IC (IC = Ionization Conditions). Sample IC includes only those systems with (1) an Mgii equivalent width detection threshold less than or equal to $`0.02`$ Å, and (2) no unresolved saturation in at least four adjacent pixels (1.33 resolution elements) in the Mgii $`\lambda 2796`$ profile. These selection criteria enforce a high level of accuracy in the number of Voigt profile components, $`N_{cl}`$, and their column densities, velocities, and $`b`$ parameters. Simulations of blended, multiple component Mgii profiles with these characteristics show that the distribution of Voigt profile parameters output from Voigt profile fitting is consistent (99% confidence) with those used to generate the profiles (Churchill (1997)). Data with lower signal–to–noise ratios result in a slight paucity of Voigt profile components, and therefore (to compensate) the resulting components have column densities and $`b`$ parameters that are too large. Profiles with severely saturated cores (e.g. DLA/Hi–Rich absorbers) systematically have fewer Voigt profile components by a factor of three and have unconstrained column densities and $`b`$ parameters. The equivalent width threshold criterion directly translates to a signal–to–noise ratio criterion of $`S/N=22`$ per resolution element in the continuum. Sample IC membership is given in column nine of Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555..
### 4.1. Ionization Conditions
#### 4.1.1 Single/Weak Systems
Normally, it is difficult to interpret the ionization conditions based upon equivalent widths because simple curve of growth arguments are muddled by the possibility of unresolved saturation in multiple component absorption. However, these arguments can hold for the Single/Weak clouds.
As shown by Churchill et al. (1999a), under the assumption of photoionization, Single/Weak clouds with $`W_r(\text{Mg}\text{ii})0.15`$ Å, cannot give rise to $`W_r(\text{C}\text{iv})0.2`$ Å (see their Figure 12). The upper limit on $`W_r(\text{C}\text{iv})`$ is significantly more restrictive when Feii is detected in the Mgii cloud (i.e. the cloud is constrained to have low ionization conditions). Based upon this analysis, we find that roughly half of the observed Single/Weak absorbers likely have multiple ionization phases, with Civ and some portion of the Ly$`\alpha `$ absorption arising in spatially distinct, higher ionization material. These findings are confirmed in a thorough study using CLOUDY (Ferland (1996)) photoionization models tuned to the measured Mgii column densities and constrained by the full complement of ionization species and transitions covered in FOS spectra (Rigby et al. (1999))
#### 4.1.2 Multicloud Systems
We now use photoionization modeling to investigate whether multiphase ionization can also be inferred for the multicloud Classic, Civ–deficient, and Double systems. We consider whether the measured Civ and/or Siiv absorption can or cannot all arise in the Mgii clouds, even under extreme assumptions about their ionization conditions. The following is based upon the methods employed by Churchill & Charlton (1999) in their study of the $`z=0.9276`$ systems toward PG $`1206+459`$.
We assumed that the Mgii clouds in a given system are in photoionization equilibrium and used CLOUDY (Ferland (1996)) to model their ionization conditions. A Haardt & Madau (1996) extragalactic spectrum at $`z=1`$ was used for the ionizing radiation. Based upon experiments with various single and extended starburst galaxy spectral energy distributions (Bruzual & Charlot (1993)) with assumed high photon escape fractions, we determined that CLOUDY models involving Mgii, Feii, Siiv, and Civ are not highly sensitive to the chosen spectrum for $`z1`$ (Churchill & Charlton (1999)); thus our general conclusions are not sensitive to the assumed spectral energy distribution.
We obtained a predicted Civ and Siiv equivalent width for each system by tuning a CLOUDY model to the observed Mgii and Feii column densities in each of its clouds. We then measured the absorption strengths in synthesized HIRES spectra using the velocities and $`b`$ parameters thermally scaled from those measured for the Mgii clouds \[see Churchill & Charlton (1999) for details.\]
Since our goal is to infer if multiphase ionization structure is possibly present in a given system, we have forced the models to yield the maximum amount of Civ and Siiv that can arise in the Mgii clouds. If a cloud has detected Feii, then, for a given $`[\alpha /\mathrm{Fe}]`$ abundance pattern under the assumption of negligible dust depletion, the Feii and Mgii column densities uniquely determine the ionization parameter (the logarithm of the ratio of the number density of hydrogen ionizing photons to the number density of hydrogen), nearly independent of metallicity. The ionization parameter uniquely dictates the Civ and Siiv column densities in the clouds. Typical ionization parameters in the clouds with Feii are $`4`$ to $`3`$. We use a solar $`[\alpha /\mathrm{Fe}]`$ abundance pattern, since this yields a higher ionization parameter for a given ratio of Mgii to Feii column densities, and thus yields a maximized Civ and Siiv strength. If there is no detected Feii for a cloud (either a limit or no coverage), then the ionization parameter is pushed to as high a value as possible without the cloud size exceeding 10 kpc. Typically, this occurs at an ionization parameter of $`1.9`$. A solar metallicity was assumed for all clouds, again in order to maximize the predicted Civ and Siiv strengths (for a high metallicity cloud, the hydrogen column density required to give rise to the observed Mgii is smaller, and the ionization parameter can be pushed to a higher value). We did not apply any other constraints from the data (e.g. Siii, Cii, etc.) than those described above.
In Figure 7, we present the predicted maximum equivalent widths vs. the observed equivalent widths for Civ (left panel) and Siiv (right panel). The mean error in the measured values is depicted by the open–box data point in the upper left of each panel. Diagonal lines demarcate those systems that are inferred to have multiphase ionization (lower right) from those that could be single phase. The maximum predicted Civ strength that can arise in photoionized Mgii clouds is more than $`3\sigma `$ short of the observed $`W_r(\text{C}\text{iv})`$ for a number of systems. Even two of the Civ–deficient systems are below the line. This implies that even though these particular systems have below average Civ absorption strengths, the Civ may still arise in a distinct, higher ionization phase and not in the Mgii clouds. For Siiv, it appears that fewer systems would require multiphase structure in this ionization species; Siiv likely resides in the Mgii clouds in a larger fraction of Mgii selected absorbers than does the Civ.
### 4.2. Mgii and Civ Kinematics
#### 4.2.1 Number of Clouds
In Figure 6, we show these same absorption strengths vs. the number of clouds, or subcomponents, $`N_{cl}`$, obtained from Voigt profile decomposition. The data point types are the same as for Figure 4.
There is a strong correlation of $`W_r(\text{Mg}\text{ii})`$ with $`N_{cl}`$, the number of Voigt profile components (also see Petitjean & Bergeron (1990)). A linear least–squares fit, which we have drawn as a dotted line in Figure 6, yielded a slope of $`0.069\pm 0.005`$ Å cloud<sup>-1</sup>. Note that this relationship is likely resolution dependent. Also note that the fit is strongly influenced by the two Double systems with large $`N_{cl}`$.
We find that there is not as “linear” a dependence for $`W_r(\text{Fe}\text{ii})`$ nor for $`W_r(\text{C}\text{iv})`$ with Mgii kinematics. There is, however, a tight correlation of $`W_r(\text{Ly}\alpha )`$ with the number of Mgii clouds. We obtained a linear least–squares fit (drawn as a dotted line) with slope $`0.57\pm 0.08`$ Å cloud<sup>-1</sup>. The linear relationship for $`W_r(\text{Mg}\text{ii})`$ vs. $`N_{cl}`$ suggests that majority of neutral hydrogen equivalent width arises in the Mgii clouds themselves.
#### 4.2.2 Mgii Kinematics
In Figure 5, the absorption strengths of Mgii, Feii, Ly$`\alpha `$, and Civ are plotted vs. the Mgii kinematic spread, $`\omega _v`$. The data point types are the same as for Figure 4.
In Figure 5, the scatter in $`W_r(\text{Mg}\text{ii})`$ arises because the Mgii equivalent width is dominated by the largest clouds, which are usually clustered in subgroups with velocity spreads less than $`20`$ km s<sup>-1</sup> \[see Figure 12 of Charlton & Churchill (1998)\]. The kinematic spread, on the other hand, is sensitive to the presence of small $`W_r(\text{Mg}\text{ii})`$ clouds with large velocities ($`\omega _vv^2`$). The “limit” along the upper left of the data, shown as a dotted line, is due to saturation in the profile line cores; there is a maximum equivalent width for a given velocity spread when the profile is black bottomed. The “limit” along the lower right is due to detection sensitivity; for the signal–to–noise ratios of the sample, there is a minimum detectable $`W_r(\text{Mg}\text{ii})`$ for a given velocity spread. Higher quality data would be required to determine if there is an actual paucity of systems with large kinematic spreads and extremely weak Mgii absorption.
The strong correlation between $`W_r(\text{C}\text{iv})`$ and $`\omega _v`$ has previously been discussed by Churchill et al. (1999b). The correlation is driven by the three Double systems ($`z=0.8514`$ toward Q $`0002+051`$, $`z=0.9110`$ toward PKS $`0823223`$, and $`z=0.9276`$ toward PG $`1206+459`$) and the two kinematically extreme Classic systems ($`z=1.3250`$ toward PG $`0117+213`$ and $`z=0.7908`$ toward PKS $`2145+067`$), which all have $`\omega _v75`$ km s<sup>-1</sup>. We have drawn in the maximum–likelihood linear fit to the data, which has slope $`15.4\pm 0.2`$ mÅ $`(\text{km s}\text{-1})^1`$. More data on Double systems at intermediate redshifts, selected by $`W_r(\text{Mg}\text{ii})>0.6`$ Å, would be useful for determining how tight the relationship between $`W_r(\text{C}\text{iv})`$ and Mgii kinematics remains for the largest kinematic spreads.
Three of the five Classic systems above the correlation line are from the lowest signal–to–noise spectra; there could be missed, small equivalent width components resulting in a small $`\omega _v`$. Even so, there is a significant scatter in $`W_r(\text{C}\text{iv})`$ at a given $`\omega _v`$ for $`\omega _v60`$ km s<sup>-1</sup>. The Single/Weak systems have $`W_r(\text{C}\text{iv})0.5`$ Å and $`\omega _v5`$ km s<sup>-1</sup>, whereas the Civ–deficient systems have $`W_r(\text{C}\text{iv})0.2`$ Å and $`\omega _v45`$ km s<sup>-1</sup>.
#### 4.2.3 Civ Kinematics
In Figure 8 we display the suite of Civ doublets feasured in FOS spectra (we exclude ground based data; see Paper I) that are not blended with other transitions. The velocity window for the Civ doublets is 2000 km s<sup>-1</sup>. Above each doublet member are ticks giving the velocities of the Mgii Voigt profile components. The Mgii $`\lambda 2796`$ profiles are also shown with the Voigt profile models superimposed on the data, which are ordered by increasing kinematic spread, $`\omega _v`$. The velocity window for the Mgii data is 600 km s<sup>-1</sup>.
To demonstrate which profiles are resolved, and to quantify the degree to which they are resolved, we fit the individual Civ members with single Gaussians while holding the width constant at the value of the instrumental spread function, $`\sigma =98`$ km s<sup>-1</sup>. We have superimposed these unresolved fits on the Civ $`\lambda \lambda 1548,1550`$ profiles. We note that several of the spectra were obtained prior to the refurbishing mission in which COSTAR was installed. For “large” aperture aquisition modes ($`1`$″ slit), the instrumental spread function of the pre–COSTAR instrument has extended wings. PKS $`0454220`$ is the only object acquired in this mode; thus, the unresolved fit to the $`z=0.4744`$ system should be viewed with some discretion. Several of the unresolved fits are consistent with the Civ data, suggesting that the adopted instrumental spread function accurately represents the pre–COSTAR and post–COSTAR instrument for the smaller apertures modes.
We computed the quantity,
$$F_r=\frac{W(\mathrm{measured})W(\mathrm{unresolved})}{W(\mathrm{measured})},$$
(2)
where $`W(\mathrm{measured})`$ is the measured Civ equivalent width adopted for Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555., and $`W(\mathrm{unresolved})`$ is the Civ equivalent width from the unresolved fits. Thus, $`F_r`$ measures the fraction of the integrated flux due to resolved velocity structure in the Civ doublet. We have noted the $`F_r`$ value and its $`1\sigma `$ uncertainty for each Civ profile in Figure 8. There is a $`3\sigma `$ correlation between $`F_r`$ and $`\omega _v`$. We have thus demonstrated that the tight $`W_r(\text{C}\text{iv})`$ vs. $`\omega _v`$ correlation is due to resolved velocity structure in the Civ absorbing gas.
Velocity structure is not unexpected. Observed at higher resolution, Civ profiles from Milky Way halo gas exhibit velocity structure over a $`200`$ km s<sup>-1</sup> spread, though the components are typically blended together (Sembach et al. 1999b ). For higher redshift galaxies, the Civ profiles also show distinct velocity structure (Petitjean & Bergeron (1994)) with velocity spreads up to 200 km s<sup>-1</sup>. In Figure 9, we present an example of a kinematically complex Civ doublet ($`9`$ km s<sup>-1</sup> resolution) and two Feii transitions (Mgii was not covered) for the damped Ly$`\alpha `$ absorber at $`z=1.7763`$ toward Q$`1331+170`$<sup>3</sup><sup>3</sup>3Interestingly, this DLA has $`W_r(\text{C}\text{iv})=1.63`$ Å, twice that of the DLA/Hi–Rich systems in our lower redshift sample. It has $`W_r(\text{Mg}\text{ii})=1.3`$ Å (Steidel & Sargent 1992). The ticks above the Civ profiles were obtained by a crude Voigt profile fit to the strongest components to parameterize the kinematics; we used this fit to synthesize the Civ doublet as it would be observed with FOS resolution and pixel sampling (infinite signal–to–noise ratio). This is shown in the separate, bottom panel of Figure 9, with the Civ $`\lambda 1548`$ and $`\lambda 1551`$ component ticks above the continuum (not the Mgii ticks as in Figure 8). An unresolved fit is superimposed on the synthetic FOS doublet for which we measured $`F_r=0.27\pm 0.02`$. This value is comparable to $`F_r`$ for the $`z=0.5505`$ system toward Q $`1241+174`$ and the $`z=0.9110`$ double system toward PG $`0823223`$.
In cases where the Mgii profiles are highly asymmetric, it appears that the Civ profiles exhibit similar asymmetry in the same sense. That is, the flux discrepancies between the unresolved fits and the Civ profiles are aligned in velocity with the kinematic outliers seen in Mgii. Since we have demonstrated that, in most cases, the Civ does not arise in the same phase as the Mgii clouds (see § 4.1.2), we infer that the Civ often arises in a physically distinct structure from the Mgii, but is aligned kinematically with the Mgii.
## 5. Absorption–Galaxy Properties
We tested for correlations between the galaxy properties and all the absorption line data presented in Paper I, including all possible combinations of equivalent width ratios. We performed Kendall and Spearman non–parametric rank correlation tests using the program ASURV (LaValley, Isobe, & Feigelson (1992)). The ASURV algorithm provides for measurements that are either upper or lower limits.
Out of 114 tests performed, seven resulted in correlations above the $`2.5\sigma `$ level<sup>4</sup><sup>4</sup>4For 114 correlations tests, only one test should result in a $`2.5\sigma `$ or greater significance level at random.; these are: $`W_r(\text{Mg}\text{ii})`$, Mgii doublet ratio, and $`W_r(\text{Si}\text{iv})/W_r(\text{C}\text{iv})`$ vs. impact parameter; $`W_r(\text{C}\text{ii})`$ and $`W_r(\text{Mg}\text{ii})`$ vs. $`M_K`$; and $`W_r(\text{Ly}\alpha )/W_r(\text{C}\text{iv})`$ and $`W_r(\text{Si}\text{ii})/W_r(\text{Si}\text{iii})`$ vs. $`BK`$. When DLA/Hi–Rich systems were removed from the sample, the significance levels dropped slightly below $`2.0\sigma `$. No test resulted in a correlation at a $`3.0\sigma `$ or greater significance level.
### 5.1. Global Ionization and Density Structure?
In Figure 10 we present plots of the absorption properties used in our multivariate analysis, $`\omega _v`$, $`W_r(\text{Mg}\text{ii})`$, $`W_r(\text{Fe}\text{ii})`$, $`W_r(\text{Ly}\alpha )`$, and $`W_r(\text{C}\text{iv})`$, vs. host galaxy properties.
$`W_r(\text{Mg}\text{ii})`$ is anti–correlated with impact parameter (decreases with galactocentric distance) with a $`2.7\sigma `$ significance and is correlated with $`M_K`$ (increases with decreasing luminosity), also with a $`2.7\sigma `$ significance. Both these trends arise because DLAs, which give rise to the largest $`W_r(\text{Mg}\text{ii})`$, are often observed at low impact parameters and often have low luminosities ($`M_K>24`$).
The $`W_r(\text{C}\text{iv})`$ vs. $`K`$ luminosity relationship can provide a test for halo models in which a virialized hot phase ($`T10^6`$ K) pressure confines the low ionization phase (e.g. Mo & Miralda–Escudé (1996)). These models predict an anticorrelation between $`W_r(\text{C}\text{iv})`$ and galaxy mass (i.e. $`K`$ luminosity); Civ is predicted to be enhanced in smaller, lower mass galaxies, which have low gas pressures in their halos. As seen in Figure 10, there is a visual, yet not statistically significant, trend for the most massive galaxies to have smaller $`W_r(\text{C}\text{iv})`$ in our small sample. This is driven by the high luminosity \[and very red; see $`W_r(\text{C}\text{iv})`$ vs. $`BK`$\] Civ–deficient systems.
In Figure 11, we have plotted the equivalent width ratios of Siiv to Civ, Ly$`\alpha `$ to Civ, Ly$`\alpha `$ to Mgii, Feii to Civ, and the Civ and Mgii doublet ratios vs. impact parameter. We have selected combinations only of the gas properties used in our multivariate analysis, with the exception of the ratio $`W_r(\text{Si}\text{iv})/W_r(\text{C}\text{iv})`$. This ratio is anticorrelated with impact parameter at the $`2.8\sigma `$ level when all galaxies are included and at the $`2.1\sigma `$ level when DLA/Hi–Rich systems are excluded. If this trend were to hold for a larger sample, it would be consistent with the scenario in which halo gas is photoionized by the extragalactic UV background flux resulting in a increasing ionization level with increasing galactocentric distance (see Savage, Sembach, & Lu (1997)).
The Mgii doublet ratio is correlated with impact parameter at the $`2.7\sigma `$ significance level. This reflects the fact that the gas tends to become optically thin at larger galactocentric distances, where $`W_r(\text{Mg}\text{ii})`$ is systematically smaller. Again, the DLAs dominate the trend.
The ratio $`W_r(\text{Ly}\alpha )/W_r(\text{Mg}\text{ii})`$ is not correlated with impact parameter when all galaxies in the sample are included. However, when the DLA/Hi–Rich systems are excluded a $`2.5\sigma `$ level correlation is observed. This is governed by the general decrease in Mgii equivalent widths with impact parameter, since Ly$`\alpha `$ shows no trend with impact parameter (when DLA/Hi–Rich systems are excluded). This particular trend is difficult to interpret because of the very different curve of growth behavior for Mgii and Ly$`\alpha `$. It is not possible to address the possibility of a metallicity or ionization gradient. The location of the Lyman limit break was covered for only six of the galaxies and was detected for all six, sampling impact parameters from $`10`$ to $`40`$ kpc.
### 5.2. Ionization Conditions and Colors
In Figure 12, we have plotted the Civ to Mgii, Ly$`\alpha `$ to Civ, and Ly$`\alpha `$ to Mgii equivalent width ratios vs. host galaxy $`BK`$ color. Both $`W_r(\text{Ly}\alpha )/W_r(\text{C}\text{iv})`$ and $`W_r(\text{Ly}\alpha )/W_r(\text{Mg}\text{ii})`$ are correlated with $`BK`$ at a $`2.6\sigma `$ significance level. This trend in our small sample is governed by both the DLA/Hi–Rich systems and the Civ–deficient systems being associated with redder galaxies.
### 5.3. Correlation Test: Overall Results
The degree of scatter in the absorption properties presented here are similar to that found by Churchill et al. (1996) for the Mgii absorption properties only. None of the presented absorption properties correlated with the galaxy properties at a high significance level. This is not to say that definite trends, or even statistically significant correlations, do not exist. The large scatter in the properties signifies large local variations from line of sight to line of sight through the galaxies; however, even with our small sample, we have uncovered some suggestive global trends.
We also find that the inclusion or exclusion of DLA/Hi–Rich systems in any given test significantly alters the statistics. The host galaxies of DLAs are seen to have a wide range of luminosities, morphologies, colors, and surface brightnesses (e.g. Le Brun et al. (1997); Rao & Turnshek (1998)), and do not represent a homogenous population, despite the fairly homogeneous absorption properties by which they are selected.
## 6. Discussion
We reemphasize that a taxonomic scheme based upon a multivariate clustering analysis will, by its very nature, result in a discretization of what may actually be a continuum of properties. Nonetheless, if we adopt the clustering analysis results at face value (see § 3.2) and segregate the wide variety of Mgii absorber properties into individual “classes”, we may be able to find order in an otherwise complex array of gas properties and kinematics. The hope is to gain further insights into defining distinctive or “characteristic” properties of Mgii absorbers and to understanding the galactic processes that give rise to the observed range of these “characteristics”.
What follows is primarily a speculative discussion, given the fact that there is still very little data upon which inferrences can be based.
### 6.1. Classic vs. Civ–deficient Systems
The existence of Civ–deficient systems with $`W_r(\text{C}\text{iv})0.15`$ Å, which have kinematics similar to the Classics, indicates that the connection between Civ absorption and Mgii kinematics does not operate the same in all galaxies (or at all locations in galaxies).
There are only three Civ–deficient systems with measured galaxy properties ($`z=0.4297`$ toward PKS $`2128123`$, $`z=0.6601`$ toward Q $`1317+277`$, and $`z=0.7291`$ toward PG $`0117+213`$). The host galaxies of these three systems are reddish ($`BK=3.3`$, $`3.8`$, and $`4.0`$) and are among the more luminous ($`M_K=24.7`$, $`25.7`$, and $`26.3`$). Also, they are probed at large impact parameters ($`D=32`$, $`38`$, and $`36`$ $`h^1`$ kpc), respectively (see Figure 10). The redder colors and large luminosities would suggest massive, early–type galaxies. Based upon HST images of the $`z=0.6601`$ and $`z=0.7291`$ absorbers (Steidel (1998)), we roughly (visually) classify them to be an edge–on S0 galaxy (probed on its minor axis) and a face–on SBa galaxy, respectively.
Despite small numbers, these relatively high masses, early–type morphologies, and large impact parameters may be a clue to the observed spread in $`W_r(\text{C}\text{iv})`$ for a given Mgii absorption profile type. Semianalytical models of pressure confined gaseous halos predict smaller Civ strengths in more massive galaxies (e.g. Mo & Miralda–Escudé (1996)), and such a trend is not inconsistent with our data. The weak Civ cannot simply be due to the line of sight through the galaxy, because some Classics at large impact parameters have very strong Civ strengths. Also, based upon cross–sectional arguments from the number of Civ absorbers per unit redshift (Steidel (1993)) and upon the expected lower pressures and gas densities at larger galactocentric distances (Mo & Miralda–Escudé (1996)), it is expected that Civ absorption would be strong at large impact parameters.
Perhaps strong Civ is expected in regions of relatively pronounced star formation, which would be consistent with Civ–deficient absorbers being associated with earlier–type galaxies. Alternatively, in Civ–deficient absorbers, the Civ could be ionized away due to collisional processes, such as with the intragroup medium scenario proposed by Mulchaey et al. (1996). If so, strong Ovi absorption would be predicted; unfortunately, we cannot address Ovi absorption for the Civ–deficient systems \[only one is “measured” to have an unrestrictive upper limit on $`W_r(\text{O}\text{vi})`$\]. A further possibility is that galaxies with extended X–ray emission, usually early–type galaxies (e.g. Mathews & Brighenti (1998)), may preferentially be Civ– and Ovi–deficient because of the extremely hot environment. If so, the Mgii absorbing gas would necessarily arise in Hi gas, which is of external origin for ellipticals from galaxy mergings (Knapp, Turner, & Cunniffe (1985); van Gorkom (1992)) and is sometimes found to be in disk–like structures extending 5 to 10 times the optical radii in “dust–lane” elliptical galaxies (e.g. Morganti, Oosterloo, & Tsvetanov (1998)) Thus, one might hypothesize that Mgii absorption associated with elliptical galaxies arises preferentially in post–merger ellipticals.
### 6.2. Classic vs. Double Systems
There are at least two obvious explanations for the Double systems; they might be two Classic systems clustered in line–of–sight velocity (bound or unbound), or a primary galaxy and a satellite (e.g. possible an interacting LMC–like object).
The three Double systems in our sample have the largest $`W_r(\text{C}\text{iv})`$, which are consistent with that of two typical Classic systems. Alternatively, they could arise in a primary galaxy either undergoing a minor merger or residing in a group with several minor galaxies. Using the Local Group as a model and applying the simple cross–sectional dependence for $`W_r(\text{Mg}\text{ii})`$ with galaxy luminosity (Steidel (1995); also see McLin, Giroux, & Stocke (1998)), we estimated the probability of intercepting a “double” absorber for a random line of sight passing through the Milky Way (line–of–sight kinematics were not considered). We found a $`25`$% chance of intercepting both the LMC and the Milky Way, and a $`5`$% chance of intercepting both the SMC and the Milky Way. All other galaxies in the Local Group have negligible probabilities of being intercepted for a line of sight passing within 50 kpc of the Milky Way.
An alternative scenario is that Double systems are otherwise typical Classic systems, but for the interception of kinematic outlier clouds, either due to infalling or ejected “high velocity” gas. This picture would naturally invoke elevated star formation to explain the large $`W_r(\text{C}\text{iv})`$ in Doubles and, possibly, the correlation between $`W_r(\text{C}\text{iv})`$ and Mgii kinematics. Infalling material can enhance star formation (Hummel et al. (1990); Lutz (1992); Hernquist & Mihos (1995); Hibbard & van Gorkom (1996)), which can provide the energetics to support a multiphase halo (Dahlem (1998)) with strong Civ absorption \[i.e., as with the Galaxy (Savage et al. 1997)\]. Consistent with this scenario would be that the bluest galaxies would preferentially be associated with Double systems. Only one Double system has measured galaxy properties ($`z=0.8519`$ toward Q $`0002+051`$), and we note that its color ($`BK=2.9)`$ is among the bluest in the sample, consistent with a star forming object. An HST image (Steidel (1998)) reveals the galaxy to be compact with a spherical, featureless morphology (no evidence for merging).
### 6.3. Classic vs. DLA/Hi–Rich
Though the Classic Mgii systems are well understood to select normal, bright galaxies (Steidel et al. (1994)), the DLA/ systems are seen to select \[based upon $`N(\text{H}\text{i})2\times 10^{20}`$ cm<sup>-2</sup>\] an eclectic population of low luminosity, low surface brightness, and dwarf galaxies (Le Brun et al. (1997); Rao & Turnshek (1998), 1999). This might suggest that not all classes of Mgii absorber have a systematic connection with galaxy type or galaxy properties.
What is interesting, however, is that the Mgii kinematic spreads for DLA/Hi–Rich systems are tightly clustered around $`60`$ km s<sup>-1</sup>, have no weak, kinematic outlier clouds, and have intermediate to weak Civ absorption strengths. This is in contrast to those at redshifts $`2`$$`3`$, where strongly absorbing, higher velocity outlying clouds and often “double” profiles are characteristic of low ionization DLA profiles (Lu et al. (1996)).
Whatever physical process or environment gives rise to kinematic outlying Mgii clouds and strong Civ, it is apparently not acting in intermediate redshift DLA/Hi–Rich systems. Alternatively, these trends could be due to selection effects of sightlines through Hi–rich environments, which implies a line–of–sight dependency for absorber class.
### 6.4. Absorber Class Evolution?
We have shown that there is a signicant correlation between $`W_r(\text{C}\text{iv})`$ and the Mgii kinematic spread. But, exactly how do the Mgii kinematics vary with $`W_r(\text{C}\text{iv})`$ for a given $`W_r(\text{Mg}\text{ii})`$? In Figure 13, we have plotted the Mgii $`\lambda 2796`$ profiles from our HIRES/Keck spectra in the approximate locations they occupy in the $`W_r(\text{C}\text{iv})`$$`W_r(\text{Mg}\text{ii})`$ plane. Based upon this schematic, we have an impression of how the kinematic “complexity” of the absorption profile relates to both the Mgii and Civ absorption strengths. Roughly, each of the five classes obtained from our multivariate analysis falls in a well defined region, which we have marked to guide the reader. Note that, in general, the Civ strength is largest when the kinematic “complexity” is the greatest; Civ absorption appears to have no direct connection with the optical depth of the strongest Mgii component, nor with the total number of Voigt profile components. Most remarkably, it would appear that the gross characteristics of the Mgii kinematics can be predicted simply based upon the absorber’s location on the $`W_r(\text{C}\text{iv})`$$`W_r(\text{Mg}\text{ii})`$ plane.
Since the distribution of Mgii equivalent widths evolves with redshift (Steidel & Sargent (1992)), and this must be reflected in the Mgii kinematics as well, it seems reasonable to assume that evolution should be discernable on the $`W_r(\text{C}\text{iv})`$$`W_r(\text{Mg}\text{ii})`$ plane.
There are at least two possible types of evolution of absorber classes: (1) the number density per unit redshift of a given class could evolve, either diminishing with or increasing with time, and/or (2) any absorber in a given class could evolve into another class. The former is systematic and would indicate a global, cosmic evolution, which might imply the existence of absorber classes not seen in our intermediate redshift sample. The latter form of evolution would be stochastically related to the processes occuring in galaxies and their environs.
We present the observational $`W_r(\text{C}\text{iv})`$$`W_r(\text{Mg}\text{ii})`$ plane in Figure 14. The intermediate redshift data (taken from this study) have small, filled circle data points. We have included lower and higher redshifts data from the literature, which we have listed in Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555.. The higher redshift absorbers ($`1.2z2.1`$) are taken from Bergeron & Boissé (1984), Boissé & Bergeron (1985), Lanzetta et al. (1987), and Steidel & Sargent (1992) \[six–pointed stars\]. The lower redshift points ($`0.1z0.6`$) are taken from Bergeron et al. (1994) \[downward pointing, open triangles\], and the $`z0`$ data<sup>5</sup><sup>5</sup>5We note that the $`z0`$ sample is small and is selected based upon a bright, background quasar having a small projected separation from a nearby galaxy, as opposed to being Mgii–absorption selected (Bowen et al. 1995); this may have introduced a bias, possibly related to impact parameter distribution, as compared to the higher redshift data. are taken from Bowen et al. (1995, 1996), Jannuzi et al. (1998), and Bowen (1999, private communication) \[open diamonds\]. There are four extraordinary systems listed in Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555. that have $`W_r(\text{Mg}\text{ii})`$ ranging from $`4`$ to $`7`$ Å. These have not been presented in Figure 14.
Three points from Figure 14 are that (1) there are absorbers in the higher, lower, and $`z0`$ redshift samples with Civ and Mgii strengths typical of each of the five classes found in our clustering analysis of the intermediate redshift systems; (2) there are both higher and $`z0`$ redshift absorbers occupying regions of the $`W_r(\text{C}\text{iv})`$$`W_r(\text{Mg}\text{ii})`$ that are not represented in our intermediate redshift sample; and (3) as with the intermediate redshift systems, the higher redshift systems exhibit a large range of $`W_r(\text{C}\text{iv})`$ for a given $`W_r(\text{Mg}\text{ii})`$. Though there is no suggestion for such a spread in the lower redshift systems, there are too few measurements to characterize the spread in $`W_r(\text{C}\text{iv})`$ values.
Given the fairly systematic dependence of kinematics in the $`W_r(\text{C}\text{iv})`$$`W_r(\text{Mg}\text{ii})`$ plane, we can infer that the Mgii kinematics of the lower and higher redshift systems that occupy locations consistent with a given absorber class are likely to be similar to that class’ kinematics.
The higher redshift systems with $`W_r(\text{Mg}\text{ii})>2`$ Å, are a class of absorber not present in our sample. Note the large spread in $`W_r(\text{C}\text{iv})`$ for these systems, which would suggest that some are “Civ deficient”. This may be indicative that the physical processes giving rise to the range of $`W_r(\text{C}\text{iv})`$ in intermediate redshift systems are also occuring in these $`W_r(\text{Mg}\text{ii})>2`$ Å, higher redshift systems.
In Figure 15, we present the HIRES/Keck Mgii $`\lambda 2796`$ profiles of four $`z1`$ Mgii systems with $`W_r(\text{Mg}\text{ii})>2.0`$ Å. Note that the profiles are optically thick with virtually unity doublet ratio over a the full velocity span, which is typically $`250\text{km s}\text{-1}`$ or greater. We also show the $`z0`$ Mgii profile measured in the spectrum of SN 1993J in M81 using the GHRS on HST (Bowen et al. (1995); spectrum provided courtesy of Dr. D. Bowen). In comparison to the smaller equivalent width, intermediate redshift absorbers, the Mgii kinematics and absorption strengths of these $`W_r(\text{Mg}\text{ii})>2`$ Å systems are clearly unique, exhibiting “double” black–bottomed profiles.
Of the systems shown in Figure 15, only the $`z=1.7945`$ system toward B2 $`1225+317`$ is represented on Figure 14. Though the Civ strengths of the other systems shown in Figure 15 are unmeasured or unpublished, it is clear that these systems constitute their own unique “class” (with respect to the five classes found in our clustering analysis). As shown in Figure 15, the Mgii kinematics are suggestive of “double–DLAs”, and, in fact, the $`z=1.7945`$ system toward B2 $`1225+317`$ is a DLA (Bechtold, Green, & York (1987)). In high quality data, “Double–DLAs” might be expected to have slightly asymmetric damping wings, reflecting the different column densities in the two systems.
Steidel & Sargent (1992) found that the number density of these large equivalent systems decreases with redshift over the range $`0.3z2.2`$ (the full range over which the Mgii doublet is observable from the ground). Based upon the SN 1993J spectrum of M81 and the Galaxy in absorption (the line of sight passes through half the disk and halo of M81, half the disk and halo of the Galaxy, and through “intergalactic” material apparently from the strong dwarf–galaxy interactions taking place with both galaxies), one possibility is that these systems arise from two galaxies with low impact parameters that happen to have a line–of–sight superposition.
For this scenario to be consistent with the Mgii absorber evolution, the number of galaxy pairs would need to decrease in step with the evolution of the absorbers themselves over the same redshift regime. In fact, over the redshift interval $`1z2`$, it is seen that the galaxy pair fraction, where pairs are defined to have projected separations less than $`20`$ kpc, evolves proportional to $`(1+z)^p`$, with $`2p4`$ (Neuschaefer et al. (1997), and references therein). Le Fevre et al. (1999) found a similar, better constrained result with $`p=2.7\pm 0.6`$ for $`0z1`$. These compare well with $`p=2.2\pm 0.7`$ for Mgii absorbers with $`W_r>1.0`$ Å (Steidel & Sargent (1992); the evolution is probably stronger for those with $`W_r>2.0`$ Å).
As such, galaxy pair evolution remains a plausible scenario as a contribution to the evolution of large equivalent width, higher redshift Mgii absorbers and the class we loosely have called “Double–DLAs”. We note, however, that outflows from very luminous, star bursting galaxies at these high redshifts could also account for some of these “super” systems, since the low ionization gas can be very prominent in absorption (Pettini et al. (1999)).
## 7. Summary
We have performed a multivariate analysis of the absorption strengths and kinematics of Mgii and the absorption strengths of Feii, Ly$`\alpha `$, and Civ for a sample of 45 Mgii absorption–selected systems. Descriptions of the survey and analysis of the data used herein have been presented in Churchill et al. (1997, 1999a) and in Paper I. The multivariate analysis was performed using the STATISTICA package (www.statsoft.com). We applied both a “Tree Clustering” and “$`K`$–means Clustering” analysis to the 30 systems in Sample CA from the data listed in Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555..
We have also compared the low, intermediate, and high ionization absorption properties of 16 systems with the $`B`$ and $`K`$ luminosities, $`BK`$ colors, and impact parameters of their host galaxies. The full complement of absorption line data was taken from Paper I. The galaxy properties, listed in Table Low and High Ionization Absorption Properties of Mgii Absorption–Selected Galaxies at Intermediate Redshifts. II. Taxonomy, Kinematics, and Galaxies<sup>1</sup><sup>1</sup>affiliation: Based in part on observations obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among Caltech, the University of California, and NASA. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. <sup>2</sup><sup>2</sup>affiliation: Based in part on observations obtained with the NASA/ESA Hubble Space Telescope, which is operated by the STScI for the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5–26555., are taken from the survey of Steidel et al. (1994) and the previous studies of Churchill et al. (1996) and Steidel et al. (1997). We tested for absorption–galaxy property correlations using Kendall and Spearman non–parametric rank correlation indicators as implemented with the program ASURV (LaValley, Isobe, & Feigelson (1992)), which incorporates upper limits on the data.
The main results are summarized as follows:
1. The clustering analysis revealed that there is a wide range of properties for Mgii–selected absorbers and that these can be categorized into three main types (Figure 2). Based upon strong $`W_r(\text{Ly}\alpha )`$ and $`W_r(\text{Fe}\text{ii})`$, there is the class of DLA/Hi–Rich Mgii absorbers. For the remaining systems, there is a spread in the Civ strengths for a given $`W_r(\text{Mg}\text{ii})`$ that gives rise to ‘Civ–weak” and “Civ–strong” systems.
2. The Civ–weak class separates into the Single/Weak and the Civ–deficient classes (Figures 2 and 3). Single/Weak systems are characterized by a single unresolved Mgii absorption line with $`W_r(\text{Mg}\text{ii})0.15`$ Å, a velocity spread of $`\omega _v6`$ km s<sup>-1</sup>, and a range of Civ strengths, but with $`W_r(\text{C}\text{iv})`$ no greater than $`0.5`$ Å (in our sample). The Mgii strengths and kinematics are the dominant properties in distinguishing them as a separate class. Civ–deficient systems have multiple Voigt profile components, $`N_{cl}>1`$, a range of kinematic spreads, $`15\omega _v45`$ km s<sup>-1</sup>, and $`W_r(\text{C}\text{iv})`$ no stronger than $`0.4`$ Å. The Civ strength, and to a lesser degree the Mgii kinematics, are the dominant properties distinguishing this as a Civ–weak class.
3. The Civ–strong class separates into the Classics and the Doubles (Figures 2 and 3). Classic systems, like the Civ–deficient systems, have multiple Voigt profile components, and a similar, but slightly larger, range of kinematic spread. The Civ strengths, on the other hand, are greater than $`0.4`$ Å. Separate from the Classics are the Doubles, which are characterized by roughly twice the number of Voigt profile components, $`N_{cl}`$, twice the Civ strength, and twice the Mgii kinematic spread, as compared to the Classics.
4. The Mgii, and Ly$`\alpha `$ strengths are tightly correlated with the number of Mgii Voigt profile components, $`N_{cl}`$ (see Figure 6). The tight correlation between $`W_r(\text{Ly}\alpha )`$ and $`N_{cl}`$ implies that the majority of the neutral hydrogen is arising in the phase giving rise to the Mgii absorption. In single cloud, $`N_{cl}=1`$, systems (the Single/Weak class), $`W_r(\text{Ly}\alpha )`$ ranges from $`0.2`$$`1.2`$ Å and $`W_r(\text{C}\text{iv})`$ ranges from less than $`0.15`$ to $`0.7`$ Å, suggesting a range of metallicities and/or ionization conditions in this class of object.
5. There is a highly significant correlation between $`W_r(\text{C}\text{iv})`$ and $`\omega _v`$ (see Figure 5; also see Churchill et al. 1999b ). This correlation is driven by the five absorbers (two Classics and three Doubles) with the largest $`\omega _v`$. For $`\omega _v60`$ km s<sup>-1</sup> there is a $`1`$ Å spread in $`W_r(\text{C}\text{iv})`$, due to the Civ–deficient systems and the Single/Weak systems with larger $`W_r(\text{C}\text{iv})/W_r(\text{Mg}\text{ii})`$. More systems with $`\omega _v80`$ km s<sup>-1</sup> are needed in order to determine if all “kinematically active” Mgii absorbers have such strong Civ.
6. Assuming the Mgii clouds are in photoionization equilibrium, we showed that in many systems a substantial fraction of the Civ absorption is arising in a separate phase from the Mgii (see Figure 7). We also showed that some fraction of the systems likely have Siiv in a separate phase from the Mgii. The FOS profiles of the larger Civ absorbers (Classics with large Civ and Doubles) are resolved due to the kinematic spread, or velocity structure, of the Civ phase. A quantity expressing the degree to which the profiles are resolved, $`F_r`$, correlates at the 99.99% confidence level with the Mgii kinematic spread. Furthermore, there is a clear impression that the asymmetries in the resolved Civ profiles trace the Mgii kinematics. The above facts lead us to suggest that, often, some of the Civ arises in a physically distinct phase from the Mgii gas, but is kinematically clustered with the Mgii clouds.
7. For our small sample of 16 galaxies, we find no significant trends (greater than $`3\sigma `$) of Civ and Ly$`\alpha `$, or their ratios with each other and with Mgii, with host galaxy $`B`$ and $`K`$ luminosities, $`BK`$ colors, and impact parameters. DLA/Hi–Rich systems, which have systematically smaller impact parameters and redder colors, significantly affect the outcome of correlation tests. The Civ and Ly$`\alpha `$ absorption properties do not smoothly depend upon global galaxy properties. At $`2.8\sigma `$, we find that $`W_r(\text{Si}\text{iv})/W_r(\text{C}\text{iv})`$ decreases with impact parameter, which could be interpreted as very tentative evidence for a global galactocentric ionization/density gradient (see Figure 11). The available data also show marginal trends for redder galaxies to have lower ionization conditions (Figure 12). This is suggestive that Civ phases may be systematically smaller in the redder galaxies.
8. We found that the Mgii kinematics of a given absorber is related to its location in the $`W_r(\text{C}\text{iv})`$$`W_r(\text{Mg}\text{ii})`$ plane. In a comparison of our intermediate redshift sample ($`0.4z1.4)`$ with published low redshift ($`0z0.6`$) and high redshift ($`1.2z2.1`$) samples, we found evidence for redshift evolution in the $`W_r(\text{C}\text{iv})`$$`W_r(\text{Mg}\text{ii})`$ plane. This implies that other “classes” of absorbers are present at high redshift, possibly including a “Double DLA/Hi–Rich” class. There is also a group of “super” Mgii systems at $`z2`$ that have a large range of $`W_r(\text{C}\text{iv})`$ values. We disussed a scenario in which the evolution of these strongest Mgii absorbers could be due to the evolution of the number of galaxy pairs and/or accreting LMC–like satellite galaxies.
### 7.1. Further Ruminations
In general, the observed range of Mgii kinematics and Civ absorption strengths could be due to a number of factors. These include global differences in gaseous conditions related to environment (i.e. group or isolated galaxies) or galaxy morphology, local variations dominated by passage through different parts of host galaxies (e.g. halos, outer disks, spiral arms, etc.), or small scale fluctuations in galactic interstellar and halo structures. An additional important issue is whether characteristic absorption properties might be tied to the evolutionary stage of the host galaxy. This would imply that a given host galaxy may not always be of the same “absorber class” throughout its evolution.
At higher redshifts ($`z2`$), galaxy–galaxy interactions were more common and no doubt played an important role in the kinematics and multiphase ionization conditions in many Mgii absorbers (e.g. Carilli & van Gorkom (1992); Bowen & Blades (1993); Bowen et al. (1995)). A local analogue is seen in absorption associated with M61, a nearly face–on, later–type galaxy with enhanced star formation. The Mgii velocity spread is $`300`$ km s<sup>-1</sup> at an impact parameter 21 kpc and there are two nearby galaxies (Bowen et al. (1996)).
Since it is the kinematic complexity and spread and not the total amount of the Mgii that appears to be the important property for predicting the Civ strength in $`z1`$ galaxies, it is not unreasonable to conjecture that sources of mechanical energy (i.e. active star formation) may be of central importance (see Guillemin & Bergeron (1997)). Extended multiphase halos are seen preferentially around late–type spiral galaxies with signs of elevated star formatiopn (see Dahlem (1998) and references therein). In fact, a basic scenario which might explain both the taxonomy of Mgii absorbers (especially the Double and Civ–deficient systems) and the CivMgii kinematics connection, is one in which star formation processes accelerate hot, ionized gas outward, which subsequently cools and fragments into parcels of low ionization gas (e.g. Houck & Bregman (1990); Li & Ikeuchi (1992); Avillez (1999)). These parcels would be observed as outlying kinematic components in the Mgii profiles.
These points are somewhat suggestive of a causal connection between evolution in the cosmic star formation rate and absorbing gas cross sections, kinematics, and ionization conditions. If true, we could hypothesize that scenarios of a cosmic mean history of galaxy evolution inferred from the global record of star formation history should be fully consistent with one based upon an “absorption line perspective”.
The global star formation rate appears to drop below $`z1.0`$ (Lilly et al. (1996); Madau, Lucia, & Dickinson (1998); however, see Cowie, Songaila, & Barger (1999)), implying a reduction in the (infalling) gas reservoir for galaxies. It could be that a majority of galaxies transform into self–regulatory systems by $`z1`$ and begin to evolve in a more isolated fashion and in a direction dependent upon their ability to continue forming stars. Galaxies that are rich in gas and capable of forming large numbers of molecular clouds in their interstellar media (i.e. late–type galaxies) would continue to form stars and exhibit evolution, whereas those less capable would exhibit no discernable evolution. Such a scenario is consistent with the differential luminosity evolution reported by Lilly et al. (1995) and is not inconsistent with the tentative trends we see between galaxy color and ionization conditions (also see Guillemin & Bergeron (1997)).
It is unfortunate that there currently are no Mgii absorption–line samples for the highest redshifts ($`z>2.2`$), so an absorption line perspective cannot yet be fully appreciated \[however, see the theoretical predictions of Rauch, Haehnelt, & Steinmetz (1997)\].
Support for this work was provided by the NSF (AST–9617185), and NASA (NAG 5–6399 and AR–07983.01–96A) the latter from the Space Telescope Science Institute, which is operated by AURA, Inc., under NASA contract NAS5–26555. R. R. M. was supported by an NSF REU supplement. B. T. J. acknowledges support from NOAO, which is operated by AURA, Inc., under cooperative agreement with the NSF. We thank Eric Feigelson for several very informative discussions on the statistical treatment of data and as a valuable resource of additional information. Sandhya Rao and David Turnshek kindly shared unpublished data of Mgii/DLA–selected galaxy properties, for which we are grateful. We thank David Bowen for providing Civ data prior to their publication and for an electronic version of the Mgii $`\lambda 2796`$ profile from the GHRS spectrum of SN 1993J. We also thank Bill Matthews for an enlightening discussion on the properties of gas and star formation in elliptical galaxies and the anonymous referee for a careful reading of the manuscript that lead to significant improvements.
|
warning/0005/astro-ph0005486.html
|
ar5iv
|
text
|
# Cosmic Ray Rejection and Readout Efficiency for Large-Area Arrays
## 1 Introduction
The optimal rejection of cosmic ray glitches from astronomical images is of critical importance for large-area pixelized detectors in space. The detectors (CCDs, etc) are generally stable and repeatable, so they can be carefully calibrated. They are often sensitive to cosmic rays and other radiation. The signals from cosmic rays can be the largest contamination. But the contamination is far from Gaussian; it tends to be dominated by “glitches” which have a large effect on one or a few pixels for a short duration. Finding, limiting, and rejecting affected data is a problem common to many observation and data reduction strategies.
Cosmic rays affect ground-based detectors as well, but to a much smaller extent. This is partly because the atmosphere and magnetic field act as a shield, eliminating most of the cosmic rays, and partly because atmospheric emission and scattering are variable and limit the extent to which it is possible to uncover and understand other systematic errors.
This study was focused on the Next Generation Space Telescope (Stockman et al. 1998); however, many of the results are applicable over a wide range of observatories. Although we specifically consider IR detectors, the results can be applied to visible light detectors or any detectors where both readout noise and Poisson statistics are present.
## 2 Noise Estimation
The ultimate limit on the signal-to-noise ratio is the Poisson variation in the photon arrival rate. For large numbers of photons, $`N`$, and low photon occupation numbers, the variance is approximately equal to the number of photons, so the signal-to-noise ratio is $`\sqrt{N}`$ if other sources of noise are negligible. For a fixed telescope, this is proportional to $`\sqrt{AZ\xi T}=\sqrt{ST}`$ where $`A`$ is the area of the telescope, $`Z`$ is the source strength, $`\xi `$ is the efficiency of the telescope and detector, and $`T`$ is the observing time. Other sources of noise can reduce the signal-to-noise ratio. The ratio of the Poisson variance to the real variance is a measure of the efficiency of the readout scheme. Variance is used rather than sigma, so the efficiency relates directly to time. In what follows, we will use this efficiency to compare various strategies.
In IR detectors, the photons cannot yet be reliably counted individually. The detector and readout electronics add noise (Fanson et al. 1998, Tian et al. 1996). This can be summarized as the readout noise, $`R`$, expressed as a number of photons or electrons. Typical values range from a few to $`40`$ e for well-designed systems. A nondestructive readout allows the use of multiple samples on the same pixel and electrons to reduce the readout error. However, the additional samples make a more complex system with more electronics and computational complexity. The process of sampling the detector can add electronic noise directly, or heat the detector, causing additional noise. With the current state of the art detectors, the noise can be reduced by increasing the number of samples.
There are often sources of noise with a power spectrum that scales inversely with frequency. Although this noise is poorly understood, it is often present, and in many detectors it becomes the dominant source of noise at very low frequencies. This sort of noise increases with time just like the photon noise from a source, so it can be treated like a background (or foreground) contamination.
The readout noise decreases with the number of samples, and the signal (and Poisson variance) increases with time. Thus after a sufficient time and number of samples, the readout noise will be insignificant relative to the photon noise, and the ideal efficiency of $`ϵ=1`$ will be approached.
However, there are two obstacles to long integration times. At some point, the wells of the detector fill up, and the detectors become nonlinear and ultimately insensitive to additional photons. Also, particularly in space based detectors, cosmic rays strike the detector at a rate of 5-30 cm<sup>-2</sup>sec<sup>-1</sup> (Tribble 1995, Barth & Isaacs 1999) (much higher in radiation belts or during solar flares) and add a large noise signal.
## 3 Sampling Strategy
The sampling strategy is an integral part of the larger observation strategy. The best strategy depends, of course, on the goal of the observation. Possible goals include finding sources, surveying a region of the sky, identifying sources, examining a source for variation, or mapping diffuse emission. Here we concentrate on detecting dim sources and assume that band limiting filters are used for purposes of identification and analysis.
If only a fixed number of samples, $`n`$, are available, there is an advantage of using them at the beginning and the end of the integration (Fowler sampling, Fowler & Gatley 1990, and Fowler et al. 1996). This approach gives the longest effective integration time, and the variance from the readout noise is reduced by $`4/n`$, if the sample noise is uncorrelated. Half of the samples are used at the beginning and half at the end, and the signal is obtained by differencing, which adds the variance from the beginning and the end resulting in the factor of 4. The efficiency is then $`ϵ=ST/(ST+4R^2/n)=1/[1+4R^2/(nST)]`$, which approaches 1 for long times $`T`$, large signals $`S`$ or many samples $`n`$. This does not account for cosmic ray hits or other sources of noise which also have effects.
A fixed sample rate has advantages, as the electronics can be made simple, and the effects of the readout on the chip (heating, electronic noise) can at least be made constant. Also, the detector charge history can be searched for cosmic ray strikes and other glitches. Uniformly weighting each of the readouts and fitting to a straight line effectively uses only 2/3 of the integration time (difference between the mean negative weighted times and the mean positive weighted times). Hence, the maximum efficiency is .667 for dim sources. For bright sources, as we shall see in section 5, this limitation can be overcome, and the efficiency can approach 1.
In the limit where the variance is dominated by the readout noise, it is $`V=12R^2n/(n+1)/(n+2)`$ so the efficiency, $`ϵ=ST/[ST+12R^2n/((n+1)(n+2))]`$ is only 1/3 of the Fowler readout. But increasing the integration time has a dramatic impact because it increases the number of samples and the signal, so the signal-to-noise ratio increases as $`T`$ for read noise dominated uniform sampling.
## 4 Hardware Considerations
Hardware limitations often run counter to the algorithmic requirements. For instance, in order to make Fowler sampling efficient, it is desirable to have rapid sampling at the beginning and at the end of the integration. But the high frequencies required for high speed sampling add more noise, and A$``$D converters generally have more noise at higher speed. Ultimately, this limitation is related to the noise voltage on a resistor, which is proportional to the square root of the bandwidth, which is inversely related to the sampling time.
Multiple readouts on each detector chip can relax the constraint, but only at the cost of additional wires, readout electronics and complexity. Also, each readout path can have its own offset and gain requiring additional parameters in the data processing.
The sheer volume of data can present a problem. An 8K$`\times `$8K detector running at a sample rate of 1 Hz produces 6 TB in just 12.4 hours of observation. Storing and processing this volume of data daily requires a dedicated computer system.
## 5 Processing
Fowler sampled data are easy to deal with. Ideally, they consist of $`n/2`$ samples at the beginning of an integration and $`n/2`$ samples at the end of the integration. The data from the beginning and end are summed independently, and the final result is the difference between the two. For integrations that are of the same order as the sample time, the readout sequence ends up taking a substantial fraction of the total time. Since the mean time of the readout is the effective time, the effective integration time is reduced to $`Tnt/2`$, where $`t`$ is the time for a single sample. This reduces the efficiency to $`1nt/2T`$ even if the readout noise is completely negligible compared to the signal. The readout noise must be added, of course, but if the number of samples is large relative to $`R^2`$, where $`R`$ is the readout noise in photons (or its equivalent electrons), the readout noise becomes small, and any signal is dominated by the photon noise.
For a typical size detector pixel ($`600\mu `$m<sup>2</sup>), the mean time between cosmic ray events is $`40000`$ sec in space. The problem with a cosmic ray is that all of the information is lost in the affected pixel due to the cosmic ray, and the efficiency drops to zero. So the integration time is a balance between making longer integrations to reduce the readout noise effect and losing information to cosmic rays. Also, if a pixel well is filled, the information is lost.
For a data set formed by uniformly sampling with non-destructive reads, the important signal is the time derivative or slope. Finding the slope of a set of data can be done in a straightforward manner. But there are several important sources of noise, and if attention is paid to each of these, the process can be optimized as shown in $`\mathrm{\S }`$7. Full wells are recognized, and data taken after a well is filled are ignored. This reduces the effect on the pixel from losing all of the information to effectively losing only the time after the well is full. Cosmic ray glitches are identified and eliminated from the data before fitting the slope. Finally, the weighting is adjusted to weight high signal pixels towards the ends of the integration and low signal pixels uniformly. This optimizes the fitting so there is no penalty for the high signal data over the Fowler sampled data. Thus, if the integration is carried out long enough, the readout noise can be effectively removed for even the low signal data, although this will suffer a factor of $`\sqrt{3}`$ noise increase over Fowler sampled data. But if the integration time is a factor of 2 or more longer, uniform sampled data have a higher efficiency.
Many other strategies for sampling can be devised. However, they tend to combine the disadvantage of the hardware complexity of Fowler sampling with the disadvantage of the software complexity of uniform sampling. Unless there are peculiar features of the hardware or noise, they offer little improvement over the two cases presented here.
In Figure 1 we compare the efficiency of uniform sampling with optimum processing to Fowler sampling with its simpler optimum processing. A range of brightnesses is presented along with a range of integration times. We assume a cosmic ray rate of $`25\times 10^6`$ s<sup>-1</sup> per pixel (5 cm<sup>-2</sup> s<sup>-1</sup> for 27 $`\mu \mathrm{m}`$ detectors 10 $`\mu \mathrm{m}`$ thick) and a full well of 70000 e. This introduces the sudden drop to zero in the Fowler sampling and the sharp decrease in the uniform sampling for the brightest sources. We also assume a minimum sample time of 10 seconds, and a maximum number of samples of 64 per integration. This means that for integrations less than 640 seconds, readout is continuous and uniform even for the Fowler case. The uniform curves are higher because the data are optimally processed, while the identical Fowler data are Fowler processed. This limits the Fowler efficiency to 50%. For longer times, it can be seen that the Fowler sampled data have a peak at $`2000`$ sec for bright sources and up to 5000 sec for dim sources. The fall-off is due to cosmic ray hits. Changes in the size or rate of cosmic ray hits change the exact placement of the peaks but not their character.
The uniformly sampled data also suffer from cosmic ray hits, but since they are removed by the processing, their effect is small, and efficiency continues to increase until the wells are filled. Thus, the efficiency of the uniform sampling can exceed that of the Fowler case provided a sufficiently long integration time is used. As noted above, this is approximately a factor of 2 for low level sources.
## 6 Data Handling
Moore’s law (memory doubles every 18 months, Moore 1965) suggests that we should not have to worry about data storage (at least not for long), but detector arrays also double on the same time scale, and the problems of data storage and handling remain. Keeping only the appropriate bits can substantially reduce the data volume. Obviously, any large offset can be removed, and the size of the data can be scaled to the largest measurements. Excessive precision includes noise that is not useful and cannot be compressed. Insufficient precision allows the uncertainties to be dominated by the digitization rather than the original measurement. Making the least significant bit the size of the uncertainty, ($`1\sigma `$), increases the variance by 8%, an additional bit reduces the increase to 2%, and 2 extra bits results in only a .5% variance increase. Thus we arrive at the well-known injunction to keep one digit into the noise but no more.
If the noise of a system is constant, the A$``$D can be arranged so that the $`1\sigma `$ noise is $`a`$ times as large as the least significant bit. The value of $`a`$ is then typically chosen to be in the range 1-10. The precise value depends on the relative cost of keeping extra bits of marginal value to losing a small part of the signal to noise ratio. If the noise of the system is variable, the process is more difficult. The A$``$D can be adjusted for the lowest noise data. The data can then be divided by the noise, multiplied by $`a`$, and the integer result stored. If each datum has a unique noise, a precise noise estimate must be stored for each one, again increasing the number of bits that must be stored.
Poisson distributed data present an efficient alternative. Since the uncertainty is $`\sqrt{N}`$, dividing by the uncertainty gives $`\sqrt{N}`$, and both the signal and the noise have the same record, $`a\sqrt{N}`$. Hence, there is no need for a separate noise file, and the data can be restored to sufficient accuracy simply by dividing by $`a`$ and squaring the result. Although we have looked at this problem in the context of astronomical observations, this particular data compression technique is general and can be used anywhere the variance is proportional to the signal.
In the case under consideration, the noise is not purely Poisson noise, but the variance can be well approximated by $`N+V`$, where $`V`$ is related to the readout noise and the number of samples. Since we are dealing with a variation on a small perturbation of the noise, a reasonable approximation will work. After compression, neighboring values will be combined, by truncating any residual fraction and keeping only the integer part. The key is the range of values that will be combined, not the absolute value which can be recalculated. Consider the form
$$D=a\sqrt{N+f},$$
(1)
where $`a`$ is the scaling factor as before, and $`f`$ is an offset. Now for large signals $`Da\sqrt{N}`$ as desired. For small signals:
$$a/\sqrt{V}=dD/dN=a/(2\sqrt{N+f})$$
(2)
which is satisfied at $`N=0`$ for $`f=V/4`$. The encoding is fast, requiring only an add, square root and multiply. The decoding is also fast and straight-forward. This process reduces the number of bits by a factor of 2 with essentially no loss of information. The advantage remains even after additional lossless compression (Nieto-Santisteban et al. 1999).
## 7 Algorithm
The algorithm to process Fowler sampled data has already been discussed. We now turn to the problem of optimizing an algorithm to process uniformly sampled data. All of the sources of error must be considered to make this a general algorithm. This will unnecessarily complicate it for some cases, but the computer implementation can take advantage of special cases.
The algorithm will assume a nondestructive set of samples for an integration. We will denote the raw data $`P_i`$, and assume $`n+1`$ samples at times \[0,1,…n\]. Thus, the index number is the time, and we look for a signal per sample time. We will also frequently use data differences, $`D_i=P_iP_{i1}`$, which approximate the signal.
The first issue to consider is full wells. Conceptually, the detector collects electrons until the well is full and then stops. Real detectors often show some nonlinearity before the well is full, and may show some “bleeding” into neighboring pixels. In any case, each pixel’s data must be examined to determine if the pixel’s well is full. The examination is most efficient if it proceeds from the last datum towards the first and stops when the first sample less than full is reached. By using all of the data up to the full well, maximum dynamic range is maintained. The limit of dynamic range is then determined by the sampling frequency and the full well size. To account for residual nonlinearity, it is most time efficient to use a look-up table to provide a linearization function.
Next, the data must be searched for cosmic rays and glitches (glitches of either sign might arise from cosmic ray strikes on electronic elements other than the detector, or other sources). Since glitches can appear at any time, all of the data must be examined and compared to what is expected. So, we must make an estimate of what is to be expected. A simple and robust signal estimate is the difference between the last and first data divided by the time: $`s=(P_nP_0)/n`$. A difference between $`s`$ and $`D_i`$ should be within the noise. One might object that $`D_i`$ is included in $`s`$ as $`s=D_i/n`$, but a little algebra shows that this only reduces the difference by a factor of $`(n1)/n`$. This can be ignored for large $`n`$ or included in the expected uncertainty. The expected uncertainty must include 2 readout uncertainties, as $`D_i`$ is a difference, and the Poisson statistics of the photons. But a large cosmic ray hit included in the estimate of $`s`$ could perturb it to the point that other $`D_i`$ are interpreted as outliers. The solution to this problem is to identify the worst offender. If it is within bounds, then all of the other data are as well. If it is not, it must be discarded and excluded from the estimate of $`s`$. The test is then repeated on the remaining data. There remains the question of how to set the bound. We leave this as a tunable parameter to adapt the algorithm to different cosmic ray rates, detector characteristics and observation goals.
Thus, $`s`$ is a good estimate of the signal. However, if the readout noise is the dominant source of noise, a linear fit to the data can reduce the variance by $`12n/[(n+1)(n+2)]`$. If, on the other hand, the noise is dominated by the Poisson statistics of the photons, the linear fit effectively uses only 2/3 of the time. In that case, the noise is worse for the linear fit by 23%.
An optimum weighting is needed. The set of differences, $`D_i`$, has all of the information. The covariance matrix for the set of differences can be expressed as the sum of two matrices. Since the photon noise for each difference is independent of each of the other differences, the matrix describing the covariance due to the photon noise is $`sI`$, where $`s`$ is the signal and $`I`$ is the identity matrix. Neighboring sample differences are correlated by the readout noise. Each sample, $`R_i`$, with its noise, is used twice, once positive (in $`D_i`$) and once negative (in $`D_{i+1}`$). Thus, the readout covariance matrix has a main diagonal of $`2R^2`$ with a diagonal of $`R^2`$ on either side of the main diagonal. The sum of these two matrices can be inverted to form a weight matrix. Since each difference is an estimate of the signal, the sums of the columns (or rows as the weight matrix is symmetric) form a weight vector that optimizes the signal when applied to the differences. By subtracting the weight vector from the weight vector shifted by 1, the difference operation is effectively transferred to the weight vector, and the result can be applied directly to the raw data.
The final weight vector depends on the number of data elements and the signal-to-noise ratio $`s/R^2`$. For large $`s/R^2`$, only the two end points are used, and for small $`s/R^2`$ the weights have a linear slope with mean of 0 (see Figure 2). Relatively few cases of $`s/R^2`$ can approximate the ideal full matrix calculation quite well. In the algorithm shown below, we use 8 approximations, and select the appropriate one by using the previously calculated $`s`$. Selecting a poor weight vector does not bias the result, it only increases the noise (and noise estimate) of the result. The error is not symmetric. Using a weight for high signal-to-noise ratio for a low signal-to-noise ratio case can result in an increase of order $`n`$ in the variance while an error in the other direction at most increases the variance by 33%. So we calculate the weight vectors for the bottom of each bin, see Figure 3.
## 8 Non-Ideal Detectors
All detectors are expected to be non-ideal, particularly after exposure to a space radiation environment (The NICMOS Data handbook). The algorithm described above was developed to demonstrate the advantages of uniform sampling, but it does not directly address detector faults. The strategy for accounting for each fault depends on its type.
Each pixel may have a different noise level and responsivity, and these parameters may change with time, particularly after a cosmic ray event depositing large amounts of charge. If a single cosmic ray can produce a responsivity change, then the algorithm would have to test for that effect and decide whether data taken after the cosmic ray are consistent with those (presumably correct) data before it. If the noise level can change with time, then the threshold criterion for recognizing a glitch may need to be adaptive rather than using a stored constant. However, neither of these cases is expected to raise the computation time very much, since most of the time is still devoted to input.
Many detectors show a distinct initial transient after the reset of the integrators. If this effect has the same general shape for all pixels, it can be modeled and subtracted from the measured data before processing to search for glitches. Such a model might be implemented as a simple formula, or as a table lookup. If the effect differs significantly between pixels, then a more complex model or table lookup may be required. However, it seems unlikely that a completely separate model for every pixel would be required. Improper treatment of such transients could affect the photometric accuracy for pixels that have glitches or become saturated early. On the other hand, such data are lost altogether for Fowler sampling. Improper treatment could also lead to false alarms, e.g. glitches that are the result only of the repeatable initial transient.
It may also be that each cosmic ray produces a temporary response that disqualifies data taken shortly afterwards. If this effect can be described by a simple dead time, it is easily handled in the code. If there is a long term effect with a constant waveform, it can be corrected by modeling and subtracting it. If it is still too difficult to correct, then data taken after the glitch can be ignored until the next exposure. The uniform sampling with cosmic ray rejection would then have a lower efficiency, but still would be superior to Fowler sampling in which such pixels can not be used at all.
The algorithm described here makes no use of spatial information. However, a processor could easily examine the neighbors of every pixel where a glitch or saturation has been found, to see if there is any odd behavior there. There are two problems: First, in this process it is very easy to introduce spatial correlations which can contaminate the results. Second, the processing can become very involved making it slow. Either of these is incompatible with our attempt to produce a simple, fast and robust process to remove CR and compress data from large-area detectors.
Some faults might not be in the detectors. A variable pointing error would produce a variable brightness incident on pixels where spatial gradients are high, leading to false detections of glitches. These could be handled by raising the threshold for glitch detection for bright pixels, possibly by using an adaptive calculation.
To summarize, detector faults will require additional code in the processing algorithm, but this additional code is not expected to produce large computational loads, nor to invalidate or bias the data. For long exposures, uniformly sampled data will still be valid over a larger fraction of the pixels than Fowler sampled data.
## 9 Algorithm Code, Processing Time and Performance
The process outlined here does not imply anything about neighboring pixels. This reduces the likelihood of generating any biases in the data. It also keeps the process simple and short. Only 6 operations per datum plus 6 operations per pixel are required to check, fit and compress the data. Rejected data are somewhat more expensive, but presumably they are relatively rare, so the entire process can take place on a modest computer for even enormous data flows.
The data from an array are usually collected in the order of columns, rows and then sample iterations. The order of the data required for the algorithm is by sample iteration first and then pixel (either columns or rows first). This has two implications. First, an entire read set must be stored in memory. For example, in the Next Generation Space Telescope (NGST) with 64 reads in an integration, this set is 8 GB (2 bytes $`\times `$ 64 reads $`\times `$ 1 M pixel per chip $`\times `$ 64 chips, see The NGST Study Team, 1997), so the task is not trivial.
Second, to speed the memory operations, it is convenient to have the addresses for the computer reads in a different order than for the A$``$D converter writes. If an even power of 2 is adopted for the number of samples, this is easily achieved by swapping some of the address bits and using a double buffer so that the computer can be processing a set of data while the next set is being collected.
Since the compression ratio is large (of order 200 is easily achievable for 64 samples), the output storage and transmission are greatly simplified. The input and output bounds on the computer are entirely dominated by the input side.
The algorithms introduced in Offenberg et al. (1999) and Nieto-Santisteban et al. (1999) have now been optimized to reduce computer requirements and improve performance. First, saturated data are marked and not used. Next, for each pixel, the set of samples (64) is fit to a straight line. The interval with largest deviation from the line (either direction) is compared with the expected noise. If it is larger than $`4.5\sigma `$ (optimum for test case), the interval is not used in the fit, and the process is repeated. Most of the processing time is used by the cosmic ray rejection.
Next, a weighted fit is applied to the remaining data. The optimum fit depends on the signal. High signal uncertainties are dominated by photon (electron) counting noise, and the optimum fit weights the endpoints. Low signal uncertainties are dominated by readout noise, and the optimum fit is uniform weighting. We calculate the weights for these and 6 intermediate signal-to-noise ratios and choose the best weighting scheme for the signal. By computing the weight table for all 8 signal-to-noise ratio levels ($`<1\%`$ of total computation time) and all possible segment lengths, we save time in the weighted fit.
After the fit, we reduce the dynamic range and equalize the noise for the different pixels by finding the square root of the slope plus an offset, which compensates for the readout noise. Finally, an adjustable scaling allows retention of $`a`$ bits of noise after conversion to an integer. Thus N (64) 16 bit samples are converted to a single 8 bit byte. This can be further compressed without loss (see Nieto-Santisteban et.al. 1999) to approximately 3 bits per pixel (if we keep 2 bits of noise).
The final results are robust. Even integration times that lead to most of the pixels being affected by cosmic rays can be effectively cleaned, allowing longer integration times than are practical with Fowler sampling. This leads to higher efficiency than Fowler sampling. The difference can be as large as a factor of 2, which translates directly to saving half of the total integration time while getting the same quality data.
Figure 4 demonstrates the effectiveness of cosmic ray rejection. The cosmic rays in the raw image (or, equivalently, in a Fowler-sampled image) make long integration ($`>1000`$ sec) undesirable. If we take non-destructive samples every 30 seconds during the integration, it is possible to get a clean image of long integration times, with high photometric reliability. The images shown in Figure 1 are stretched to the same grey-scale and assume a cosmic ray rate of 5 event/sec/cm<sup>2</sup>, which is the low end estimate for cosmic ray rates in deep space (L2 orbit), such as planned for the Next Generation Space Telescope.
The lower pair of plots in Figure 5 compares photometry for Fowler sampling and uniform sampling with processing for five 2000 second integrations. The resulting images are processed by throwing out the outliers and using IRAF DAOPHOT on the final image. The Fowler sampled data are still contaminated with cosmic rays (CR), as expected, so the ideal integration time is shorter. But the ideal integration time is longer for uniform sampling with deglitching. The upper pair of plots in Figure 5 compare the optimum twenty 500 sec Fowler sampled images with a single 10000 sec uniform sampled image after deglitching. In all cases, a total of 320 samples were executed for 10000 seconds of observation. The uniform sampled image is higher quality even though the data rate is only 1/40 as large (after data compression).
As Table 1 indicates, even though the processing (including CR rejection) of this algorithm takes longer than the processing of Fowler sampled data, the total time is dominated by I/O and in particular input. If the input speed can be raised to the data rate, the remaining processing for either the Fowler sampling or this algorithm is accommodated on a modest computer.
## 10 Conclusions
The algorithm presented here is robust as it excises points far from the line and uses a linear fit on the points near the line. It preserves the maximum dynamic range allowed by the hardware readout and rejects almost all cosmic ray hits. Its adjustable weighting efficiently uses uniformly sampled data, and yet it uses a minimum of computer resources.
These studies are supported by the NASA Remote Exploration and Experimentation Project (REE), which is administered at the Jet Propulsion Laboratory under Dr. Robert Ferraro, Project Manager.
## Appendix A Cosmic Ray Rejection Program
The following is the complete algorithm. The program ”init\_wt”, is required only once to set up the weight table, WT. The output of the program, ”cr\_rej” is a compressed best fit image and a list of the number of glitches found at each location.
```
#include <math.h>
#include "simpleimage.h"
#define e 2.71828
#define E .0024788 //smallest S/N ratio bin
#define K 8 //number of signal (S) cuts
#define M 64 //Max number of reads
float WT[K][M][M/2+1];
void init_wt(float *a, float *f, float *v, float *q,
float B, //Bits of noise to keep
float P, //Gain (photons/count)
float V, //Readout variance (counts^2)
int N, //Number of reads
float S) //Sigma for CR cutoff
{ float *W; double Z[M],Y[M],s,w,x,y,z; register int i,k,n;
*a=B*sqrt(P); Ψ//Renormalize Bits
*f=3.*V/(N*P*(N+1)*(N+2)); //Output offset
*q=S/P; //Renormalize cutoff
*v=2.0* *q*V*(N*N+1)/(P*N*(N-1)); //Renormalize variance
for(s=E,k=0;k<K;s*=e,k++){WT[k][0][0]=0; //loop on S/N ratio, s.
for(Z[0]=y=x=n=0;++n<N;){W=WT[k][n]; //for all ramp lengths
Y[n]=x=1/(2+s-x);y++; //New x New last Y
for(z=i=0;++i<n;)z+=Z[i]+=y*(Y[i]*=x); //Update row and sum
for(y=x,i=0;++i<n;)y+=Y[i]; //Sum weight
for(i=0;i<(n+1)/2;i++)W[i]=(Z[i]-Z[i+1])*P*(N-1); //Differences
W[i]= z+(Z[n]=y);} //final weight
for(x=1/(z+y),i=0;i<N/2;i++)W[i]*=x;}} //Renormalize last row
void cr_rej(//Routine to reject cosmic rays and perform linear fit of data.
short int *DI, //Input: Data cube (np x N) containing the data.
const int N, //Input: number of images in the data set.
simpleimage *CD, //Output: contains the fitted image data.
simpleimage *CR, //Output: number of detected CR
const int np, //Input: number of pixels in the data set.
const int F, //Input: count for Full-well.
const float a, //Input: noise bits to keep in sqrt-scaling
const float f, //Input: offset for sqrt-scaling
const float v, //Input: renormalized read variance
const float V, //Input: Read variance (counts^2)
const float q) //Input: Chi^2 limit in counts
{ register int p,c,k; register short int *D, *P;
const float EVe=E*V*e; const int n=N-1;
register float s,x,y,*U,*W,*X,*Y,Z[M],*C[M]; U=Z+n;
for(P=DI+N*(p=np);p-->0;){D=P;P-=N; //Do for all pixels
if(P[1]>F){CR->setval(p,N);CD->setval(p,0);} //If no good data exit
else{while(*--D>F); Ψ//Find last good data
for(c=-1,X=Y=Z+(k=D-P);(C[++c]=Y++)<U;); //Delete Full Reads
for(Y=X;D>=P;*Y--=*D--); //Convert to floats
s=(*X-*Z)/k; //Estimate Signal
for(x=v,Y=X;Y>Z;)if((y*=y=s-*Y+*--Y)>x){X=Y;x=y;} //Find Worst Point
for(;x>s*q+v;X--){ //Is it a CR?
s+=(s+*X-*(X+1))/(n-++c); //fix s for CR
for(k=c;((C[k]=C[k-1])>X)&&--k;);C[k]=X; //fit CR in list
for(k=0,x=v,Y=Z;k<=c;++Y){W=C[k++]; //for all segments
while(Y<W)if((y*=y=s+*Y-*++Y)>x){X=Y;x=y;}}} //Find Worst Point
CR->setval(p,c); //Record # CR
for(k=0,x=EVe;s>x&&++k<K-1;x*=e); //Find proper S/N Bin
if(c){for(Y=Z,s=y=c=0;Y<U;Y=C[c++]+1){ //for all segments
W=WT[k][C[c]-Y]; //Find right weights
for(X=C[c];Y<X;)s+=*W++*(*Y++-*X--); //total this segment
y+=*W;} //Sum of weight
s=s/y+f;} //Final Sig
else for(W=WT[k][n],s=f,X=U;*W<0;s+=*W++*(*Y++-*X--));//0CR weighted fit
CD->setval(p,(s<0)?0:int(a*sqrt(s)));}}} //Record data
```
|
warning/0005/gr-qc0005023.html
|
ar5iv
|
text
|
# Semiclassical Cosmological Perturbations Generated during Inflation
## I INTRODUCTION
One of the key problems in modern cosmology is that of cosmic structure formation . If an inflationary period is present, the initial seeds for structure formation are supposed to be originated by the quantum fluctuations of the inflaton field, which is responsible for driving inflation . By semiclassical back reaction on the spacetime geometry, these quantum fluctuations will, in turn, produce fluctuations on the spacetime metric. Here we want to look at this problem within the context of a simple chaotic inflationary model by means of a recently suggested formalism. In this formalism classical metric fluctuations induced by quantum matter fluctuations are described by a Langevin-type equation . This is an alternative to the more usual approach in which some perturbative degrees of freedom of the gravitational field are also quantized .
The idea behind this approach is to relate the back-reaction problem in semiclassical gravity with the dynamics of open quantum systems. In fact, there are a number of situations in which one is interested in the observables and the dynamics of a few degrees of freedom from a whole closed quantum system undergoing unitary evolution. These degrees of freedom constitute an open system whose dynamics is no longer unitary due to its interaction with the remaining degrees of freedom of the whole system, which constitute the environment .
For the existence of a semiclassical regime for the system dynamics two requirements are needed . The first is decoherence, which guarantees that probabilities can be consistently assigned to histories describing the evolution of the system. The second is that these probabilities should be peaked near histories which correspond to solutions of classical equations of motion. The effect of the environment plays a crucial role in the semiclassical dynamics of the system. In fact, on the one hand, it may provide enough induced decoherence through the entanglement between system and environment . On the other hand, the environment back reaction on the system dynamics will produce both dissipation and noise (commonly connected by fluctuation-dissipation relations). The environment may, thus, induce a semiclassical stochastic dynamics on the system, which may be suitably described by a Langevin-type equation .
The plan of the paper is the following: In Sec. II we give a brief summary of the Einstein-Langevin equation. We apply this formalism in Sec. III to study the generation of cosmological gravitational perturbations during inflation by considering the simplest model leading to chaotic inflation. We finally discuss our main conclusions in Sec. IV. Throughout the paper we use natural units ($`\mathrm{}=c=1`$) and the $`(+,+,+)`$ sign convention of Ref. .
## II EINSTEIN-LANGEVIN EQUATION
In the context of semiclassical gravity one treats the matter fields as quantum fields on a classical curved spacetime. As a consequence of their energy density, these fields act as gravitational sources which modify the spacetime geometry. To study this back-reaction effect one usually uses the so-called semiclassical Einstein equation
$$G_{ab}[g]=\frac{8\pi }{m_p^2}T_{ab}[g,\widehat{\varphi }[g]]_{ren},$$
(1)
where the renormalized expectation value of the stress tensor operators of the quantum matter fields in some quantum state are introduced as gravitational sources. There are, however, some situations in which the fluctuations of the stress tensor operator are important . In those cases we cannot expect that the semiclassical Einstein equation provides the actual dynamics of the spacetime metric any longer, but some kind of averaged description.
It may be useful to consider the spacetime metric as an open system which interacts gravitationally with the quantum matter fields, which constitute the environment . In this case the system will exhibit a stochastic dynamics with fluctuations due to the noise induced by the environment. In order to take this effect into account, the following modified equation, known as Einstein-Langevin equation, has been suggested :
$$G_{ab}[g+h]\frac{8\pi }{m_p^2}\widehat{T}_{ab}[g+h]_{ren}=\frac{8\pi }{m_p^2}\xi _{ab}[g],$$
(2)
where $`g`$ is a solution of the semiclassical Einstein equation which is used as the background metric, whereas $`h`$ is a linear perturbation. The field $`\xi _{ab}[g]`$ is a Gaussian stochastic classical source with the following properties:
$`\xi _{ab}(x)_\xi `$ $`=`$ $`0`$ (3)
$`\xi _{ab}(x)\xi _{cd}(y)_\xi `$ $`=`$ $`{\displaystyle \frac{1}{2}}\{\widehat{t}_{ab}(x),\widehat{t}_{cd}(y)\}[g],`$ (4)
where $`\widehat{t}_{ab}(x)\widehat{T}_{ab}(x)\widehat{T}_{ab}(x)`$. We use the two different notations $`_\xi `$ and $``$ to explicitly distinguish the average associated to a classical stochastic process from the expectation value of quantum operators. The correlation function for the stochastic source, which will generate a stochastic dynamics on the spacetime geometry, was precisely chosen to take into account the quantum fluctuations of the stress tensor.
## III COSMOLOGICAL PERTURBATIONS GENERATED DURING INFLATION
Let us now consider the simplest model leading to chaotic inflation , which is driven by a massive real scalar field $`\widehat{\varphi }`$ minimally coupled to the spacetime curvature (this field is usually called the inflaton). The corresponding Lagrangian density is, thus:
$$(\widehat{\varphi })=\frac{1}{2}g^{ab}_a\widehat{\varphi }_b\widehat{\varphi }+\frac{1}{2}m^2\widehat{\varphi }^2$$
(5)
A few comments are in order. First of all, the condition for the existence of an inflationary period (characterized by an accelerated expansion of spacetime) is that the value of the field averaged over a region with a typical size equal to the Hubble radius (the so-called horizon scale) is higher than the Planck mass, $`m_p`$. In fact, in order to have enough inflation to solve the horizon and the flatness problem, more than 60 e-folds are needed. To achieve that, the scalar field should begin with a value higher than $`3m_p`$. On the other hand, as will be shown below, the small value of the CMB (Cosmic Microwave Background) large scale anisotropies measured by COBE imposes a severe constraint on the inflaton mass $`m`$, which should be of the order of $`10^6m_p`$.
We want to study small metric perturbations around a Robertson-Walker geometry. For this purpose we need to deal with the corresponding gauge freedom either by choosing a particular gauge or by working with gauge invariant quantities . We will restrict our study to scalar-type perturbations of the metric. The expression for the perturbed metric in the longitudinal gauge is then:
$$ds^2=a^2(\eta )\left((1+2\mathrm{\Phi }(x))d\eta ^2+(12\mathrm{\Psi }(x))\delta _{ij}dx^idx^j\right),$$
(6)
where the two functions $`\mathrm{\Phi }(x)`$ and $`\mathrm{\Psi }(x)`$ correspond in this case to Bardeen’s gauge invariant variables and $`a^2(\eta )`$ is the cosmological scale factor of the background Robertson-Walker geometry. As shown below, the Einstein-Langevin equation (2) is gauge invariant. Therefore, we can work in a given gauge and finally extract the desired gauge invariant quantities in a consistent way. To see how the first member of Eq. (2) is gauge invariant, one uses the following result for linear perturbations in $`h`$:
> $`A_b^a[g+h]`$ is gauge invariant if and only if $`_\stackrel{}{\varsigma }(A_b^a[g])=0`$ for any vector field $`\stackrel{}{\varsigma }(x)`$ and this is equivalent to $`A_b^a[g]\delta _b^a`$ (zero being a particular case).
The first member of our Einstein-Langevin equation is, thus, gauge invariant if $`G_{ab}^{(0)}[g](8\pi /m_p^2)\widehat{T}_{ab}^{(0)}[g]_{ren}=0`$, but this is indeed the case since the background metric $`g`$ is taken to be a solution of the semiclassical Einstein equation. On the other hand, the second member of Eq. (2) is explicitly gauge invariant since it does not depend on the perturbed metric.
It is convenient to decompose the inflaton scalar field in the following way: $`\widehat{\varphi }(x)=\varphi (t)+\widehat{\phi }(x)`$, where $`\varphi (t)`$ is the homogeneous background solution, which is compatible with the background metric through the semiclassical Einstein equation, whereas $`\widehat{\phi }(x)`$ corresponds to a free massive quantum scalar field with zero expectation value on the spacetime with the background metric: $`\widehat{\phi }(x)_g=0`$. The two main ingredients that we need for our Einstein-Langevin equation are the renormalized expectation value of the stress tensor on the spacetime with the perturbed metric $`\stackrel{~}{g}=g+h`$, and the noise kernel, which takes into account the fluctuations of the stress tensor evaluated on the background metric. The stress tensor of a minimally coupled massive scalar field is:
$$\widehat{T}_{\mu \nu }=_\mu \widehat{\varphi }_\nu \widehat{\varphi }+\frac{1}{2}\stackrel{~}{g}_{\mu \nu }(_\mu \widehat{\varphi }^\mu \widehat{\varphi }+m^2\widehat{\varphi }^2).$$
(7)
Using the decomposition for the scalar field introduced above, we rewrite the renormalized expectation value for the stress tensor as
$$\widehat{T}_{\mu \nu }[g+h]^{ren}=\widehat{T}_{\mu \nu }[g+h]_{\varphi \varphi }+\widehat{T}_{\mu \nu }[g+h]_{\varphi \phi }+\widehat{T}_{\mu \nu }[g+h]_{\phi \phi }^{ren},$$
(8)
where only the homogeneous solution for the scalar field contributes to the first term. The second term is proportional to $`\widehat{\phi }[g+h]`$, but this quantity is no longer zero since the field dynamics is considered on the perturbed spacetime. Finally, the last term corresponds to the expectation value of the stress tensor for a free scalar field on a spacetime with the perturbed metric. In the usual approach when computing fluctuations during inflation, $`\widehat{\phi }`$ is treated perturbatively. This last term being quadratic in $`\widehat{\phi }`$, is of higher order and will not be taken into account.
As for the noise kernel, after using the previous decomposition, the following expression is obtained:
$$\{\widehat{t}_{\mu \nu },\widehat{t}_{\rho \sigma }\}[g]=\{\widehat{t}_{\mu \nu },\widehat{t}_{\rho \sigma }\}_{\varphi \phi }[g]+\{\widehat{t}_{\mu \nu },\widehat{t}_{\rho \sigma }\}_{\phi \phi }[g],$$
(9)
where we have used the fact that $`\widehat{\phi }_g=0=`$ $`\widehat{\phi }\widehat{\phi }\widehat{\phi }_g`$ for Gaussian states (those considered here) on the background geometry. It is important to note that both contributions to the noise kernel (the first term is quadratic in $`\widehat{\phi }`$ whereas the second one is quartic) are “conserved” separately since both $`\varphi (t)`$ and $`\widehat{\phi }(x)`$ satisfy the Klein-Gordon equation on the background geometry. Due to this fact, the two corresponding stochastic sources can be consistently considered in an independent way. We are, thus, allowed to concentrate on the source associated to the first term from now on. The contribution of a term of the same sort as the second one has been discussed elsewhere . One can check that the space-space components coming from the stress-tensor expectation value terms that we are considering and the stochastic source are diagonal, i.e., $`\widehat{T}_{ij}=0=\xi _{ij}`$ for $`ij`$. This, in turn, implies that the two gauge invariant quantities used to characterize the scalar-type metric perturbations must be equal: $`\mathrm{\Phi }=\mathrm{\Psi }`$ .
Let us write the Einstein-Langevin equation in Fourier space and consider the $`0i`$-component:
$$2ik_i(\mathrm{\Phi }_k+\mathrm{\Phi }_k^{})=\frac{8\pi }{m_p^2}\xi _{k\mathrm{\hspace{0.33em}0}i},$$
(10)
where $`k_i`$ is the comoving momentum component associated to the comoving coordinate $`x^i`$ (throughout the paper we use the subindex $`k`$ to denote the comoving momentum vector $`\stackrel{}{k}`$ that labels the Fourier modes in flat space), primes denote derivatives with respect to the conformal time $`\eta `$ and $`a^{}(\eta )/a(\eta )`$. The first member is just the linearized Einstein tensor for the perturbed metric (6) . There should also appear a non-local term of dissipative character coming from the second term in (8), which we have not considered in this work, where we are mainly concerned about the fluctuating part.
From this equation we may obtain the metric perturbations $`\mathrm{\Phi }_k`$ in terms of the stochastic source $`\xi _{k\mathrm{\hspace{0.33em}0}i}`$. For this purpose we need the retarded propagator for the gravitational potential $`\mathrm{\Phi }_k`$, i.e., the required Green function to solve the inhomogeneous first order differential Eq. (10) with the appropriate boundary conditions:
$$\stackrel{~}{G}_k^{ret}(\eta ,\eta ^{})=i\frac{4\pi }{k_im_p^2}\left(\theta (\eta \eta ^{})\frac{a(\eta ^{})}{a(\eta )}+f(\eta ,\eta ^{})\right),$$
(11)
where $`f(\eta ,\eta ^{})`$ is a homogeneous solution related to the chosen initial conditions. If we take, for instance, $`f(\eta ,\eta ^{})=\theta (\eta _0\eta ^{})a(\eta ^{})/a(\eta )`$, we would obtain the stochastic evolution of the metric perturbations for $`\eta >\eta _0`$ due to the effect of the stochastic source after $`\eta _0`$. The correlation function for the metric perturbations is then given by the following expression:
$`\mathrm{\Phi }_k(\eta )\mathrm{\Phi }_k^{}(\eta ^{})_\xi `$ $`=`$ $`(2\pi )^3\delta (\stackrel{}{k}+\stackrel{}{k}^{}){\displaystyle ^\eta }𝑑\eta _1{\displaystyle ^\eta ^{}}𝑑\eta _2\stackrel{~}{G}_k^{ret}(\eta ,\eta _1)\stackrel{~}{G}_k^{}^{ret}(\eta ^{},\eta _2)`$ (13)
$`\xi _{k\mathrm{\hspace{0.33em}0}i}(\eta _1)\xi _{k^{}\mathrm{\hspace{0.33em}0}i}(\eta _2)_\xi .`$
And the correlation function for the stochastic source is, in turn, connected with the stress-energy fluctuations:
$$\xi _{k\mathrm{\hspace{0.33em}0}i}(\eta _1)\xi _{k\mathrm{\hspace{0.33em}0}i}(\eta _2)_\xi =\frac{1}{2}\{\widehat{t}_{0i}^k(\eta _1),\widehat{t}_{0i}^k(\eta _2)\}_{\varphi \phi }=\frac{1}{2}k_ik_i\varphi ^{}(\eta _1)\varphi ^{}(\eta _2)G_k^{(1)}(\eta _1,\eta _2),$$
(14)
where $`G_k^{(1)}(\eta _1,\eta _2)=\{\widehat{\phi }_k(\eta _1),\widehat{\phi }_k(\eta _2)\}`$ is the $`k`$-mode Hadamard function for a free minimally coupled scalar field which is in a state close to the Euclidean vacuum on an almost de Sitter background.
The so-called “slow-roll” parameters account for the fact that the background geometry is not exactly that of de Sitter spacetime (for which $`a(\eta )=1/H\eta `$ with $`\mathrm{}<\eta <0`$). It is also useful to compute the Hadamard function for a massless field and consider a perturbative expansion in terms of the dimensionless parameter $`m/m_p`$, for which observations seem to imply, as will be seen below, a value of the order of $`10^6`$. Thus, we will consider $`\overline{G}_k^{(1)}(\eta _1,\eta _2)=a(\eta _1)a(\eta _2)G_k^{(1)}(\eta _1,\eta _2)=0|\{\widehat{y}_k(\eta _1),\widehat{y}_k(\eta _2)\}|0`$ such that $`\widehat{a}_k|0=0`$ with $`\widehat{y}_k(\eta )=a(\eta )\widehat{\phi }_k(\eta )=\widehat{a}_ku_k(\eta )+\widehat{a}_k^{}u_k^{}(\eta )`$ and $`u_k(\eta )=(2k)^{1/2}e^{ik\eta }(1i/k\eta )`$ corresponding to the positive frequency $`k`$-mode for a massless minimally coupled scalar field in the Euclidean vacuum state on a de Sitter background .
The result to lowest order in the mass $`m`$ of the inflaton field and the “slow-roll” parameters is:
$`\mathrm{\Phi }_k(\eta )\mathrm{\Phi }_k^{}(\eta ^{})_\xi `$ $`=`$ $`{\displaystyle \frac{64\pi ^5}{m_p^4}}\left(a(\eta )a(\eta ^{})\right)^1\delta (\stackrel{}{k}+\stackrel{}{k}^{}){\displaystyle _{\eta _0}^\eta }𝑑\eta _1{\displaystyle _{\eta _0}^\eta ^{}}𝑑\eta _2a(\eta _1)a(\eta _2)\dot{\varphi }(\eta _1)\dot{\varphi }(\eta _2)\overline{G}_k^{(1)}(\eta _1,\eta _2)`$ (15)
$`=`$ $`64\pi ^5\left({\displaystyle \frac{m}{m_p}}\right)^2k^3\delta (\stackrel{}{k}+\stackrel{}{k}^{}){\displaystyle _{k\eta _0}^{k\eta }}d(k\eta _1){\displaystyle _{k\eta _0}^{k\eta ^{}}}d(k\eta _2){\displaystyle \frac{k\eta }{k\eta _1}}{\displaystyle \frac{k\eta ^{}}{k\eta _2}}`$ (17)
$`\left[\mathrm{cos}k(\eta _1\eta _2)\left(1+{\displaystyle \frac{1}{k\eta _1k\eta _2}}\right)\mathrm{sin}k(\eta _1\eta _2)\left({\displaystyle \frac{1}{k\eta _1}}{\displaystyle \frac{1}{k\eta _2}}\right)\right]`$
$`=`$ $`64\pi ^5\left({\displaystyle \frac{m}{m_p}}\right)^2k^3\delta (\stackrel{}{k}+\stackrel{}{k}^{})[\mathrm{cos}k(\eta \eta ^{}){\displaystyle \frac{1}{k\eta _0}}(k\eta \mathrm{cos}k(\eta \eta _0)`$ (19)
$`+k\eta ^{}\mathrm{cos}k(\eta ^{}\eta _0))+{\displaystyle \frac{k\eta k\eta ^{}}{(k\eta _0)^2}}]\text{,}`$
where we used the lowest order approximation for $`\dot{\varphi }(t)`$ during “slow-roll” (overdots denote derivatives with respect to the physical time $`t`$): $`\dot{\varphi }(t)m_p^2(m/m_p)`$. We considered the effect of the stochastic source after the conformal time $`\eta _0`$. Notice that the result (19) is rather independent of the value of $`\eta _0`$ provided that it is negative enough, i.e., it corresonds to an early enough initial time. This weak dependence on the initial conditions is rather usual in this context and can be qualitatively understood: after a sufficient amount of time, the accelerated expansion for the quasi-de Sitter spacetime during inflation effectively erases any information about the initial conditions, which is redshifted away. The actual result will, therefore, be very close to that for $`\eta _0=\mathrm{}`$:
$$\mathrm{\Phi }_k(\eta )\mathrm{\Phi }_k^{}(\eta ^{})_\xi =8\pi ^2\left(\frac{m}{m_p}\right)^2k^3(2\pi )^3\delta (\stackrel{}{k}+\stackrel{}{k}^{})\mathrm{cos}k(\eta \eta ^{})\text{.}$$
(20)
## IV CONCLUSIONS
It is of major interest to study the cosmological implications which can be extracted from our work, especially those related to large-scale gravitational fluctuations. These fluctuations are believed to play a crucial role in the generation of the large-scale structure and matter distribution observed in our present universe . They are also tightly connected with the anisotropies in the CMB radiation, which decoupled from matter about $`310^5`$ years after the Big Bang and provides us with very valuable information about the early universe .
From the analysis of our final result in Eq. (20) two main facts can be concluded. First, an almost Harrison-Zel’dovich scale-invariant spectrum seems to be obtained for large scales (small values of $`k`$). Second, no significant relaxation of the coupling parameter is found. Since we get $`\mathrm{\Phi }_k(\eta )\mathrm{\Phi }_k^{}(\eta ^{})_\xi (m/m_p)^2`$ in agreement with the usual results , the small value of the CMB anisotropies detected by COBE imposes a severe bound on the gravitational fluctuations, characterized by $`\mathrm{\Phi }_k(\eta )\mathrm{\Phi }_k^{}(\eta ^{})_\xi `$, which implies $`(m/m_p)10^6`$, whereas the mechanisms considered in those works which allowed an important relaxation of this fine tuning (due to the extremely homogeneous classical initial conditions taken for the inflaton field) resulted in $`\mathrm{\Phi }_k(\eta )\mathrm{\Phi }_k^{}(\eta ^{})_\xi (m/m_p)`$.
It can be shown that genuine quantum correlation functions can be equivalently obtained through a stochastic description based on Langevin-type equations even in regimes where the actual dynamics of the system does not admit a description in classical terms . The case of gravitational perturbations coupled to a scalar field is more subtle due to the existing gauge symmetry associated to diffeomorphic transformations and the subsequent constraints arising in the dynamics of the whole system. Nevertheless, total agreement with the purely quantum treatment is expected at least for the case in which both gravitational inhomogenities and the scalar field are treated perturbatively to linear order .
## ACKNOWLEDGEMENTS
We are grateful to Bei-Lok Hu, Esteban Calzetta and Rosario Martín for interesting discussions. This work has been partially supported by the CICYT Research Project No. AEN98-0431. A. R. also acknowledges support of a grant from the Generalitat de Catalunya.
|
warning/0005/hep-ph0005231.html
|
ar5iv
|
text
|
# 1 We show the 𝑍 boson mass as a function of 𝜇₀/𝑀_{1/2} for various gaugino masses. In this figure, we set 𝑚₀ and 𝐴₀ to be zero. We choose the 𝐵₀ parameter so that tan𝛽=10 at the point 𝑀_𝑍=91 GeV. On the left side of this figure, tan𝛽 becomes close to 1, and the Higgs potential is destabilized.
## A Notation and Convention
The superpotential of minimal supersymmetric standard model (MSSM) is presented as
$$W=Y_uQH_uU^c+Y_dH_dQD^c+Y_eH_dLE^c+\mu H_dH_u,$$
(A.1)
where the SU(2) inner product is defined as
$$H_dH_uH_d^TϵH_u,ϵ=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right).$$
(A.2)
Here, $`Q`$, $`U^c`$, $`D^c`$, $`L`$, $`E^c`$ are matter chiral superfields, and $`H_u`$ and $`H_d`$ are Higgs doublets.
We denote the soft supersymmetry breaking terms as
$`V_{soft}`$ $`=`$ $`m_{H_d}^2|H_d|^2+m_{H_u}^2|H_u|^2`$ (A.3)
$`+m_{\stackrel{~}{q}}^2\stackrel{~}{q}\stackrel{~}{q}^{}+m_{\stackrel{~}{u}}^2\stackrel{~}{u}_R\stackrel{~}{u}_R^{}+m_{\stackrel{~}{d}}^2\stackrel{~}{d}_R\stackrel{~}{d}_R^{}+m_\stackrel{~}{\mathrm{}}^2\stackrel{~}{\mathrm{}}\stackrel{~}{\mathrm{}}^{}+m_{\stackrel{~}{e}}^2\stackrel{~}{e}_R\stackrel{~}{e}_R^{}`$
$`+(A_uY_u\stackrel{~}{q}H_u\stackrel{~}{u}_R^c+A_dY_dH_d\stackrel{~}{q}\stackrel{~}{d}_R^c+A_eY_eH_d\stackrel{~}{\mathrm{}}\stackrel{~}{e}_R^c+h.c.)`$
$`+(B\mu H_dH_u+h.c.)`$
To clarify our notation, we present the left-right component in the scalar top quark mass matrix and chargino mass matrix in the following. The left-right mixing is
$$(A_t+\mu \mathrm{cot}\beta )m_t.$$
(A.4)
The chargino mass matrix is presented as
$$M_{\chi ^+}=\left(\begin{array}{cc}M_2& \sqrt{2}M_W\mathrm{cos}\beta \\ \sqrt{2}M_W\mathrm{sin}\beta & \mu \end{array}\right).$$
(A.5)
The supergravity theories are given by the Kähler potential $`K`$, the superpotential $`W`$ and the gauge kinetic function $`f`$. The scalar potential is given in supergravity as
$$V=e^K[g^{ij^{}}(D_iW)(D_j^{}W^{})3WW^{}].$$
(A.6)
Using the Kähler transformation $`G=K+\mathrm{log}W+\mathrm{log}W^{}`$, we obtain
$$V=e^G[G^iG_i3].$$
(A.7)
The no-scale Kähler potential is written as
$$G=3\mathrm{ln}(T+\overline{T}h(\varphi _i^{},\varphi _i))+\mathrm{ln}|W(\varphi _i)|^2,$$
(A.8)
where $`T`$ is a moduli field and $`\varphi `$ are fields in the visible sector. The function $`h`$ is a Kähler potential for the visible fields. The scalar potential is thus
$$V=\frac{3|W|^2}{(T+\overline{T}h(\varphi _i^{},\varphi _i))^2}\left|\frac{W}{\varphi _i}\right|^2.$$
(A.9)
If the global supersymmetric conditions, $`W/\varphi _i=0`$, are satisfied, the scalar potential for $`T`$ is flat and the gravitino mass, $`m_{3/2}=e^{G/2}`$, is not determined.
Expanding the Kähler potential with respect to visible fields, $`Q`$, we write the Kähler potential as
$$K=\widehat{K}(T,T^{})+\stackrel{~}{K}_{ij^{}}(T,T^{})Q^iQ^j^{}+\frac{1}{2}(H_{ij}(T,T^{})Q^iQ^j+h.c.)+\mathrm{},$$
(A.10)
where the $`T^\alpha `$’s are hidden sector fields (Dilaton and Moduli). The superpotential is given by
$$W=\widehat{W}+\stackrel{~}{\mu }_{ij}Q^iQ^j+\stackrel{~}{Y}_{ijk}Q^iQ^jQ^k+\mathrm{}.$$
(A.11)
Substituting VEV into $`T`$, we obtain the effective superpotential in the flat limit,
$$W_{\mathrm{eff}}(Q)=\mu _{ij}Q^iQ^j+Y_{ijk}Q^iQ^jQ^k.$$
(A.12)
The parameters $`\mu `$ and $`Y`$ are written as
$`\mu _{ij}=e^{\widehat{K}/2}{\displaystyle \frac{\widehat{W}^{}}{|\widehat{W}|}}\stackrel{~}{\mu }_{ij}+m_{3/2}(H_{ij}G^\alpha ^{}_\alpha ^{}H_{ij}),`$ (A.13)
$$Y_{ijk}=e^{\widehat{K}/2}\frac{\widehat{W}^{}}{|\widehat{W}|}\stackrel{~}{Y}_{ijk}.$$
(A.14)
In order to solve the $`\mu `$-problem, we often set $`\stackrel{~}{\mu }`$ to be zero . Then, the $`\mu `$ parameter in the flat limit is proportional to the gravitino mass.
The gaugino mass is given by the gauge kinetic function, $`f_a`$, as
$$M_a=\frac{1}{2}m_{3/2}G^\alpha _\alpha f_a.$$
(A.15)
The supersymmetry breaking scalar mass squared is
$$m_{ij^{}}^2=m_{3/2}^2(\stackrel{~}{K}_{ij^{}}G^\alpha G^\beta ^{}R_{\alpha \beta ^{}ij^{}}).$$
(A.16)
Those two parameters are proportional to the gravitino mass (squared).
The $`A`$ and $`B`$ parameters are complicated, and are not necessarily proportional to the gravitino mass. In this paper, we suppose that $`A`$ and $`B`$ are proportional to the gravitino mass for simplicity.
We note that the parameters $`m_0`$, $`A`$ and $`B`$ are zero at the Planck scale in a strict no-scale model, which is given by the Kähler potential (A.8). In this paper, we loosen the boundary condition.
|
warning/0005/hep-ph0005159.html
|
ar5iv
|
text
|
# Flavour Violation in SUSY SU(5) GUT at Large tan𝛽
## I Introduction
The minimal supersymmetric standard model (MSSM) , one of the best motivated extensions of the Standard Model (SM), has triggered intensive research both in theoretical as well as in experimental physics. Despite of non-trivial constraints on its parameter space coming from collider and low energy experiments the MSSM has so far successfully passed all the tests of precision physics.
In the most general case MSSM contains more than one hundred free parameters. Some of them may give rise to unobserved phenomena like proton decay, large electron dipole moments, large flavour violation, etc. To explain the absence of such phenomena additional assumptions are needed to explain the pattern of supersymmetry (SUSY) breaking parameters. An attempt towards that direction is to regard the MSSM as a low energy remnant of some grand unified theory (GUT) such as SU(5), which predicts unification of gauge couplings. Since the unification naturally occurs in the MSSM, one might try to apply a similar organizing principle to the soft SUSY breaking masses. Another prediction of GUTs is the unification of some or all Yukawa couplings of each generation. To achieve successful Yukawa unification the ratio between the vacuum expectation values (vevs) of the two MSSM Higgs doublets, $`\mathrm{tan}\beta ,`$ is found to be in the range $`\mathrm{tan}\beta 3050`$ for tau-bottom unification and $`\mathrm{tan}\beta 50`$ for tau-bottom-top unification . However, large mixings in the lepton sector may somewhat change this picture .
Stringent tests of SUSY GUTs are offered by flavour violation experiments. The flavour violation in SUSY theories is not suppressed by the high scale where the SUSY breaking parameters are generated but rather by the mass scale of these terms themselves which is believed to be of order TeV. In the scheme of the minimal flavour violation of Barbieri and Hall the SUSY breaking parameters are generated above the GUT scale $`M_{GUT}2\times 10^{16}`$ GeV, at the reduced Planck scale $`M_P2.4\times 10^{18}`$ GeV by gravitational interactions and are therefore universal at $`M_P.`$ The renormalization group (RG) evolution below $`M_P`$ induces non-universalities of the soft terms at the GUT scale where the SU(5) gauge group breaks down to the usual MSSM gauge symmetry group. Therefore the flavour mixings present in the Yukawa couplings at GUT scale cannot be rotated away and should be reflected at low energies in the squark and slepton mass matrices. The phenomenology of this scenario in SU(5) SUSY GUT has been studied with the assumption that the top quark Yukawa coupling is the only sizable one which implies flavour violation in the right-handed slepton and left-handed down squark sectors. Rates of the lepton flavour violating (LFV) processes $`\mu e\gamma ,`$ $`\mu e`$ conversion in nuclei and $`\tau \mu \gamma `$ are found to be large for some parts of the SUSY parameter space but due to the cancellations between gaugino and higgsino loops, they almost vanish for some other parts of the parameter space .
Another set of SUSY theories predicting large rates of flavour violation are the ones with right-handed massive neutrinos . These models are motivated by the Super-Kamiokande results which imply maximal mixing between tau and muon neutrinos. The large mixing in the neutrino sector induces large flavour mixings in the left-handed slepton and right-handed down squark sectors in these models. The rates of flavour violating processes may in this case be much larger than in the minimal model, for example, the branching ratios of LFV processes can be close to or even exceed the present experimental bounds.
The aim of the present work is to revisit the minimal flavour violation scenario in the SUSY SU(5) grand unified theory. We extend the considerations of the previous papers in several directions. Following the hints of possible Yukawa unification<sup>*</sup><sup>*</sup>*Values of $`\mathrm{tan}\beta `$ between 0.7 and 1.8 are excluded by the direct LEP2 searches for the lightest MSSM Higgs boson . At the end of LEP2 run, values of $`\mathrm{tan}\beta `$ below 2.6 will be probed . This constraint is not very restrictive but hints into the same direction as the Yukawa unification. we consider the case when all third generation Yukawa couplings are large and given by the renormalizable physics above the GUT scale. In this case, sizable flavour violation in the left-handed (right-handed) slepton (down squark) sector is induced due to the renormalization effects of down type Yukawa couplings between GUT and Planck scales, in addition to the flavour violation in the right-handed slepton sector. Thus the pattern of flavour violation in SUSY SU(5) GUT at large $`\mathrm{tan}\beta `$ resembles the flavour violation pattern in models with massive neutrinos, and deserves phenomenological studies. We do not make an attempt to identify and to study only that part of the soft terms parameter space in which tau-bottom Yukawa unification is achieved starting from the low energy parameters, but instead assume that there may be large corrections depending on the details of the GUT theory. We allow $`\mathrm{tan}\beta `$ to be a free, although large, parameter so that the effects of $`\tau `$ and $`b`$ Yukawa couplings become non-negligible. The details of our calculations are given below. In our numerical analyses we consider the case where Yukawa unification is possible as well as the case where it is not.
In addition, we calculate the new flavour physics contributions to $`K\overline{K}`$ and $`B\overline{B}`$ mixings as well as to $`bs\gamma ,`$ $`\mu e\gamma ,`$ $`\mu e`$ conversion in nuclei and $`\tau \mu \gamma `$ branching ratios. We also comment on the possibility of large GUT phases in this context. There is an increasing amount of constraints on the MSSM mass spectrum coming from direct collider searches as well as from indirect low energy measurements. We take these into account when calculating allowed ranges for the physical observables. In particular, the constraints coming from $`bs\gamma `$ turn out to allow only the LFV processes to be in the phenomenologically interesting ranges.
The main motivation for the present work is to study how sensitively the planned next generation LFV experiments will probe the SUSY model considered here. In the near future the branching ratio of the decay $`\mu e\gamma `$ will be probed with the sensitivity below $`10^{14}`$ and $`\mu e`$ conversion in nuclei below $`10^{16}`$ (in more distant future $`\mu `$ factories may further reduce these numbers by orders of magnitude ). We shall show that if these experiments will indeed reach the planned sensitivity then, in the scenario considered in this work, LFV must be observed unless the SUSY scale $`\sqrt{m_{\stackrel{~}{t}_1}m_{\stackrel{~}{t}_2}}`$ is above TeV scale. This conclusion remains valid even in the case where the only source of LFV is the right-handed slepton mass matrix as in Refs. since the deep cancellation in the branching ratios observed in Ref. occurs at slepton masses which imply multi-TeV SUSY scale. In this context it is interesting to note that if we require approximate $`\tau `$-$`b`$ Yukawa unification at GUT scale then the sparticle masses are required to be too high for direct production at future collider experiments. In this case, the observation of LFV may turn out to be the only signal of supersymmetry.
The paper is organized as follows. In the second section we present the crucial parts of model we are using throughtout this work, especially the relevant mixing matrices are introduced. In section III renormalization from the reduced Planck scale to the electroweak scale is discussed. In section IV we present our numerical results. Finally in section V we summarize our conclusions.
## II Sources of Flavour Violation
In this section we present some details of the minimal flavour violation scenario considered in our work. We assume that the supersymmetric SU(5) grand unified theory is valid at mass scales between $`M_P`$ and $`M_{GUT}.`$ In the SU(5) model there are three generations of matter multiplets $`\psi _i`$ and $`\varphi _i,`$ $`i=1,`$ 2, 3, which form the $`\mathrm{𝟏𝟎}`$ and $`\mathrm{𝟓}^{}`$ dimensional representations of SU(5), respectively, and $`\mathrm{𝟓}`$ and $`\mathrm{𝟓}^{}`$ dimensional representations of Higgs multiplets, $`\widehat{H}_2`$ and $`\widehat{H}_1,`$ respectively. The tenplets $`\psi _i`$ contain the quark doublets, the charged lepton singlets and the up-type quark singlets, while the down-type quark singlets and the lepton doublets are included in the fiveplets $`\varphi _i`$. The Higgs fiveplet $`\widehat{H}_1`$ contains the MSSM Higgs multiplet $`H_1`$ and a coloured Higgs multiplet $`H_{C1}`$, and $`\widehat{H}_2`$ contains the second MSSM Higgs multiplet $`H_2`$ and another coloured Higgs multiplet $`H_{C2}`$. An adjoint representation Higgs multiplet $`\mathrm{\Sigma }`$ causing the breaking of SU(5) should also belong to the Higgs sector of the model. We neglect the Yukawa coupling and the soft SUSY-breaking parameters associated with it. Inclusion of these terms will increase the non-universalities at GUT scale and thus increase the amount of flavour violation in the model. Thus the terms in superpotential $`W`$ relevant for our consideration are given by
$`W`$ $`=`$ $`{\displaystyle \frac{1}{4}}f_{u_{ij}}\psi _i^{AB}\psi _j^{CD}\widehat{H}_2^Eϵ_{ABCDE}+\sqrt{2}f_{d_{ij}}\psi _i^{AB}\varphi _{jA}\widehat{H}_{1B},`$ (1)
where $`A,B,\mathrm{}=1,\mathrm{},5`$ are the SU(5) indices. The relevant soft SUSY breaking terms associated with the SU(5) multiplets are
$`_{\mathrm{SUSY}\mathrm{breaking}}`$ $`=`$ $`(m_{10}^2)_{ij}\stackrel{~}{\psi }_i^{}\stackrel{~}{\psi }_j+(m_5^2)_{ij}\stackrel{~}{\varphi }_i^{}\stackrel{~}{\varphi }_j+m_{h_1}^2h_1^{}h_1+m_{h_2}^2h_2^{}h_2`$ (3)
$`+\{{\displaystyle \frac{1}{4}}A_{u_{ij}}\stackrel{~}{\psi }_i\stackrel{~}{\psi }_jh_2+\sqrt{2}A_{d_{ij}}\stackrel{~}{\psi }_i\stackrel{~}{\varphi }_jh_1+h.c.\},`$
where $`\stackrel{~}{\psi }_i`$ and $`\stackrel{~}{\varphi }_i`$ are the scalar components of the $`\psi _i`$ and $`\varphi _i`$ chiral multiplets, respectively, and $`h_1`$ and $`h_2`$ are the Higgs multiplets.
In the minimal SUGRA scenario the soft SUSY breaking parameters are generated by gravitational interactions and are universal at $`M_P`$:
$`(m_{10}^2)_{ij}=(m_5^2)_{ij}=\delta _{ij}m_0^2,`$ (4)
$`m_{h_1}^2=m_{h_2}^2=m_0^2,`$ (5)
$`A_{u_{ij}}=(A_u^{}f_u)_{ij},A_{d_{ij}}=(A_d^{}f_d)_{ij},`$ (6)
$`A_{u_{ij}}^{}=A_{d_{ij}}^{}=\delta _{ij}A_0.`$ (7)
Below $`M_P`$ till $`M_{GUT}`$ the parameters in Eq.(1) and Eq.(3) evolve with energy according to the renormalization group equations (RGE) of SUSY SU(5). These can be found for example in Ref. and we do not present them here. Because the third generation Yukawa couplings $`f_{u_{33}}`$ and $`f_{d_{33}}`$ are much larger than the Yukawa couplings of the first two generations, large hierarchies in the parameters of Eq.(7) are induced at $`M_{GUT}.`$ Therefore the soft mass terms $`m_{10}^2,`$ $`m_5^2`$ and $`A_{u,d}^{}`$ remain diagonal in the generation space but the third generation masses are smaller than the masses of the first two generations, which remain degenerate to a good approximation.
To understand the origin of flavour violation in SUSY SU(5) GUT we have to discuss the evolution of the Yukawa coupling matrices $`f_d`$ and $`f_u`$ with energy. At $`M_P`$ we can choose the basis in which $`f_u`$ is diagonal, $`f_u^P=(f_u^P)_{ii}\delta _{ij}.`$ In that case the Yukawa matrix $`f_d`$ can be diagonalized with a bi-unitary transformation,
$`V^P\mathrm{\Theta }^Pf_d^PU^P,`$ (8)
where $`V^P`$ corresponds to the Cabibbo-Kobayashi-Maskawa (CKM) matrix at Planck scale and $`\mathrm{\Theta }^P=\text{diag}(e^{i\varphi _1},e^{i\varphi _2},e^{i\varphi _3}),`$ where $`\varphi _{1,2,3}`$ are the GUT phases (satisfying $`\varphi _1+\varphi _2+\varphi _3=0`$) present in addition to the phase in $`V^P.`$ However, the mixing matrix $`U^P`$ can be rotated away by a redefinition of the fiveplet fields $`\varphi .`$ No trace of that rotation will remain in the soft SUSY breaking terms because these are universal and proportional to unit matrix at $`M_P.`$ Below $`M_P`$ the Yukawa matrices $`f_d`$ and $`f_u`$ run according to
$`16\pi ^2{\displaystyle \frac{\text{d}}{\text{d}t}}f_{u_{ij}}`$ $`=`$ $`\left[{\displaystyle \frac{96}{5}}g_5^2+3\mathrm{T}\mathrm{r}(f_u^{}f_u)\right]f_{u_{ij}}+6(f_uf_u^{}f_u)_{ij}+2(f_df_d^{}f_u)_{ij}+2(f_uf_d^{}f_d^{})_{ij},`$ (9)
$`16\pi ^2{\displaystyle \frac{\text{d}}{\text{d}t}}f_{d_{ij}}`$ $`=`$ $`\left[{\displaystyle \frac{84}{5}}g_5^2+4\mathrm{T}\mathrm{r}(f_d^{}f_d)\right]f_{d_{ij}}+6(f_df_d^{}f_d)_{ij}+3(f_uf_u^{}f_d)_{ij}.`$ (10)
Because the third generation Yukawa couplings are large, of order unity, the running of the off-diagonal elements of the Yukawa matrices is significant. Notice that despite of the chosen basis, non-zero off-diagonal elements are generated also in $`f_u.`$ To predict the running of the off-diagonal elements requires a precise knowledge of all the elements of the Yukawa matrices. This, however, goes beyond the assumptions made in our work: we assumed that only the third generation Yukawa couplings can be reliably estimated via the RGEs from the experimental data. Therefore, in our numerical estimates we assume that the off-diagonal elements of the rotation matrix $`U`$, which rotates the fiveplet fields $`\varphi `$ are small, of order or smaller than the corresponding CKM matrix elements at GUT scale.
At GUT scale the Yukawa coupling matrices $`f_d`$ and $`f_u`$ get renormalized to $`f_d^G`$ and $`f_u^G,`$ respectively, and can be diagonalized again with bi-unitary transformations. However, now any rotation of the superfields $`\psi `$ and $`\varphi `$ is reflected in the soft SUSY breaking parameters because these are not universal at GUT scale any more. These rotations give rise to large flavour changing effects, as will be discussed in the following.
At $`M_{GUT}`$ the SU(5) gauge group breaks spontaneously into the usual MSSM gauge group $`\text{SU(3)}_C\times \text{SU(2)}_L\times \text{U(1)}_Y.`$ The MSSM superpotential $`W`$ valid below GUT scale is
$`W`$ $`=`$ $`Q_i^o(f_{u_{ij}})U_i^{co}H_2Q_i^o(f_{d_{ij}})D_j^{co}H_1E_i^{co}(f_{e_{ij}})L_j^oH_1\mu H_1H_2`$ (11)
and the soft SUSY breaking terms are given by
$`_{\mathrm{soft}}`$ $`=`$ $`\stackrel{~}{L}_i^o(m_{\stackrel{~}{L}}^2)_{ij}\stackrel{~}{L}_j^o+\stackrel{~}{E}_i^{co}(m_{\stackrel{~}{E}}^2)_{ij}\stackrel{~}{E}_j^{co}+\stackrel{~}{Q}_i^o(m_{\stackrel{~}{Q}}^2)_{ij}\stackrel{~}{Q}_j^o+\stackrel{~}{U}_i^{co}(m_{\stackrel{~}{U}}^2)_{ij}\stackrel{~}{U}_j^{co}+\stackrel{~}{D}_i^{co}(m_{\stackrel{~}{D}}^2)_{ij}\stackrel{~}{D}_j^{co}`$ (14)
$`+m_{H_1}^2H_1^{}H_1+m_{H_2}^2H_2^{}H_2+(\stackrel{~}{Q}_i^o(A_{u_{ij}})\stackrel{~}{U}_j^{co}H_2\stackrel{~}{Q}_i^o(A_{d_{ij}})\stackrel{~}{D}_j^{co}H_1\stackrel{~}{E}_i^{co}(A_{e_{ij}})\stackrel{~}{L}_j^oH_1`$
$`+B_HH_1H_2+{\displaystyle \frac{1}{2}}M_1\stackrel{~}{B}\stackrel{~}{B}+{\displaystyle \frac{1}{2}}M_2\stackrel{~}{W}^a\stackrel{~}{W}^a+{\displaystyle \frac{1}{2}}M_3\stackrel{~}{g}^a\stackrel{~}{g}^a+h.c.).`$
Here all the fields are explicitly written in the flavour eigenstate basis denoted by the superscript zero. The boundary condition for the soft terms are specified via
$`(m_{\stackrel{~}{Q}}^2)_{ij}=(m_{\stackrel{~}{U}}^2)_{ij}=(m_{\stackrel{~}{E}}^2)_{ij}=(m_{10}^2)_{ij},`$ (15)
$`(m_{\stackrel{~}{D}}^2)_{ij}=(m_{\stackrel{~}{L}}^2)_{ij}=(m_5^2)_{ij},`$ (16)
$`m_{H_1}^2=m_{h_1}^2,m_{H_2}^2=m_{h_2}^2,`$ (17)
$`M_1=M_2=M_3=M_0.`$ (18)
Since there is no splitting among the gaugino masses above the GUT scale, it is most convenient to stipulate $`M_0`$ at the GUT scale. For the Yukawa couplings and trilinear terms this model predicts
$`f_e^G=f_d^G,A_e^G=A_d^G`$ (19)
at GUT scale.
From the GUT scale down to the electroweak scale the Yukawa couplings and the soft parameters evolve with energy via the MSSM RGEs . Once the electroweak symmetry is broken, rotations of the superfields
$`\begin{array}{ccc}D^o=V_dD,& E^o=V_eE,& U^o=V_uU,\\ D^{co}=U_d^{}D^c,& E^{co}=U_e^{}E^c,& U^{co}=U_u^{}U^c,\end{array}`$ (22)
bring quarks and leptons into their mass eigenstates with diagonal Yukawa couplings $`f_d^{}`$, $`f_e^{}`$ and $`f_u^{}`$
$`f_{d_i}^{}=(V_d^Tf_dU_d^{})_{ii},f_{e_i}^{}=(U_e^{}f_eV_e)_{ii},f_{u_i}^{}=(V_u^Tf_uU_u^{})_{ii}.`$ (23)
At the GUT scale the down quark and lepton masses are predicted to be equal. Therefore the diagonalizing matrices are related at $`M_{GUT}`$ as
$`V_d^G=U_e^G,U_d^G=V_e^G.`$ (24)
This implies that at low energies the left-handed quark rotation matrix, the CKM matrix in the basis in which up quark Yukawa matrix is diagonal, can be related to the right-handed lepton rotation matrix and the right-handed down quark rotation matrix can be related to the left-handed lepton rotations via the RGEs.
At the same time the rotation in Eq.(22) changes the basis of the superpartners of quarks and leptons. For example, the mass eigenstates of the charged sleptons and sneutrinos can be expressed as
$`\stackrel{~}{E}=\mathrm{\Gamma }_E\left(\begin{array}{c}V_e^{}\stackrel{~}{E}^o\\ U_e^{}\stackrel{~}{E}^{co}\end{array}\right),\stackrel{~}{\nu }=\mathrm{\Gamma }_\nu V_e^{}\stackrel{~}{\nu }^o,`$ (27)
where $`\mathrm{\Gamma }_E`$ and $`\mathrm{\Gamma }_\nu `$ are $`6\times 6`$ and $`3\times 3`$ rotation matrices, respectively. The slepton mass matrices are given by
$`m_{\stackrel{~}{E}}^2`$ $`=`$ $`\mathrm{\Gamma }_E\left(\begin{array}{cc}V_e^{}m_{\stackrel{~}{L}}^2V_e+m_e^2m_Z^2\mathrm{cos}2\beta (\frac{1}{2}\mathrm{sin}^2\theta _W)& \mu m_e\mathrm{tan}\beta +m_eU_e^{}A_e^{{}_{}{}^{}}U_e\\ & \\ \mu ^{}m_e\mathrm{tan}\beta +U_e^{}A_e^{}U_em_e& U_e^{}m_{\stackrel{~}{E}}^2U_e+m_e^2m_Z^2\mathrm{cos}2\beta \mathrm{sin}^2\theta _W\end{array}\right)\mathrm{\Gamma }_E^{}`$ (31)
$`m_{\stackrel{~}{\nu }}^2`$ $`=`$ $`\mathrm{\Gamma }_\nu \left(V_e^{}m_{\stackrel{~}{L}}^2V_e+{\displaystyle \frac{1}{2}}m_Z^2\mathrm{cos}2\beta \right)\mathrm{\Gamma }_\nu ^{},`$ (33)
and analogously for squarks.
Notice that this definition of the diagonalizing matrices $`\mathrm{\Gamma }_{E,D,U}`$ differs from the one originally given in Ref. where the matrices $`\mathrm{\Gamma }`$ relate the flavour eigenstates directly to the mass eigenstates. The advantage of the present notation in calculating the flavour violating observables in our scenario is the following. It allows us to perform the MSSM RGE running of the soft terms (which are diagonal in the flavour space) and the CKM matrix elements (which induce the flavour mixings) separately without constructing the squark mass matrices at high scales and without running each element of the $`6\times 6`$ sparticle mass matrices. The slepton and squark mass matrices are then constructed at low energies in terms of the low energy values of the soft terms, quark masses and the mixing matrices.
We use the standard notation for the neutralino and chargino mass matrices. The neutralino mass matrix $`M_{\stackrel{~}{N}}`$ in the basis $`(\stackrel{~}{B},\stackrel{~}{W}^0,\stackrel{~}{H}_1^0,\stackrel{~}{H}_2^0)`$ can be diagonalized as
$`M_{\stackrel{~}{N}}^{diag}=NM_{\stackrel{~}{N}}N^{}`$ (34)
where $`N_{ij}`$ is a $`4\times 4`$ unitary matrix. Similarly, the two diagonalizing matrices $`O_{L,R}`$ in the chargino sector can be found from
$`M_{\stackrel{~}{\chi }}^{diag}=O_L\left(\begin{array}{cc}M_2& \sqrt{2}\mathrm{sin}\beta m_W\\ \sqrt{2}\mathrm{cos}\beta m_W& \mu \end{array}\right)O_R^{}`$ (37)
For the further details see, e.g. , Ref. .
## III Renormalization Procedure
Before calculating the rates of flavour violating observables, let us describe our procedure of calculating the input SUSY parameters via the RGE evaluation. We start the RGE running at $`M_Z`$ where we introduce as the low energy input the values of the gauge coupling constants, the tau and bottom quark Yukawa couplings corresponding to the tau and bottom quark masses $`m_\tau (M_Z)=1.784`$ GeV and $`m_b(M_Z)=3.0`$ GeV , respectively. We parametrize the CKM matrix in the standard way with $`\theta _{12}=0.22,`$ $`\theta _{23}=0.04`$ and $`\theta _{13}=0.003.`$ We evolve these quantities from $`M_Z`$ to $`m_t`$ using two loop SM RGE-s for five flavours. At $`m_t`$ we include top quark Yukawa coupling corresponding to the pole mass $`m_t=174`$ GeV and run the SM RGE-s for six flavours up to the scale $`Q`$ where superparticles are introduced. At the scale $`Q`$ we convert the SM couplings to the MSSM ones fixing the value of $`\mathrm{tan}\beta `$ and include the SUSY loop corrections to the bottom quark and tau lepton Yukawa couplings according to Ref. . Then the gauge and Yukawa couplings are evaluated at GUT scale with the help of two loop MSSM RGE-s. For the running of the CKM matrix elements we use the one loop RGE-s from Ref. .
The crucial issue in the present context is the correct tau-bottom Yukawa unification as predicted by SU(5). The Yukawa unification has been studied extensively in literature and it turns out that unification is possible only for $`\mu <0`$ and 30
<
tanβ
<
50.
<
30𝛽
<
5030\mathrel{\vbox{\hbox{$<$}\nointerlineskip\hbox{$\sim$}}}\tan\beta\mathrel{\vbox{\hbox{$<$}\nointerlineskip\hbox{$\sim$}}}50. This can be understood as follows. The correction to the bottom Yukawa coupling ,
$`f_b{\displaystyle \frac{m_b}{1+\delta _b}}{\displaystyle \frac{1}{v_1}}`$ (38)
which is numerically the most important one, is proportional to $`\mathrm{tan}\beta `$ and its sign depends on $`sign(\mu ).`$ Tau-bottom Yukawa unification can be obtained if the top quark mass is approximately at a fixed point at $`\mathrm{tan}\beta 1.6.`$ On the other hand, small values of $`\mathrm{tan}\beta `$ are, anyway, excluded by the LEP 2 search for the lightest Higgs boson . To achieve the tau-bottom Yukawa unification therefore a sizable negative $`\delta _b`$ is needed, which requires large values of $`\mathrm{tan}\beta `$ and fixes the sign of the $`\mu `$ parameter to be $`sign(\mu )=1.`$ On the other hand, it has been argued that there might be important threshold corrections due to the coloured triplet Higgses at GUT scale modifying sizably the bottom quark Yukawa coupling already at $`M_{GUT}.`$ In any case, the coupling which receives large corrections is the bottom Yukawa coupling, tau Yukawa coupling is less sensitive to these corrections and thus the prediction of tau Yukawa coupling at GUT scale is more robust. Therefore, in our numerical calculations we use actually the matching condition
$`f_d^G[\text{SU(5)}]=f_\tau ^G[\text{MSSM}]`$ (39)
at $`M_{GUT}`$ to give a numerical value to the SU(5) Yukawa coupling $`f_{d_{33}},`$ and allow the value of the MSSM coupling $`f_b^G`$ to be numerically different. To achieve semi-realistic Yukawa coupling evolution we fix the magnitude of $`\delta _b`$ by hand for each considered value of $`\mathrm{tan}\beta `$ but keep its sign to be determined by the $`sign(\mu ).`$ In such a case the Yukawa unification is achieved for $`sign(\mu )=1`$ and not achieved for $`sign(\mu )=+1.`$ In our numerical examples we consider both possibilities. For definiteness we consider two values of $`\mathrm{tan}\beta ,`$ $`\mathrm{tan}\beta =35`$ (corresponds to $`f_t^G2f_\tau ^G`$) and $`\mathrm{tan}\beta =48`$ (corresponds to $`f_t^Gf_\tau ^G`$).
At $`M_{GUT}`$ we fix the leptonic mixing matrices $`V_e`$ and $`U_e`$ according to Eq.(24). Unlike the quark mixing matrices $`V_e`$ and $`U_e`$ do not run with energy due to the absence of neutrino Yukawa couplings in the MSSM. Choosing the basis in which the top Yukawa matrix is diagonal we have $`V_d=V_{CKM}`$ and therefore $`|U_e|_{ij}=|V_{CKM}^G|_{ij}.`$ There is no experimental restriction on the mixing matrix of the right-handed quark fields. Therefore we assume that large couplings $`f_{d_{33}},f_{u_{33}}`$ in Eq.(10) generate small non-zero $`(31)`$ and $`(32)`$ elements for $`U_d`$ and $`V_e`$ but the angle $`\theta _{12}`$ remains negligible in these matrices. Numerically we consider three cases: $`V_e^{(31),(32)}=U_e^{(31),(32)},`$ $`V_e^{(31),(32)}=0.1\times U_e^{(31),(32)}`$ and $`V_e^{(31),(32)}=0.`$
Further, we run the SU(5) gauge and Yukawa couplings from $`M_{GUT}`$ to $`M_P`$ where we fix the values of $`m_0`$ and $`A_0.`$ For the trilinear coupling we take always $`A_0=0`$ in our numerical calculations. Then we run all the parameters, including the SU(5) soft terms down to the GUT scale, apply the boundary conditions Eq.(18) and run all the MSSM parameters down to the scale $`Q`$ which we take to be 200 GeV. At that scale we fix the SUSY parameters $`\mu `$ and $`B`$ via the conditions of spontaneous symmetry breaking and calculate the masses and mixing matrices of the sparticles. The chargino and neutralino mass matrices are given by the standard expressions which can be found for example in Ref. . In order not to run into contradiction with the mass bounds on the SUSY particles from direct searches we take M0
>
150
>
subscript𝑀0150M_{0}\mathrel{\vbox{\hbox{$>$}\nointerlineskip\hbox{$\sim$}}}150 GeV and m0
>
150
>
subscript𝑚0150m_{0}\mathrel{\vbox{\hbox{$>$}\nointerlineskip\hbox{$\sim$}}}150 GeV; the lightest sparticle masses corresponding to $`m_0=M_0=150`$ GeV and $`\mathrm{tan}\beta =48`$ are roughly $`M_{\stackrel{~}{\tau }_1}=73`$ GeV, $`M_{\stackrel{~}{\chi }_1^+}=98`$ GeV and $`M_{\stackrel{~}{\chi }_1^0}=58`$ GeV and are just on the limit of the present LEP bounds.
## IV Rates of Flavour Changing Processes
### A $`K_LK_S`$ System
We start our numerical analyses by calculating the new SUSY flavour changing physics contribution to physical observables in the $`K_LK_S`$ system. Recently it has been emphasized that in models with massive neutrinos the SUSY flavour changing contribution to $`ϵ_K`$ may be large, especially if new GUT phases are present which may maximize the effect. Because at large $`\mathrm{tan}\beta `$ the pattern of flavour violation in the SUSY SU(5) is the same as in the models with right-handed neutrinos we study this issue carefully.
The dominant new physics contribution to $`\mathrm{\Delta }S=2`$ processes comes from box diagrams with gluinos and down squarks running in the loop . Because in our scenario the first two generation sfermions are almost exactly degenerate the appropriate tool to use is the mass insertion approximation . The current state of the art on this subject is summarized in Ref. in which the low energy $`\mathrm{\Delta }S=2`$ effective Hamiltonian is calculated including NLO QCD corrections and the relevant hadronic matrix elements are evaluated using the lattice results for the $`B_K`$-parameters. We use the expressions for the $`\mathrm{\Delta }S=2`$ Wilson coefficients, NLO QCD corrections and hadronic matrix elements as well as numerical input exactly as in Ref. and therefore we do not copy them here. However, we need to specify the model dependent input.
In the mass insertion approximation the flavour violation is characterized with the dimensionless parameters $`(\delta _{LL})_{ij},`$ $`(\delta _{RR})_{ij},`$ $`(\delta _{LR})_{ij}`$ and $`(\delta _{RL})_{ij}`$ defined as
$`\left(\begin{array}{cc}\delta _{LL}& \delta _{LR}\\ \delta _{RL}& \delta _{RR}\end{array}\right)={\displaystyle \frac{1}{\stackrel{~}{m}^2}}\left(\begin{array}{cc}(m_{\stackrel{~}{D}}^2)_{LL}& (m_{\stackrel{~}{D}}^2)_{LR}\\ (m_{\stackrel{~}{D}}^2)_{RL}& (m_{\stackrel{~}{D}}^2)_{RR}\end{array}\right)`$ (44)
where $`(m_{\stackrel{~}{D}}^2)_{MN},`$ $`M,N=L,R`$ are the corresponding $`3\times 3`$ components of the $`6\times 6`$ down squark mass matrix and $`\stackrel{~}{m}^2`$ is an average squark mass appropriate for the problem under consideration. We calculate the down squark mass matrix as described in detail in previous sections. The mixings among the left down squark fields are generated by the CKM matrix $`V_{CKM}.`$ The mixings among the right down squarks are assumed to be given by the matrix $`U_{d_{ij}}=(V_{CKM}^G)_{ij},`$ $`i,j=2,3`$ with vanishing angle $`\theta _{12}.`$ Thus the off-diagonal elements of $`U_d`$ are smaller than the elements of $`V_{CKM}.`$ Because only the $`(12)`$ components of $`\delta _{MN}`$ enter to the $`\mathrm{\Delta }S=2`$ process we identify $`\stackrel{~}{m}^2=(m_{\stackrel{~}{Q}}^2)_{11}.`$
The $`K_LK_S`$ mass difference and the CP-violating parameter $`ϵ_K`$ are given by:
$`\mathrm{\Delta }M_K`$ $`=`$ $`2\text{Re}K^0|H_{eff}^{\mathrm{\Delta }S=2}|\overline{K}^0,`$ (45)
$`ϵ_K`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}\mathrm{\Delta }M_K}}\text{Im}K^0|H_{eff}^{\mathrm{\Delta }S=2}|\overline{K}^0.`$ (46)
It turns out that the new physics contribution to $`\mathrm{\Delta }M_K`$ is very small in our model, typically at the permil level of the measured value. However, the new contribution to $`ϵ_K`$ which is a small quantity in the SM can be sizable. In Fig. 1 we plot the gluino mediated contribution to $`ϵ_K`$ as a function of the average squark mass $`\stackrel{~}{m}`$ fixing the gluino mass to be $`M_{\stackrel{~}{g}}=420`$ GeV and assuming that the new GUT phases maximize the effect. For the squark masses of order 500 GeV and large $`\mathrm{tan}\beta `$ the contribution may exceed 25% of the measured value
$`ϵ_K=2.310^3.`$ However, because of the assumption of the GUT phases and because of large theoretical errors in the hadronic matrix elements no useful constraints can be derived from $`ϵ_K`$ on the parameters of the model. The reason why the contribution to $`ϵ_K`$ in SUSY SU(5) is smaller than in the models with massive neutrinos is the smallness of the $`(\delta _{MN})_{12}`$ elements. The magnitude of the largest mixing parameter $`(\delta _{LL})_{12}`$ is typically below $`10^4`$ and for the chosen parameters $`(\delta _{RR})_{12}`$ is about factor of three smaller. The (LR) parameters are smaller by three orders of magnitude. We conclude that the small mixing angles in our approach suppress the contribution to the $`K\overline{K}`$ mixing.
### B $`B_d\overline{B}_d`$ System
Here we estimate the new physics contribution to the $`B_d`$ meson system. The SM contribution to $`\mathrm{\Delta }M_{B_d}=2|M_{12}|`$ is given by
$$M_{12}^{SM}=\frac{G_F^2}{12\pi ^2}\eta _{QCD}B_{B_d}f_{B_d}^2M_{B_d}M_W^2(V_{td}V_{tb}^{})^2S_0(z_t),$$
(47)
where
$$S_0(z_t)=\frac{4z_t11z_t^2+z_t^3}{4(1z_t)^2}\frac{3z_t^3\mathrm{ln}z_t}{2(1z_t)^3},$$
(48)
where $`z_t=m_t^2/m_W^2`$ and $`\eta _{QCD}=0.55\pm 0.01`$. With $`B_{B_d}=1.29\pm 0.08\pm 0.06`$ and $`f_{B_d}=175\pm 25`$ MeV it successfully predicts the measured value $`\mathrm{\Delta }M_{B_d}=0.470\pm 0.019`$ ps$`^1.`$ To estimate the magnitude of the dominant gluino induced contribution to $`B\overline{B}`$ mixing in our scenario we use again the mass insertion approximation. Neglecting the small LR and RL contributions it reads
$`M_{12}^{SUSY}={\displaystyle \frac{\alpha _s^2}{216\stackrel{~}{m}^2}}{\displaystyle \frac{1}{3}}B_{B_d}f_{B_d}^2M_{B_d}\{((\delta _{LL})_{31}^2+(\delta _{RR})_{31}^2)(66\stackrel{~}{f}_6(x)+24xf_6(x))+`$ (53)
$`(\delta _{LL})_{31}(\delta _{RR})_{31}[(3624\left({\displaystyle \frac{M_{B_d}}{m_b+m_d}}\right)^2)\stackrel{~}{f}_6(x)+(72+384\left({\displaystyle \frac{M_{B_d}}{m_b+m_d}}\right)^2)xf_6(x)]\},`$
$`f_6(x)={\displaystyle \frac{1}{6(x1)^5}}\left(x^39x^29x+17+6(1+3x)\mathrm{ln}x\right),`$
$`\stackrel{~}{f}_6(x)={\displaystyle \frac{1}{3(x1)^5}}\left(x^39x^2+9x+1+6x(1+x)\mathrm{ln}x\right).`$
where $`x=M_{\stackrel{~}{g}}^2/\stackrel{~}{m}^2.`$ In this case the common squark mass is taken to be $`\stackrel{~}{m}^2=(m_{\stackrel{~}{Q}}^2)_{33}`$ and the mixing elements $`\delta _{31}`$ are calculated as in the previous subsection.
In the case where the sparticle masses are assumed to take their lower allowed limits, the SUSY contribution to $`\mathrm{\Delta }M_{B_d}`$ turns out to be at most of order a few percent. Taking into account the large errors in the input parameter $`f_{B_d}^2B_{B_d}`$ no useful constraints on our model can be derived drom the measurement of $`\mathrm{\Delta }M_{B_d}`$.
### C $`bs\gamma `$
The radiative decay $`bs\gamma `$ is known to get large contributions from SUSY particle loops and therefore this process implies strong constraints on the allowed SUSY parameter space . Besides the SM $`W`$-boson-$`t`$-quark contribution there are also charged Higgs, chargino, neutralino and gluino contributions. At the electroweak scale all of them have been calculated in Ref. . While in the SM the NLO QCD corrections are included to $`bs\gamma `$, in SUSY theories the NLO analyses has been performed only for specific scenarios . The complication here lies in the new flavour structure of SUSY theories, as compared to the SM, which extends the operator basis beyond the SM one. Therefore the LO QCD corrections to the gluino contribution which gives the dominant new flavour physics contribution was calculated very recently in Ref. . Before that, the gluino contribution in a generic SUSY flavour model was studied in without including QCD corrections.
Here we shortly review the results of Ref. . The effective Hamiltonian for $`bs\gamma `$ is expressed in two parts
$`_{eff}=_{eff}^{CKM}+_{eff}^{\stackrel{~}{g}},`$ (54)
where $`_{eff}^{CKM}`$ is the effective Hamiltonian in which the structure of flavour violation is the same as in the SM and the gluino contribution $`_{eff}^{\stackrel{~}{g}}`$ exhibits the new flavour structure. The Wilson coefficients of the first term,
$$_{eff}^{CKM}=\frac{4G_F}{\sqrt{2}}V_{tb}^{}V_{ts}^{}\underset{i}{}C_i(\mu )𝒪_i(\mu ),$$
(55)
contain the SM as well as the charged Higgs, chargino and neutralino contributions. The dominant operators in Eq.(55) are the dimension six magnetic operators ($`\overline{m}_b(\mu )`$ is the running mass)
$`𝒪_7={\displaystyle \frac{e}{16\pi ^2}}\overline{m}_b(\mu )(\overline{s}\sigma ^{\mu \nu }P_Rb)F_{\mu \nu },𝒪_8={\displaystyle \frac{g_s}{16\pi ^2}}\overline{m}_b(\mu )(\overline{s}\sigma ^{\mu \nu }T^aP_Rb)G_{\mu \nu }^a,`$ (56)
and the operators $`𝒪_{7,8}^{}`$ obtained by $`LR.`$ The gluino effective hamiltonian
$$_{eff}^{\stackrel{~}{g}}=\underset{i}{}C_{i,\stackrel{~}{g}}(\mu )𝒪_{i,\stackrel{~}{g}}(\mu )+\underset{i}{}\underset{q}{}C_{i,\stackrel{~}{g}}^q(\mu )𝒪_{i,\stackrel{~}{g}}^q(\mu ).$$
(57)
contains in addition to the dimension six magnetic operators $`𝒪_{7b,\stackrel{~}{g}},`$ $`𝒪_{7b,\stackrel{~}{g}}^{},`$ $`𝒪_{8b,\stackrel{~}{g}},`$ $`𝒪_{8b,\stackrel{~}{g}}^{}`$ also operators of dimension five,
$`𝒪_{7\stackrel{~}{g},\stackrel{~}{g}}=eg_s^2(\mu )(\overline{s}\sigma ^{\mu \nu }P_Rb)F_{\mu \nu },𝒪_{8\stackrel{~}{g},\stackrel{~}{g}}=g_s(\mu )g_s^2(\mu )(\overline{s}\sigma ^{\mu \nu }T^aP_Rb)G_{\mu \nu }^a,`$ (58)
in which the chirality-violating parameter is the gluino mass in the corresponding Wilson coefficients, as well as new four-quark operators $`𝒪_{i,\stackrel{~}{g}}^q,`$ $`q=u,d,c,s,b,`$ which are listed in Ref. . These new operators change the structure of the LO QCD corrections in general SUSY flavour models.
The Wilson coefficients in Eq.(55) including the LO QCD corrections in the MSSM are well known. Their explicit expressions can be found in and we do not rewrite them here. Instead, let us for a moment concentrate on the study of the effective hamiltonian Eq.(57) in our model.
At the electroweak scale the relevant Wilson coefficients are given by
$`C_{7b,\stackrel{~}{g}}(\mu _W)`$ $`=`$ $`{\displaystyle \frac{e_d}{16\pi ^2}}C(R){\displaystyle \underset{k=1}{\overset{6}{}}}{\displaystyle \frac{1}{m_{\stackrel{~}{d}_k}^2}}\left(\mathrm{\Gamma }_{DL}^{kb}\mathrm{\Gamma }_{DL}^{ks}\right)f_2(x_{gd_k}),`$ (59)
$`C_{7\stackrel{~}{g},\stackrel{~}{g}}(\mu _W)`$ $`=`$ $`M_{\stackrel{~}{g}}{\displaystyle \frac{e_d}{16\pi ^2}}C(R){\displaystyle \underset{k=1}{\overset{6}{}}}{\displaystyle \frac{1}{m_{\stackrel{~}{d}_k}^2}}\left(\mathrm{\Gamma }_{DR}^{kb}\mathrm{\Gamma }_{DL}^{ks}\right)f_4(x_{gd_k}),`$ (60)
$`C_{8b,\stackrel{~}{g}}(\mu _W)`$ $`=`$ $`{\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{k=1}{\overset{6}{}}}{\displaystyle \frac{1}{m_{\stackrel{~}{d}_k}^2}}\left(\mathrm{\Gamma }_{DL}^{kb}\mathrm{\Gamma }_{DL}^{ks}\right)\left[\left(C(R){\scriptscriptstyle \frac{1}{2}}C(G)\right)f_2(x_{gd_k}){\scriptscriptstyle \frac{1}{2}}C(G)f_1(x_{gd_k})\right],`$ (61)
$`C_{8\stackrel{~}{g},\stackrel{~}{g}}(\mu _W)`$ $`=`$ $`M_{\stackrel{~}{g}}{\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \underset{k=1}{\overset{6}{}}}{\displaystyle \frac{1}{m_{\stackrel{~}{d}_k}^2}}\left(\mathrm{\Gamma }_{DR}^{kb}\mathrm{\Gamma }_{DL}^{ks}\right)\left[\left(C(R){\scriptscriptstyle \frac{1}{2}}C(G)\right)f_4(x_{gd_k}){\scriptscriptstyle \frac{1}{2}}C(G)f_3(x_{gd_k})\right],`$ (62)
Here the matrices $`\mathrm{\Gamma }_{DL}`$ and $`\mathrm{\Gamma }_{DR}`$ are the $`6\times 3`$ submatrices of $`\mathrm{\Gamma }_D,`$
$`\mathrm{\Gamma }_D^{6\times 6}=\left(\begin{array}{cc}\mathrm{\Gamma }_{DL}^{6\times 3}& \mathrm{\Gamma }_{DR}^{6\times 3},\end{array}\right)`$ (64)
and the ratios $`x_{gd_k}`$ are defined as $`x_{gd_k}M_{\stackrel{~}{g}}^2/m_{\stackrel{~}{d}_k}^2.`$ The Casimir factors $`C(R)`$ and $`C(G)`$ are $`C(R)=4/3`$ and $`C(G)=3`$ and the functions $`f_i(x)`$, $`i=1,\mathrm{},4`$, are given by
$`f_1(x)`$ $`=`$ $`{\displaystyle \frac{1}{12(x1)^4}}\left(x^36x^2+3x+2+6x\mathrm{log}x\right),`$ (65)
$`f_2(x)`$ $`=`$ $`{\displaystyle \frac{1}{12(x1)^4}}\left(2x^3+3x^26x+16x^2\mathrm{log}x\right),`$ (66)
$`f_3(x)`$ $`=`$ $`{\displaystyle \frac{1}{2(x1)^3}}\left(x^24x+3+2\mathrm{log}x\right),`$ (67)
$`f_4(x)`$ $`=`$ $`{\displaystyle \frac{1}{2(x1)^3}}\left(x^212x\mathrm{log}x\right).`$ (68)
The Wilson coefficients of the corresponding primed operators are obtained through the interchange $`\mathrm{\Gamma }_{DR}^{ij}\mathrm{\Gamma }_{DL}^{ij}.`$ At the low scale $`\mu _b`$, the LO renormalization group improved coefficients become
$`C_{7\stackrel{~}{g},\stackrel{~}{g}}(\mu _b)`$ $`=`$ $`\eta ^{\frac{27}{23}}C_{7\stackrel{~}{g},\stackrel{~}{g}}(\mu _W)+{\displaystyle \frac{8}{3}}\left(\eta ^{\frac{25}{23}}\eta ^{\frac{27}{23}}\right)C_{8\stackrel{~}{g},\stackrel{~}{g}}(\mu _W),`$ (69)
$`C_{7b,\stackrel{~}{g}}(\mu _b)`$ $`=`$ $`\eta ^{\frac{39}{23}}C_{7b,\stackrel{~}{g}}(\mu _W)+{\displaystyle \frac{8}{3}}\left(\eta ^{\frac{37}{23}}\eta ^{\frac{39}{23}}\right)C_{8b,\stackrel{~}{g}}(\mu _W)+R_{7b,\stackrel{~}{g}}(\mu _b),`$ (70)
where $`\eta =\alpha _s(\mu _W)/\alpha _s(\mu _b)`$. The remainder function $`R_{7b,\stackrel{~}{g}}(\mu _b)`$ turns out to be numerically very small and we neglect it in our numerical computation. The low-scale Wilson coefficients for the corresponding primed operators are obtained by replacing in (70) all the unprimed coefficients with primed ones.
The decay width of $`bs\gamma `$ can be written
$$\mathrm{\Gamma }(bs\gamma )=\frac{m_b^5G_F^2|V_{tb}V_{ts}^{}|^2\alpha }{32\pi ^4}\left|C_7^{eff}\right|^2,$$
(71)
where in our model
$`\left|C_7^{eff}\right|^2=\left|C_7+C_7^{\stackrel{~}{g}}\right|^2+\left|C_7^{{}_{}{}^{}\stackrel{~}{g}}\right|^2.`$ (72)
Here
$`C_7^{\stackrel{~}{g}}`$ $`=`$ $`{\displaystyle \frac{16\sqrt{2}\pi ^3\alpha _s(\mu _b)}{G_FV_{tb}V_{ts}^{}}}\left[C_{7b,\stackrel{~}{g}}(\mu _b)+{\displaystyle \frac{1}{m_b}}C_{7\stackrel{~}{g},\stackrel{~}{g}}(\mu _b)\right],`$ (73)
$`C_7^{{}_{}{}^{}\stackrel{~}{g}}`$ $`=`$ $`{\displaystyle \frac{16\sqrt{2}\pi ^3\alpha _s(\mu _b)}{G_FV_{tb}V_{ts}^{}}}\left[C_{7b,\stackrel{~}{g}}^{}(\mu _b)+{\displaystyle \frac{1}{m_b}}C_{7\stackrel{~}{g},\stackrel{~}{g}}^{}(\mu _b)\right],`$ (74)
and $`C_7`$ stands for the total contribution from the effective Hamiltonian Eq.(55). In the SM its value is $`C_7^{SM}=0.3`$ . In the MSSM $`C_7`$ receives large contribution from charged Higgs loops which add constructively with $`C_7^{SM}`$ and at large $`\mathrm{tan}\beta `$ also from chargino loops which add constructively (destructively) with $`C_7^{SM}`$ for $`sign(\mu )=1`$ ($`sign(\mu )=+1`$). Therefore for large $`\mathrm{tan}\beta `$ and small Higgs and sparticle masses the $`C_7`$ may be completely dominated by SUSY. Therefore the present experimental result
$$2\times 10^4<\mathrm{BR}(\overline{B}X_s\gamma )<4.5\times 10^4,$$
(75)
which favours the following allowed range:
$`0.25<|C_7^{eff}|<0.375,`$ (76)
implies strong constraints on the SUSY masses.
Let us now study the gluino contribution to $`C_7^{eff}`$ numerically. In Fig. 2 we plot the values of $`C_7^{\stackrel{~}{g}}`$ as a function of the lightest bottom squark mass $`m_{\stackrel{~}{b}_1}`$ for a fixed value of the gluino mass $`M_{\stackrel{~}{g}}=420`$ GeV and for two values of $`\mathrm{tan}\beta `$ and $`sign(\mu ),`$ as indicated in the figure. Note that this is the smallest gluino mass consistent with chargino mass bounds in our model. The values of $`C_7^{\stackrel{~}{g}}`$ might be sizable for small sparticle masses but decreases rapidly if the masses are higher. The important behaviour to notice is that the sign of $`C_7^{\stackrel{~}{g}}`$ depends on the $`sign(\mu )`$ exactly the same way as the dominant contribution $`C_7.`$ This implies that $`C_7`$ and $`C_7^{\stackrel{~}{g}}`$ add up constructively and no cancellation between them is possible. It has been argued in Ref. that the constraints on the SUSY parameter space coming from $`C_7`$ which are rather restrictive can be relaxed by new flavour contributions. However, in our model this does not happen and the $`bs\gamma `$ bounds cannot be relaxed with help of $`C_7^{\stackrel{~}{g}}.`$
The gluino contribution $`C_7^{\stackrel{~}{g}}`$ is induced by the mixings in the left-handed down squark sector, thus by $`V_dV_{CKM}.`$ However, the coefficient $`C_7^{{}_{}{}^{}\stackrel{~}{g}}`$ is almost entirely induced by the mixings in the right-handed down squark sector. If these mixings are similar in size to the CKM mixings, as we argue here, then in our scenario with large $`\mathrm{tan}\beta `$, $`C_7^{{}_{}{}^{}\stackrel{~}{g}}`$ is as large as $`C_7^{\stackrel{~}{g}}.`$ While its absolute contribution to $`C_7^{eff}`$ is subdominant (see Fig. 2) it still may have important phenomenological consequences. Namely, while the mixing induced time dependent CP asymmetry in $`BM_s\gamma `$, where $`M_s`$ is some CP eigenstate, is predicted to be very small, a few percent, in the SM; our model, where it can be expressed as
$`{\displaystyle \frac{\mathrm{\Gamma }(t)\overline{\mathrm{\Gamma }}(t)}{\mathrm{\Gamma }(t)+\overline{\mathrm{\Gamma }}(t)}}=A_t\mathrm{sin}\mathrm{\Delta }M_{B_d}t,A_t={\displaystyle \frac{2\text{Im}\left[e^{i\theta _{B_d}}(C_7+C_7^{\stackrel{~}{g}})C_7^{{}_{}{}^{}\stackrel{~}{g}}\right]}{|C_7^{eff}|^2}},`$ (77)
with $`\theta _{B_d}=arg(M_{12}^{B_d})`$ being the phase in the $`B\overline{B}`$ mixing amplitude, asymmetries $`A_t`$ of more than 10% (which would be a clear and powerful signal of beyond the Sm physics) are allowed. Similar conclusions hold also in models with right-handed neutrinos .
### D Lepton Flavour Violation
So far we have shown that in SUSY SU(5) the new flavour physics contribution to flavour changing hadronic observables is subdominant. At the same time the SUSY contribution to $`bs\gamma `$ induced by the effective Hamiltonian Eq.(55) constrains severely the SUSY scale in our model. Now we turn to study the LFV processes which are completely dominated by the new physics. The amplitude for the process $`l_jl_i\gamma ^{}`$ where the photon is off-shell can be written as
$`M=eϵ^\alpha \overline{l}_i\left[q^2\gamma _\alpha \left(A_1^LP_L+A_1^RP_R\right)+m_{l_j}i\sigma _{\alpha \beta }q^\beta \left(A_2^LP_L+A_2^RP_R\right)\right]l_j,`$ (78)
where $`q`$ is the photon momentum and $`A_{1,2}^{L,R}`$ are the form factors giving rise to the process. Note that all the form factors contribute to $`\mu e`$ conversion but only $`A_2^{L,R}`$ give rise to the lepton radiative decays. The form factors are induced by two types of loop diagrams with neutralinos and charged sleptons in the loop, and charginos and sneutrinos in the loop:
$`A_{1,2}^{L,R}=A_{1,2}^{(n)L,R}+A_{1,2}^{(c)L,R},`$ (79)
where schematically written
$`A_{1,2}^{(n)L,R}`$ $`=`$ $`A_{1,2}^{(n)L,R}(M_{\stackrel{~}{\chi }^0},m_{\stackrel{~}{l}},N,\mathrm{\Gamma }_E),`$ (80)
$`A_{1,2}^{(c)L,R}`$ $`=`$ $`A_{1,2}^{(c)L,R}(M_{\stackrel{~}{\chi }^+},m_{\stackrel{~}{\nu }},O_{L,R},\mathrm{\Gamma }_\nu ),`$ (81)
depend on the slepton masses and mixings as well as on the neutralino and chargino masses and mixings. We have adopted the formulas for the form factors from Ref. and we do not present them here. The decay rate of $`l_jl_i\gamma `$ is then given by
$`\mathrm{\Gamma }(l_jl_i\gamma )={\displaystyle \frac{e^2}{16\pi }}m_{l_j}^5\left(|A_2^L|^2+|A_2^R|^2\right).`$ (82)
We start with studying the decay $`\mu e\gamma `$ in the case where the only source of LFV is the mixing in the right-handed slepton sector as given by Eq.(24), and the mixing matrix $`V_e`$ is the unit matrix. This is the case studied in Ref. . In Fig. 3 we plot the branching ratio of $`\mu e\gamma `$ on the plane of the lightest charged slepton mass $`m_{\stackrel{~}{l}_1}`$ and the lightest chargino mass $`M_{\stackrel{~}{\chi }_1^+}`$ for two values of $`\mathrm{tan}\beta =35,\mathrm{\hspace{0.17em}48}`$ and $`sign(\mu )=1`$ (to achieve Yukawa unification). We have taken into account the experimental constraints coming from $`bs\gamma `$ by requiring that the total value of $`C_7^{eff}`$ is in the allowed range (76). As seen in Fig. 3 the constraint Eq.(76) puts strong lower bounds on the lightest sparticle masses. Our values of the branching ratio of $`\mu e\gamma `$ are much smaller than the quoted values in Ref. . The reason is twofold. First, the $`bs\gamma `$ constraint pushes the sparticle masses to high valuesIt has been noticed in Ref. that for a particular corner of the parameter space where $`M_0m_0A`$ one can supress $`bs\gamma `$ and still have light gauginos. However, also LFV processes are suppressed for this parameter space. and this has not been taken into account in Ref. . Second, the authors of Ref. fix the top Yukawa coupling at $`M_{GUT}`$ to be $`f_t^G=1.4`$ which for large $`\mathrm{tan}\beta `$ implies by far a too large top quark mass. For $`m_t=174`$ GeV and $`\mathrm{tan}\beta =35`$, the correct value is $`f_t^G=0.56.`$ For large part of the parameter space the destructive interference between the gaugino and higgsino contributions suppresses the $`\mu e\gamma `$ branching ratio to almost vanishing values.
The situation changes completely if we allow also flavour mixings in the left-slepton sector and allow $`sign(\mu )`$ to be also positive. Taking $`V_e^{ij}=U_e^{ij}=(V_{CKM}^G)^{ij}`$ as we predict in our model we plot the $`\mu e\gamma `$ branching ratio for $`\mathrm{tan}\beta =35`$ in Fig. 4 and for $`\mathrm{tan}\beta =48`$ in Fig. 5. The branching ratios are about two orders of magnitude higher than in Fig. 3 and no cancellation occurs for any sparticle masses. This implies that $`\mu e\gamma `$ is dominated by the sneutrino-chargino contribution. The LFV pattern here is exactly the same as in models with right-handed neutrinos. Notice that for $`sign(\mu )=+1`$ the sparticle masses are allowed to be much smaller than in the other case. This is because for $`sign(\mu )=+1`$ the chargino contribution to $`bs\gamma `$ interferes destructively with the SM and charged Higgs contributions. In particular, for $`\mathrm{tan}\beta =48`$ and for very small chargino and slepton masses the chargino contribution can be so large that it cancels the SM and $`H^+`$ contributions and gives the allowed $`C_7^{eff}`$ value with an opposite sign. This is seen in Fig. 5 for $`sign(\mu )=+1`$ in which a small parameter region around $`M_{\stackrel{~}{\chi }_1^+}100`$ GeV and $`m_{\stackrel{~}{l}_1}300`$ GeV is allowed. This region can be excluded by collider searches for a very light chargino or by improving the bound on the $`\mu e\gamma `$ branching ratio by a factor of few.
We have shown that the non-vanishing flavour mixings in the left-slepton sector are crucial to ensure the detectibility of $`\mu e\gamma `$ in the planned experiments. Since there is no experimental information on these mixings the central question to ask now is how small the off-diagonal elements of $`V_e`$ can be and still allow successful determination of $`\mu e\gamma .`$ To analyze this question we plot in Fig. 6 the $`\mu e\gamma `$ branching ratio against the lightest slepton mass $`m_{\stackrel{~}{l}_1}`$ for a fixed $`M_2=460`$ GeV which is roughly the minimal chargino mass for $`sign(\mu )=1.`$ The curves denoted by $`a`$ correspond to $`V_e=\text{1},`$ curves denoted by $`b`$ to $`V_e^{ij}=0.1\times U_e^{ij},`$ $`ij`$ and curves denoted by $`c`$ to $`V_e^{ij}=U_e^{ij}.`$ As can be seen, if the off diagonal elements of $`V_e`$ are as small as 10% of the corresponding $`U_e`$ elements then the deep cancellation is superseeded. Let us also mention that for the chosen chargino mass in Fig. 6 the SUSY scale, $`M_{SUSY}=\sqrt{m_{\stackrel{~}{t}_1}m_{\stackrel{~}{t}_2}},`$ is $`M_{SUSY}1200`$ GeV for the minimally allowed slepton mass. Thus the LFV processes are sensitive to SUSY scale above TeV.
Our calculations show that the rate of $`\mu e`$ conversion in nuclei is about $`6\times 10^3`$ times the branching ratio of $`\mu e\gamma .`$ The qualitative behaviour of the $`\mu e`$ conversion rate with the sparticle masses is the same as in the case of $`\mu e\gamma .`$ Thus the results for $`\mu e`$ conversion can be obtained by rescaling the figures for $`\mu e\gamma .`$ Therefore we do not present any new plots for the $`\mu e`$ conversion process. We conclude that the planned $`\mu e`$ conversion experiments are as sensitive to our models as the planned $`\mu e\gamma `$ experiments .
Finally let us discuss the decay $`\tau \mu \gamma .`$ In Fig. 7 we plot the branching ratio of the decay $`\tau \mu \gamma `$ for the same choice of parameters as in Fig. 6. Even for $`\mathrm{tan}\beta =48`$ the branching ratio is always below a few times $`10^9`$ and unobservable in the planned experiments. Thus if $`\tau \mu \gamma `$ will be discovered at these experiments this implies some other LFV scenario than the one considered in this work.
## V Conclusions
Motivated by the sensitivity of the running or approved experiments to flavour violating processes we have studied flavour violation in the minimal SUSY SU(5) GUT at large $`\mathrm{tan}\beta .`$ In this case the flavour mixing occurs both in the left and right slepton and squark mass matrices and are enhanced by the large value of $`\mathrm{tan}\beta .`$ We have calculated the new physics contributions to $`K\overline{K}`$ and $`B\overline{B}`$ mixings and to the decays $`bs\gamma ,`$ $`\mu e\gamma ,`$ $`\tau \mu \gamma `$ and to $`\mu e`$ conversion in nuclei. To predict reliably the rates of these processes we have correctly taken into account the measured values of the low energy parameters as well as the constraints on the SUSY particle masses.
We found that in our model the new physics contributions to $`\mathrm{\Delta }M_K`$ and $`\mathrm{\Delta }M_B`$ are negligible, but might reach a 10% level in $`ϵ_K`$ if there exist new GUT phases. No useful constraints on the model parameters can be derived from these processes.
The decay $`bs\gamma `$ receives contributions from two sources of flavour violation: from the loops proportional to the CKM matrix elements and from the loops exhibiting new flavour violation in the squark mass matrices. The latter contribution interferes constructively with the dominant chargino contribution. At large $`\mathrm{tan}\beta `$ the experimental constraints on the $`bs\gamma `$ branching ratio imply stringent constraints on the SUSY particle masses, especially for $`sign(\mu )=1`$ as required by Yukawa unification. In this case, the SUSY scale is constrained to be at least TeV. For such a high squark masses the new flavour physics contribution to $`bs\gamma `$ branching ratio is a few percent. Nevertheless, this may induce CP asymmetries considerably larger than in the SM.
There is a competition between the sensitivity of the future LFV experiments to the new flavour physics and the constraints on the SUSY scale coming from the $`bs\gamma `$ branching ratio. If the branching ratio of the decay $`\mu e\gamma `$ will be tested down to $`10^{14}`$ and the SUSY scale is below 1 TeV then, the present scenario predicts that $`\mu e\gamma `$ should be discovered in these experiments. The branching ratio of the decay $`\tau \mu \gamma `$ is, however, predicted to be below a few times $`10^9`$ and should not be seen at LHC in the minimal SUSY SU(5).
Acknowledgements
We are very grateful to K. Jakobs and E. Ma for discussions and to J. Hisano for clarifying communication. The work of GB is supported by the DFG, that of KH is partially supported by the Academy of Finland project no. 163394 and the one of MR by U.S. Department of Energy under Grant No. DE-FG03-94ER40837.
|
warning/0005/cond-mat0005199.html
|
ar5iv
|
text
|
# Self-dual Ginzburg-Landau vortices in a disk
## Abstract
We study the properties of the Ginzburg-Laundau model in the self-dual point for a two-dimensional finite system . By a numerical calculation we analyze the solutions of the Euler-Lagrange equations for a cylindrically symmetric ansatz. We also study the self-dual equations for this case. We find that the minimal energy configurations are not given by the Bogomol’nyi equations but by solutions to the Euler Lagrange ones. With a simple approximation scheme we reproduce the result of the numerical calculation.
Pacs: 11.27.+d,74,74.20.De,74.60.Ec
Keywords: Vortices, Superconductivity, Ginzburg-Landau, Mesoscopics.
The study of vortex solutions in Ginzburg Landau (GL) theories has been the subject of continuous interest in different areas of condensed matter and high energy physics.
Static solutions in 2-dimensional infinite samples are characterized by the dimensionless GL parameter $`\kappa `$ which is defined as the ratio of the magnetic penetration length $`\lambda `$ and the coherence length $`\zeta `$. It has been known for several years that the GL model in the infinite plane possesses very special properties at $`\kappa ^2=1/2`$. For instance the second order static equations of motion are equivalent to a set of much simpler first order equations referred as self-dual or Bogomol’nyi equations (BE).
Although the existence of self-dual equations was first pointed out within the study of superconductors by Harden and Arp most of the research on the subject has been done in connection with the role this type of equations play in High Energy Physics. Indeed, self-dual equations were introduced in this context independently by Bogomol’nyi and de Vega and Schaposnik in their study of vortex solutions of the relativistic version of the GL model (Abelian Higgs model). Since then, the properties of the solutions, the connection with Supersymmetry, Topological Field Theories and Duality have been established not only for the Abelian Higgs model but also for related theories in different number of space time dimensions and non-Abelian gauge groups .
Very recently, Akkermans and Mallick have addressed the study of the GL model at the self-dual point for a 2-dimensional disk of finite radius $`R`$ . Their study is relevant for the case of vortices in mesoscopic systems, where the size of the sample is of the order of $`\lambda `$ and $`\zeta `$. As evidenced by recent experiments , the superconducting behavior of mesoscopic disks is radically different from that of the same material in the macroscopic regime . An interesting question that arises is then to determine in which way size effects manifest at the self-dual point and reciprocally to analyze if the special properties that the model show at the self-dual point allow for a simpler interpretation of the experimental results
In this work we shall re-analyze the properties of the GL model for finite systems in the self-dual point. We will perform a numerical study of the equation of motion and explore the role played by the self-dual equations for this case. As a result of our analysis it will be shown that some of the approximations made in are not correct. We shall in turn present a simple approximation scheme which correctly reproduces the numerical calculation.
The GL expression for the energy of a 2-dimensional sample $`\mathrm{\Omega }`$ can be written as
$$E=d^2x\left\{\frac{1}{16\pi }F_{ij}F_{ij}+\frac{1}{2}|D_i\varphi |^2+V(|\varphi |)\right\}$$
(1)
where $`F_{ij}=_iA_j_jA_i`$, and $`D_i\varphi =_i\varphi iqA_i\varphi `$ with $`i=1,2`$. Here $`A_i`$ denotes the electromagnetic vector potential, $`\varphi `$ is a complex scalar field (order parameter) and $`q`$, the charge. Writing the potential as
$$V(|\varphi |)=\beta /2\left(|\varphi |^2\eta ^2\right)^2,$$
(2)
the penetration length and the coherence length are $`\lambda ^2=1/(4\pi q^2\eta ^2)`$ and $`\zeta ^2=1/(2\eta ^2\beta )`$ while the GL parameter is $`\kappa ^2=\beta /(2\pi q^2)`$.
For arbitrary $`\kappa `$ the minimal energy configurations satisfy the second order Euler Lagrange equations,
$$D_iD_i\varphi =2\frac{\delta V}{\delta \varphi ^{}}$$
(3)
$$\frac{1}{4\pi }_iF_{ij}=j_j=q/2i(\varphi ^{}D_j\varphi \varphi D_j\varphi ^{})$$
(4)
Using the identity
$$\frac{1}{4}|D_i\varphi iϵ_{ij}D_j\varphi |^2=\frac{1}{2}|D_i\varphi |^2\pm \frac{1}{2q}ϵ_{ij}_iJ_j\pm \frac{q}{2}B|\varphi ^2|$$
(5)
the energy at the self-dual point can be re-written as
$$E=d^2x\left(\frac{1}{16\pi }(F_{ij}\pm q2\pi ϵ_{ij}(|\varphi |^2\eta ^2))^2+\frac{1}{4}|D_i\varphi iϵ_{ij}D_j\varphi |^2\frac{1}{2q}ϵ_{ij}_iJ_j\right)\frac{q\eta ^2}{2}\mathrm{\Phi }.$$
(6)
where $`\mathrm{\Phi }=_\mathrm{\Omega }d^2xB=_\mathrm{\Omega }\stackrel{}{A}𝑑\stackrel{}{x}`$ is the total magnetic flux through the sample. Assuming that the current is zero at the boundary, a lower bound for the energy is obtained,
$$E|\frac{q\eta ^2}{2}\mathrm{\Phi }|.$$
(7)
Energy configurations saturating the bound must satisfy the self-dual or Bogomol’nyi equations,
$$F_{ij}\pm 2\pi qϵ_{ij}(|\varphi |^2\eta ^2)=0,$$
(8)
$$D_i\varphi iϵ_{ij}D_j\varphi =0.$$
(9)
In the plane these equations are totally equivalent to the Euler Lagrange equations. To obtain finite energy conditions, one has to demand
$$\underset{\rho \mathrm{}}{lim}D_i\mathrm{\Phi }=0\underset{\rho \mathrm{}}{lim}|\mathrm{\Phi }|^2=\eta ^2$$
(10)
which in turns implies that
$$\underset{\rho \mathrm{}}{lim}J_i=0;$$
(11)
On the other hand, writing $`\varphi =|\varphi |e^{i\chi }`$, the current takes the form:
$$J_i=q|\varphi |^2(_i\chi qA_i).$$
(12)
As $`\varphi `$ is a single valued field, the phase $`\chi (\rho ,\theta )`$ must satisfy
$$\chi (\rho ,2\pi )\chi (\rho ,0)=2\pi n,$$
(13)
with $`n`$ an integer. This condition together with (11) implies that the total flux has to be an integer multiple of the quantum of the flux $`\mathrm{\Phi }_0=2\pi /q`$
$$\mathrm{\Phi }=_\mathrm{\Omega }A_i𝑑x^i=\frac{1}{q}_\mathrm{\Omega }_i\chi dx^i=n\mathrm{\Phi }_0,$$
(14)
For each integer $`n`$, equations (8), (9) admit a family of solutions depending on $`2|n|`$ parameters that can be identified as the 2 dimensional coordinates of $`|n|`$ non interacting vortices with flux quantum $`\mathrm{\Phi }_0`$ (the upper or lower sign in the equations has to be chosen according to the sign of $`n`$)().
Let us now concentrate to the case of finite geometries with a boundary. When the current is zero at the boundary, the Bogomol’nyi inequality still provides a bound for the energy and the flux is quantized in units of $`\mathrm{\Phi }_0`$. However, for a given flux, there is a minimal area of the sample for which the equations admit solutions. Indeed, integrating the first equations we get the inequality:
$$\mathrm{\Phi }=_\mathrm{\Omega }B=_\mathrm{\Omega }2\pi q(\eta ^2|\varphi |^2)_\mathrm{\Omega }2\pi \eta ^2=\eta ^2\frac{q}{2}area(\mathrm{\Omega }).$$
(15)
In the infinite plane the requirement of zero current at the boundary is the natural boundary condition since it is the only way of obtaining finite energy solutions. However there is no compelling reason to do so for a finite region. The appropriate boundary condition is ,
$$D_{}\varphi =0.$$
(16)
Notice that this boundary condition only implies vanishing of the normal component of the current at the boundary. The tangential component is left in principle undetermined. Nevertheless, it is possible to show that for configurations satisfying the self-dual equations, it holds the relation,
$$J_i=\pm \frac{q}{2}ϵ_{ij}_j|\varphi ^2|$$
(17)
That is, for solutions of the BE equations both components of the current must vanish at the boundary. As discussed above, this implies that the total flux is quantized.
Suppose now that instead of imposing a boundary condition over $`J_{}`$, we fix the total flux $`\mathrm{\Phi }`$. It is clear, that if $`\mathrm{\Phi }/\mathrm{\Phi }_0`$ is not an integer, a tangential component of the current will be established at the boundary and minimal energy configurations will not be given by solutions to the BE but by solutions to the Euler Lagrange equations. More interestingly, when $`\mathrm{\Phi }/\mathrm{\Phi }_0=m`$, with $`m`$ integer, although the BE do admit solutions, they are not the minimal energy configuration. In fact, it is energetically favorable to create $`nm`$ vortices and to develop a tangential current at the boundary.
Let us illustrate this in a cylindrically symmetric ansatz:
$`\varphi (x)`$ $`=`$ $`f(\rho )e^{in\theta }`$ (18)
$`A_\theta (x)`$ $`=`$ $`A(\rho )`$ (19)
$`A_\rho (x)`$ $`=`$ $`0.`$ (20)
Defining dimensionless variables $`x(r)=nqA(r)`$, $`z(r)=f(r)/\eta `$, $`r=(\rho /\lambda )`$, the self-dual equations become
$`x^{}\pm {\displaystyle \frac{r}{2}}(z^21)`$ $`=`$ $`0`$ (21)
$`z^{}{\displaystyle \frac{xz}{r}}`$ $`=`$ $`0`$ (22)
and the boundary condition (16) translates into a Neumann condition for the order parameter:
$$z^{}(R^{})=0$$
(23)
where $`R^{}=R/\lambda `$. Thus, we see that unless $`x(R^{})=0`$ (which implies that $`\mathrm{\Phi }=A=2\pi n/q`$) the BE (22) cannot be satisfied at $`r=R^{}`$.
Minimal energy solutions are then obtained by solving the second order Euler-Lagrange equations. In our ansatz they read,
$$\frac{d^2x}{dr^2}\frac{1}{r}\frac{dx}{dr}xz^2=0$$
(24)
$$\frac{d^2z}{dr^2}+\frac{1}{r}\frac{dz}{dr}\frac{x^2z}{r^2}+\kappa ^2z(1z^2)=0,$$
(25)
while the energy can be expressed as,
$$E=\pi \eta ^2_0^R^{}r𝑑r[(\frac{x^{}}{r})^2+z^2+\frac{x^2z^2}{r^2}+\kappa ^2(z^21)^2].$$
(26)
Regularity of cylindrical coordinates impose conditions at $`r=0`$, and together with the Neumann condition (23) and the definition of $`x(R^{})`$ in terms of $`n`$ and $`\mathrm{\Phi }`$ we have the following set of boundary conditions:
$`x(0)`$ $`=`$ $`nx(R^{})=n{\displaystyle \frac{\mathrm{\Phi }}{\mathrm{\Phi }_0}}`$ (27)
$`z(0)`$ $`=`$ $`0z^{}(R^{})=0.`$ (28)
We have analyzed the existence of solutions of the system (24,25,28) by numerical integration. We employed a relaxation method for boundary value problems . In such method, the differential equations are discretized in a convenient mesh and converted to a set of coupled algebraic equations. The system is then solved using Newton’s iterative method, starting from an initial guess and improving it iteratively. Given $`\mathrm{\Phi }`$ and $`R^{}`$ a solution for each integer $`n`$ is found corresponding to a local minimum of the energy. Then, the $`n`$ giving the lowest energy is selected.
In figure 1 we show the solutions corresponding to $`\mathrm{\Phi }=6\mathrm{\Phi }_0,`$ $`R^{}=10`$. For these values, the minimal energy configuration corresponds to $`n=1`$. The energy of this solution is $`E=3.51\pi \eta ^2`$. As the external flux is an integer number of flux quanta, the Bogomoln’nyi equations also have solutions for this case with energy $`E=6\pi \eta ^2`$. Thus, the system lowers its energy by allowing a tangential component of the current $`J_{||}(R^{})0`$.
Notice nevertheless that there is a point $`R_0^{}`$ such that $`J(R_0^{})=0`$. Following reference , we will consider the disk as $`\mathrm{\Omega }=\mathrm{\Omega }_1\mathrm{\Omega }_2`$ where $`\mathrm{\Omega }_1`$, is the inner disk $`0rR_0^{}`$ and $`\mathrm{\Omega }_2`$, the outer ring $`R_0^{}rR^{}`$. We will express the energy as:
$$E(\mathrm{\Omega })=E(\mathrm{\Omega }_1)+E(\mathrm{\Omega }_2)$$
(29)
As the current vanishes at $`R_0^{}`$, the authors in assumed that the minimal energy configurations satisfy the Bogomol’nyi equations in $`\mathrm{\Omega }_1`$; however it can be shown that this is not the case. Consider the function $`L[r]=z^{}(r)x(r)z(r)/r`$, evaluated with the solutions of the Euler-Lagrange equations. If the Bogomol’nyi equations are satisfied on $`\mathrm{\Omega }_1`$, then $`L[r]=0`$ in this region. But, as the current is not zero on the external boundary $`r=R^{}`$, the Bogomol’nyi equations are not satisfied on $`\mathrm{\Omega }_2`$, and thus $`L[r]0`$ on $`\mathrm{\Omega }_2`$. Clearly, regularity of the solutions of ordinary differential equations prevents the existence of a function that vanishes in the whole region $`0<r<R_0^{}`$ but is different from zero for $`r>R_0^{}`$. Although the solutions of the self-dual equation minimize the energy on the internal region $`\mathrm{\Omega }_1`$, any regular extension of the solution to the whole disk will not minimize the total energy.
Having said this we should notice that even though the self-dual solutions are not exact solutions in the inner disk $`\mathrm{\Omega }_1`$, they are in fact a very good approximation. A numerical analysis of both solutions shows that for a wide range of parameters, they differ only in around one part in a thousand and the same is true for the energy. Then we will take,
$$E(\mathrm{\Omega }_1)\pi \eta ^2|n|$$
(30)
Let us now analyze the contribution of the $`\mathrm{\Omega }_2`$ region to the energy.
$$E(\mathrm{\Omega }_2)=\pi \eta ^2_{R_0^{}}^R^{}r𝑑r[(\frac{x^{}}{r})^2+z^2+\frac{x^2z^2}{r^2}+\frac{1}{2}(z^21)^2]$$
(31)
We first review the main steps in . There it was assumed that as the fields are concentrated in a region of width of order one from the border, this expression could be approximated as
$$E(\mathrm{\Omega }_2)\pi \eta ^2r[(\frac{x^{}}{r})^2+z^2+\frac{x^2z^2}{r^2}+\frac{1}{2}(z^21)^2]|_{r=R^{}}.$$
(32)
Then, the condition $`\frac{\delta E}{\delta z}=0`$ would give
$$\frac{x^2(R)}{R^2}=1z^2$$
(33)
As a next step Akkermans and Mallick neglected the magnetic energy contribution (first term in (32)) arriving to the following expression for the energy,
$$E(\mathrm{\Omega }_2)\pi \eta ^2R^{}(\frac{x^2(R^{})}{R^2}\frac{1}{2}\frac{x^4(R^{})}{R^4})$$
(34)
Finally, the authors neglected, in the large $`R`$ limit, the quartic term in (34) ending with the expression
$$E(\mathrm{\Omega }_2)\pi \eta ^2\frac{x^2(R^{})}{R^{}}$$
(35)
The reasoning above suffers from two main drawbacks. Although the field $`z`$ is practically constant, the field $`x`$ is not, making the approximation of the integral not valid. In fact, our numerical simulation shows that equation (33) is not fulfilled. Second, as shown in fig 2 the magnetic energy is of the same order than the third term in (31).
The correct field distribution in $`\mathrm{\Omega }_2`$ can instead be obtained by a simple approximation using
$$\nu =\frac{(n\frac{\mathrm{\Phi }}{\mathrm{\Phi }_0})}{R^{}}$$
(36)
as an expansion parameter. Indeed, defining
$$\stackrel{~}{x}=x/\nu $$
(37)
the equations of motion become
$$\frac{d^2\stackrel{~}{x}}{dr^2}\frac{1}{r}\frac{d\stackrel{~}{x}}{dr}z^2\stackrel{~}{x}=0$$
(38)
$$\frac{d^2z}{dr^2}+\frac{1}{r}\frac{dz}{dr}\nu ^2\frac{\stackrel{~}{x}^2z}{r^2}+\frac{1}{2}z(1z^2)=0$$
(39)
with the boundary conditions
$$\stackrel{~}{x}(R_0^{})=0\stackrel{~}{x}(R^{})=R^{}$$
(40)
$$z(R_0^{})=1z^{}(R^{})=0$$
(41)
In order to solve these equations, we make an expansion of the form
$`\stackrel{~}{x}`$ $`=`$ $`\stackrel{~}{x}_0+\nu ^2\stackrel{~}{x}_2+O(\nu ^4)`$ (42)
$`z`$ $`=`$ $`1+\nu ^2z_2+O(\nu ^4)`$ (43)
The boundary conditions will be satisfied by the solution $`\stackrel{~}{x}_0`$ while homogeneous conditions are valid for $`\stackrel{~}{x}_2`$ and $`z_2`$. To lowest order we obtain
$$\frac{d^2\stackrel{~}{x}_0}{dr^2}\frac{1}{r}\frac{d\stackrel{~}{x}_0}{dr}\stackrel{~}{x}_0=0$$
(44)
with solution
$$\stackrel{~}{x}_0(r)=c_0r\left[I_1(r)\frac{I_1(R_0^{})}{K_1(R_0^{})}K_1(r)\right]$$
(45)
Here $`c_0^1=[I_1(R^{})\frac{I_1(R_0^{})}{K_1(R_0^{})}K_1(R^{})]`$ and $`I_1(r)`$, $`K_1(r)`$ are Bessel functions.
With these results we can solve the order $`\nu ^2`$ equations:
$$\frac{d^2\stackrel{~}{x}_2}{dr^2}\frac{1}{r}\frac{d\stackrel{~}{x}_2}{dr}\stackrel{~}{x}_2=2z_2\stackrel{~}{x}_0,$$
(46)
$$\frac{d^2z_2}{dr^2}+\frac{1}{r}\frac{dz_2}{dr}z_2=\frac{\stackrel{~}{x}_0^2}{r^2}$$
(47)
with homogeneous boundary conditions for both functions. Using the Green functions for each equation we obtain the solution
$$z_2(r)=y_2(r)_{R_0^{}}^r\frac{y_1(r^{})f(r^{})}{W_{y_1,y_2}}𝑑r^{}+y_1(r)_r^R^{}\frac{y_2(r^{})f(r^{})}{W_{y_1,y_2}}𝑑r^{}$$
(48)
where
$`y_1(r)`$ $`=`$ $`K_0(r){\displaystyle \frac{K_0(R_0^{})}{I_0(R_0^{})}}I_0(r)`$ (49)
$`y_2(r)`$ $`=`$ $`K_0(r){\displaystyle \frac{K_0^{}(R^{})}{I_0^{}(R^{})}}I_0(r)`$ (50)
$`f(r)`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{x}_0^2}{r^2}}`$ (51)
and $`W_{y_1,y_2}`$ is the Wronskian.
The same steps are followed to calculate $`\stackrel{~}{x}_2`$ for which we obtain
$$\stackrel{~}{x}_2(r)=Y_2(r)_{R_0^{}}^r\frac{Y_1(r^{})g(r^{})}{W_{Y_1,Y_2}}𝑑r^{}+Y_1(r)_r^R^{}\frac{Y_2(r^{})g(r^{})}{W_{Y_1,Y_2}}𝑑r^{}$$
(52)
In this case
$`Y_1(r)`$ $`=`$ $`r[K_1(r){\displaystyle \frac{K_1(R_0^{})}{I_1(R_0^{})}}I_1(r)]`$ (53)
$`Y_2(r)`$ $`=`$ $`r[K_1(r){\displaystyle \frac{K_1(R^{})}{I_1(R^{})}}I_1(r)]`$ (54)
$`g(r)`$ $`=`$ $`2z_2\stackrel{~}{x}_0`$ (55)
Having obtained the solutions $`\stackrel{~}{x}=\stackrel{~}{x}_0+\nu ^2\stackrel{~}{x}_2`$ and $`z=1+\nu ^2z_2`$ we can now consider the expression for the energy in this expansion:
$`E(\mathrm{\Omega }_2)`$ $`=`$ $`\pi \eta ^2{\displaystyle _{R_0^{}}^R^{}}dr\{{\displaystyle \frac{\nu ^2}{r}}[({\displaystyle \frac{d\stackrel{~}{x}_0}{dr}})^2+\stackrel{~}{x}_0^2]+\nu ^4[r({\displaystyle \frac{dz_2}{dr}})^2+{\displaystyle \frac{2}{r}}({\displaystyle \frac{d\stackrel{~}{x}_0}{dr}}{\displaystyle \frac{d\stackrel{~}{x}_2}{dr}}+\stackrel{~}{x}_0^2z_2+\stackrel{~}{x}_0\stackrel{~}{x}_2)`$ (56)
$`+`$ $`rz_2^2]+O(\nu ^6)\}`$ (57)
Using the equations of motion, the energy takes the form
$$\frac{E(\mathrm{\Omega })}{\pi \eta ^2}=n+\nu ^2\frac{d\stackrel{~}{x}_0}{dr}(R^{})+\nu ^4_{R_0^{}}^R^{}𝑑r\frac{\stackrel{~}{x}_0^2z_2}{r}$$
(58)
The above expression gives us the energy of the vortex configuration up to the order $`\nu ^4`$. However an obvious drawback of this expression is the presence of the point $`R_0^{}`$, which should be located numerically, and then preventing any analytical predictability power of the equation. Nevertheless we can convince ourselves that the point $`R_0^{}`$ can be shrunk to zero. The reason is that at leading order we can approximate the whole solution as a superposition of a Bogolmo’nyi vortex and a vortex concentrated at the boundary. Because both kind of solutions are exponentially small in complementary regions, any contribution from the non-linearity of the equations is exponentially suppressed. In fact, a simple plot of the solutions (45,48,52) shows that the solution with $`R_0^{}=0`$ only differs in about $`10^3`$ with the one with non-zero $`R_0^{}`$. Therefore, we will take $`R_0^{}=0`$.
In this case the energy can be found by a simple numerical integration. Notice that the integrals only depends on $`R^{}`$ and not on $`\mathrm{\Phi }`$ or $`n`$. The result can be compared with the obtained in reference . In the following table we show a comparison between the energy values obtained from our approximate equation (58), the ones obtained from the expression of reference and and the exact ones, corresponding to vortex solutions with $`R=10`$ and $`n=1`$. We see that even for big values of $`\nu `$, our approximate equation gives a result in excellent agreement with the exact ones.
$`\mathrm{\Phi }`$ Energy Energy Exact from (58) of Ref. energy 1.0 1.0 1.0 1.0 3.0 1.42 1.4 1.42 5.0 2.64 2.6 2.64 8.0 5.78 5.9 5.79 12.0 11.51 13.1 11.27 18.0 18.81 29.9 18.0
Furthermore we can find a large $`R`$ expansion of equation (58). Using the asymptotic expansion of the Bessel functions, the first term in (58) has the asymptotic form $`\nu ^2(R^{}+0.5+O(1/R^{}))`$. The coefficient of $`\nu ^4`$ term was found by a numerical fit. The resulting approximate expression for the energy is
$$\frac{E}{\pi \eta ^2}=n+\nu ^2R^{}a(R^{})\nu ^4R^{}b(R^{})+O(1/R^{},\nu ^6)$$
(59)
where
$$a(R^{})=1+1/(2R^{}),b(R^{})=0.139+0.111/R^{}$$
(60)
The most stable configuration corresponds to a vortex solution with vortex number $`n`$ such that is an absolute minimum of the energy. Equation (59) can be approximately minimized with respect to $`n`$, with $`b(R^{})/a(R^{})`$ as the expansion parameter. We find that the vortex number that minimizes the energy is given by:
$$n=\left[\frac{\mathrm{\Phi }}{\mathrm{\Phi }_0}\frac{R^{}}{2a}\left(\frac{a^3b}{a^33/2b}\right)+\frac{1}{2}\right]$$
(61)
where $`[x]`$ means the integer part of x.
The magnetization of the system shown in fig.3 is obtained from the Gibbs free energy, $`G=E2\pi \eta ^2(\mathrm{\Phi }/\mathrm{\Phi }_0)^2/R^2`$, as
$$M=\frac{G}{\mathrm{\Phi }}=\frac{\pi \eta ^2}{\mathrm{\Phi }_0}\left(\frac{2a(R^{})}{R^{}}\left(\frac{\mathrm{\Phi }}{\mathrm{\Phi }_0}n\right)\frac{4b(R^{})}{R^3}\left(\frac{\mathrm{\Phi }}{\mathrm{\Phi }_0}n\right)^34\frac{\mathrm{\Phi }}{\mathrm{\Phi }_0}\frac{1}{R^2}\right)$$
(62)
where $`n`$ is given in equation (61).
In this work we have analyzed the existence of vortex solutions of the Ginzburg Landau theory at the self dual point $`\kappa ^2=1/2`$ for a two dimensional disk of finite radius. Our original aim was to pursue further the interesting proposal made by Akkermans and Mallick of exploiting the properties of the Bogomol’nyi equations for the study of the vortices at the self-dual point. Unfortunately our numerical study revealed that some of the assumptions made there are not entirely correct.
We have shown that the minimal energy configurations do not satisfy the Bogomol’nyi equations in the inner disk. Nevertheless they provide a very good approximation to the actual solutions. Concerning the behaviour of the fields in the outer ring we have provided a simple analytical approximation scheme which do conform the numerical simulation and allows to obtain a selection rule for the number of vortices as a function of the external flux.
|
warning/0005/astro-ph0005581.html
|
ar5iv
|
text
|
# Detection of Strong Clustering of Extremely Red Objects: Implications for the Density of 𝒛>𝟏 Ellipticals
## 1 Introduction
Near-infrared surveys prompted the discovery of a population of objects with very red optical-infrared colors (Extremely Red Objects, EROs hereafter; e.g. Elston et al. 1988, McCarthy et al. 1992, Hu & Ridgway 1994; Thompson et al. 1999; Yan et al. 2000; Scodeggio & Silva 2000). In general, objects have been classified as EROs when they had redder colors than late type galaxies (with negligible dust extinction) at any redshift. However, depending on the depth of the photometry and on the available filters, different selection criteria have been used to select EROs. In this paper, EROs are defined as objects with $`RKs`$5 (see Sect. 5 for more details on this choice).
The red colors of EROs are consistent with two classes of galaxies: they could be old, passively evolving elliptical galaxies at $`z\stackrel{>}{}1`$ which are so red because of the large $`K`$–correction. EROs may also be strongly dust–reddened star–forming galaxies or AGN. The observational results of the last few years showed that both classes of galaxies are indeed present in the ERO population: on one hand, a few objects were spectroscopically confirmed to be $`z>1`$ ellipticals (Dunlop et al. 1996, Spinrad et al. 1997, Liu et al. 2000, and marginally, Soifer et al. 1999), or to have surface brightness profiles consistent with being dynamically relaxed early type galaxies (e.g. Stiavelli et al. 1999, Benitez et al. 1999). On the other hand, other EROs have been detected in the sub–mm (Cimatti et al. 1998, Dey et al. 1999, Smail et al. 1999, Andreani et. al. 2000), thus providing examples of high–redshift starburst galaxies reddened by strong dust extinction and characterized by high star formation rates. The relative contribution of the two classes of objects to the whole ERO population is still unknown, but there are preliminary indications, based on near-IR and optical spectroscopy and on surface brightness analysis, that ellipticals may represent the largest fraction of this population (e.g. Cimatti et al. 1999, 2000; Liu et al. 2000; Moriondo et al. 2000). A small fraction of low-mass-stars and brown dwarfs among EROs is also expected in case of unresolved objects (e.g. Thompson et al. 2000, Cuby et al. 1999).
The importance of studying EROs is clear especially for the clues that they could provide on the formation and evolution of elliptical galaxies. For instance, existing realizations of hierarchical models of galaxy formation predict a significant decline in the comoving density of the ellipticals with $`z`$, as they should form through merging at $`z\stackrel{<}{}2`$ (Kauffmann 1996, Baugh et al. 1996a), so that a measure of a decline of their comoving density would provide a stringent proof of these models. Conflicting results have been found so far about such issue: some works claim the detection of a deficit of $`z>1`$ ellipticals (e.g. Kauffmann et al. 1996, Zepf 1997, Franceschini et al. 1998, Barger et al. 1999, Menanteau et al. 1999), whereas others find that a constant comoving density of ellipticals up to $`z2`$ is consistent with the data (Totani & Yoshii 1997, Benitez et al. 1999, Broadhurst & Bowens 2000, Schade et al. 1999). A potentially serious problem in these studies is suspected to be the influence of the field-to-field variations in the density of EROs due to the small fields of view usually covered in the near-infrared. For instance, Barger et al. (1999) found a very low surface density of EROs in a 60 arcmin<sup>2</sup> survey to $`K=20`$, while on a similar area McCracken et al. (1999) observed a density three times larger at the same $`K`$ level.
Similar uncertainties have been found in the attempts of deriving the fraction of high redshift galaxies among IR selected samples, which is believed to be a stringent test for the formation of the massive galaxies (Broadhurst et al. 1992, Kauffmann & Charlot 1998). Fontana et al. (1999), using photometric redshifts, found that the fraction of high–$`z`$ galaxies in a collection of small deep fields complete to $`K=21`$ was low and comparable with the predictions of the cold dark matter (CDM) models, but not with passive evolution (PLE) models. The preliminary results of Eisenhardt et al. (2000), suggesting a much higher fraction of high–$`z`$ galaxies, are instead consistent with both CDM and PLE models.
The main aim of our survey was to encompass the difficulties induced by the cosmic variance, obtaining a sample of EROs on a large area, at moderately deep $`K`$ levels, in order to minimize and possibly to detect the effects of their clustering, and to compare the observed density with that expected in the case of passive evolution of ellipticals. Our survey is larger by more than a factor of four than the Thompson et al. (1999) survey, and by more than an order of magnitude than all the other previous surveys for EROs, at the same limiting magnitudes. With the large area covered we aimed also to detect a sample, or place limits to the surface density of the very rare class of extreme EROs with $`RKs7`$.
In this paper, the observational results of this survey and their main implications are presented. A more detailed interpretation of our findings will be presented in a forthcoming paper (Daddi et al., in preparation).
The paper is organized as follows: we first describe the data reduction and analysis, then we present the counts of field galaxies. In Sect. 4 the sample of EROs is described. Sect. 5 contains the analysis of the clustering of field galaxies and EROs. The main implications of our findings are discussed in Sect. 6. H$`{}_{0}{}^{}=50`$ km s<sup>-1</sup> Mpc<sup>-1</sup> throughout the paper.
## 2 Observations, data reduction and photometry
### 2.1 $`Ks`$-band imaging
The $`Ks`$ observations were made with the ESO NTT 3.5m telescope in La Silla, during the nights of 27–30 March 1999, using the SOFI camera (Moorwood et al. 1998) with a field of view of about 5$`\mathrm{}\times `$5$`\mathrm{}`$. SOFI is equipped with a Hawaii HgCdTe 1024x1024 array, with a scale of 0.29$`\mathrm{}`$/pixel. The $`Ks`$ filter has $`\lambda _\mathrm{c}=2.16\mu m`$ and $`\mathrm{\Delta }\lambda 0.3\mu m`$ and it is slightly bluer than the standard $`K`$ filter in order to reduce the thermal background.
The center of the observed field is at $`\alpha `$ = 14<sup>h</sup>49<sup>m</sup>29<sup>s</sup> and $`\delta `$ = 09$`\mathrm{°}00\mathrm{}00\mathrm{}`$ (J2000). The observed field is one of the fields described in Yee et al. (2000) to which we refer for details about its selection. The main criteria were not to have any apparent nearby clusters and to be at high galactic latitude.
The images were taken with a pattern of fixed offsets of 144$`\mathrm{}`$ (about half of the SOFI field of view) over a grid of 9$`\times `$13 pointings. The total area covered by the observations was about 24$`\times `$34 arcmin, with a local integration time of 12 minutes in the central deepest region of the field. In the shallower region, the effective integration time is reduced to not less than 6 minutes. The total amount of time required to cover the whole field was about 5.5 hours.
The data reduction was carried out using the IRAF software. The images were flat-fielded with twilight flats. The sky background was estimated and subtracted for each frame using a clipped average of 6–8 adjacent frames (excluding the central frame itself). The photometric calibration was achieved each night with the observation of 5-7 standard stars taken from Persson et al. (1998). The zero-points have a scatter of $``$ 0.015 magnitudes in each night and a night-to-night variation within 0.02 magnitudes. Each frame was scaled to the same photometric level correcting for the different zero-points and airmasses. Accurate spatial offsets were measured for each frame using the area in common with the adjacent frames. The images were then combined, masking the known bad pixels, in order to obtain the final mosaic. The cosmic rays were detected and replaced by the local median using the task cosmicrays of the IRAF package ccdred. The effective seeing of the final coadded mosaic ranges from 0.9$`\mathrm{}`$ to 1.1$`\mathrm{}`$.
### 2.2 $`R`$-band imaging
The $`R`$-band data were taken in May 19–21 1998 with the 4.2m William Herschel Telescope on La Palma. The observations were done using the prime focus camera, equipped with a thinned 2048x4096 pixels EEV10 chip, with a scale of 0.237$`\mathrm{}`$/pixel. This gives a field of view of about 8.1$`\mathrm{}\times 16.2\mathrm{}`$. A standard Johnson $`R`$–filter was used. The whole field has been covered by a mosaic of 6 pointings. Each pointing consisted of at least 3 exposures of 1200s taken with small offsets. The total integration times per pointing was therefore 3600s, with the exceptions of two pointings with 4800s and 6000s.
The photometric calibration was achieved with standard stars taken from Landolt (1992) with a scatter in the zeropoints, from the different stars used, below 0.01 magnitudes. The images were de-biased and then flatfielded, using a master flatfield constructed from the science exposures, scaled to the same zeropoint and then combined. The seeing of the final $`R`$-band mosaic was between 0.7$`\mathrm{}`$ and 0.8$`\mathrm{}`$.
### 2.3 Sample selection and $`Ks`$ photometry
The software SExtractor (Bertin & Arnouts 1996) was run on the $`Ks`$ mosaic with a background weighted threshold in order to take into account the depth variations across the area, as defined by SExtractor. Among the detected objects, all those with S/N$`>5`$ in a 2<sup>′′</sup> circular aperture (twice the average seeing FWHM) were selected and added to the catalog. A few spurious detections (e.g. close to image defects) have been excluded after a visual inspection of the image. The final catalog includes 4585 objects. In the central deepest region, the 5$`\sigma `$ limiting aperture magnitude is $`Ks`$(2<sup>′′</sup>)=19.6, whereas in the remaining area such limit is $`Ks`$(2<sup>′′</sup>)$`\stackrel{>}{}`$19.2 because of the reduced integration time.
Isophotal magnitudes were measured with a limiting threshold of about 0.7$`\sigma _{\mathrm{sky}}`$ corresponding in the central area to a surface brightness limit of about $`\mu _{\mathrm{lim}}21`$ mag arcsec<sup>-2</sup>. The aperture correction from 2<sup>′′</sup> to total magnitudes was estimated throughout the area by measuring the difference between the isophotal and the 2$`\mathrm{}`$ aperture magnitudes for the stars with $`Ks<`$16. A differential correction, in the range of 0.16–0.30 magnitudes, was measured for different regions of the mosaic with a typical scatter less than 0.03 magnitudes. For the bright objects the isophotal magnitudes were on average consistent with the Kron automatic aperture magnitudes. However, we adopted the isophotal magnitudes because the Kron magnitudes are rather unstable at faint flux levels, where the low signal often does not allow to define the correct automatic aperture.
The total $`Ks`$ magnitudes were then defined as the brightest between the isophotal and the corrected aperture magnitude. This allowed to safely assign a total magnitude for both the faint and the bright objects. The typical $`Ks`$ magnitude where the corrected aperture magnitude begins to be adopted as the total one is $`Ks`$$``$18 in the central deepest region.
The completeness of our catalog has been estimated by adding artificial objects to the $`Ks`$ mosaic in empty positions, using the IRAF package artdata. Point-like sources as well as objects with De Vaucouleurs and exponential profiles (convolved with the seeing PSF) were simulated, and SExtractor was run with the same detection parameters as for the real data. The 85% completeness magnitude for the deepest area is $`Ks=19.2`$ for point-like sources. The completeness decreases to $``$ 70% for the worst case that we have tested, i.e. for exponential galaxies with 0.7$`\mathrm{}`$ half–light radius. In the shallower area the corresponding limiting magnitude is $`Ks18.8`$. Most of the $`Ks\stackrel{>}{}18.5`$ galaxies are anyway expected to be only barely resolved with the $`Ks`$ 1$`\mathrm{}`$ seeing (Saracco et al. 1997), and this certainly occurs for the distant $`z\stackrel{>}{}1`$ ellipticals, and thus their completeness limits can be assumed to be similar to those for stars.
### 2.4 $`R`$-band photometry and colors
In order to recover the $`R`$-band counterparts of the $`Ks`$-selected objects, a coordinate mapping between the $`Ks`$ and the $`R`$ images was derived. SExtractor was then run in ASSOC mode with a search box of 2$`\times `$ FWHM<sub>R</sub>. The regions around bright stars or defects in the $`R`$ and $`Ks`$ band images were excluded from this analysis. The final effective area is 701 arcmin<sup>2</sup> at $`Ks18.8`$ and 447.5 arcmin<sup>2</sup> at $`18.8<Ks19.2`$.
Whenever an object had S/N$`<`$3 in the $`R`$-band image 3$`\sigma `$ limits were assigned. The 3$`\sigma `$ limiting magnitude in a 2$`\mathrm{}`$ diameter aperture is $`R>26.2`$ for most of the area, reaching $`R>26.5`$ in the deepest pointing. When S/N$`>`$3, 2$`\mathrm{}`$ diameter corrected aperture magnitudes were assigned to each object. The aperture correction was derived in the same way as for the $`Ks`$-band, with slightly smaller corrections because of the better $`R`$-band seeing. The magnitudes were dereddened for Galactic extinction. At the Galactic coordinates of the center of our field ($`l5\stackrel{}{.}5`$, $`b57\mathrm{°}`$), the extinction coefficients from Burstein & Heiles (1982) and from Schlegel et al. (1998) are $`A_\mathrm{B}=0.04`$ and $`A_\mathrm{B}=0.13`$ respectively. Since the two values are derived in different ways, neither of the two can be discarded. The average was therefore adopted, obtaining a correction of 0.052 magnitudes in $`R`$ and negligible in $`Ks`$. This introduces an uncertainty of $``$ 0.03 magnitudes in the dereddened $`R`$ magnitudes.
Finally, the $`RKs`$ colors are defined for all the objects as the difference between the $`R`$ and $`Ks`$ corrected aperture magnitudes. Thanks to the depth of the $`R`$-band data, colors as red as $`RKs=7`$ could be measured down to the $`Ks`$ magnitude limits of our survey. Because of the long integrations used in the $`R`$-band the objects with $`R\stackrel{<}{}20`$ were saturated. This effect is obviously more important for stars than for galaxies; moreover, since we are interested in the extremely red galaxy population, the effects due to saturation have no impact on our results.
### 2.5 Star–Galaxy classification
The star–galaxy (S/G) separation was done by means of the SExtractor CLASS\_STAR parameter in both the $`R`$ and $`Ks`$ band. This classification was found to be reliable for objects with $`Ks\stackrel{<}{}`$17.5 and $`R\stackrel{<}{}`$24. Because of the seeing variations through the area in both bands, a variable CLASS\_STAR threshold was adopted in different subareas. Given the better seeing of the $`R`$-band, the classification was based mostly on that band, switching to the $`Ks`$ CLASS\_STAR for objects close to the saturation level in $`R`$. Objects which are fainter than $`Ks`$ = 17.5 and $`R`$ = 24 have not been classified and have been considered to be galaxies. This means that a S/G separation can be provided only for objects with colors $`RKs\stackrel{<}{}`$5 at $`Ks`$=19, $`RKs\stackrel{<}{}`$6 at $`Ks`$=18, and so on (see Fig. 5 and the upper panel of Fig. 3). From the small number of objects classified as stars which are close to the diagonal straight line which indicates our S/G classification limit in the color - magnitude plane, we can safely conclude that our inability to properly classify very red, faint objects has almost no effect on the total star number counts and, therefore, on the galaxy number counts. In addition, it is possible that near the faint limit of our survey a very small fraction of very compact, blue galaxies, such as for instance AGN or compact narrow emission line galaxies (e.g. Koo & Kron 1988), could have been incorrectly classified as stars.
## 3 $`Ks`$-band number counts
Galaxy number counts in the $`Ks`$ band can provide more advantages in studying galaxy evolution and cosmological geometry than optical counts because they are much less sensitive to the evolution of stellar population and to the dust extinction. Our survey, which covers the magnitude range $`14<Ks<19.2`$, represents the widest among the previous deep surveys at levels fainter than $`Ks>18`$.
Table 1 summarizes the number of galaxies and stars detected in each $`Ks`$ bin and Fig. 1 shows the corresponding differential counts in 0.5 magnitude bins. No correction for incompleteness was applied. Galaxies start to dominate over stars at $`Ks16.5`$ and their surface density is about a factor of 8 higher than the stellar surface density at $`Ks19.`$
The slopes of the galaxy number counts were derived over the magnitude range covered by our survey. At bright magnitudes a slope of $`\gamma =0.53\pm 0.02`$ is found in the range $`14<Ks<17.5`$. We confirm that the $`K`$-band galaxy counts show a flattening at $`Ks17.5`$, where the best fit slope changes from $`\gamma =0.53`$ to $`\gamma =0.32\pm 0.02`$ (see Fig. 2). The leveling off of the counts below a slope of 0.4 indicates that the differential contribution to the extragalactic background light (EBL) in the $`K`$ band peaks at $`Ks1718`$ and then starts to decrease at fainter fluxes. The contribution to the EBL over the magnitude range $`14Ks19.2`$ sampled by our survey is about $`4.20`$ nW/m<sup>2</sup>/sr, which constitutes about $`53\%`$ of the estimated EBL from discrete sources in the $`K`$-band (cf. Pozzetti et al. 1998, Madau & Pozzetti 2000).
Fig. 2 shows the differential galaxy number counts in our survey compared with a compilation of $`K`$-band published surveys. No attempt was made to correct for different filters. As shown in the figure, our counts are in very good agreement with the average counts of previous surveys (Hall & Green 1998).
## 4 The sample of EROs
In Fig. 3 the $`RKs`$ vs. $`Ks`$ color - magnitude diagram is plotted for both stars and galaxies in our sample. The diagonal straight line in the upper panel indicates our S/G classification limit in this plane. Because objects above this line are not classified, no star can appear in the upper right corner. However, this figure shows that there are very few stars with a color redder than $`RKs>5`$, even in the region of this color - magnitude diagram where our morphological classification is still reliable. This suggests that very few stars should be present in the sample of objects for which no morphological classification was possible.
Fig. 4 shows the distribution of colors for galaxies fainter than $`Ks=16.5`$, from which the median color at different $`Ks`$ magnitude levels have been calculated (see Table 2). In spite of the presence of objects with color upper and lower limits, the use of the median colors (instead of the mean colors) allows unbiased estimates in the $`Ks`$ range we are considering. The faintest bin with 18.8$`<Ks`$19.2 includes only the objects detected in the deeper region. The median $`RKs`$ color of the galaxies increases by 0.5 magnitudes from $`Ks=16.5`$ to $`Ks18`$ and then it remains almost constant up to the limits of our survey ($`Ks`$=19.2). This trend is similar to what is found by Saracco et al. (1999) for the median $`JKs`$ galaxies color, which also reaches a maximum at $`Ks`$ 18–19 and then it becomes bluer, while the median $`BK`$ color gets significantly bluer at brighter ($`K17`$) magnitudes (Gardner et al. 1993).
Our wide-field survey allows us to select a statistically significant and complete sample of EROs which can provide stringent constraints on the density of high-$`z`$ ellipticals (see Sect. 1). For this reason, EROs are defined as objects with $`RKs5`$ because this corresponds to select passively evolving ellipticals at $`z\stackrel{>}{}0.9`$ (see Fig. 13). Table 3 shows the results of the selection for different magnitude and color thresholds. Among the others, the threshold $`RKs5.3`$ is used because it corresponds to the selection of $`z\stackrel{>}{}1`$ elliptical galaxies (see Fig. 13). We detected 279 EROs with $`Ks18.8`$ from the whole area and 119 EROs with $`18.8<Ks19.2`$ in the deeper area, yielding a total sample of 398 objects (see Fig. 5 and Table 3). This is by far the largest sample of EROs obtained to date. A small but complete sample of EROs with $`RKs7`$ has also been selected, and we estimate for the first time their surface density to be $`0.01`$ arcmin<sup>-2</sup> at $`Ks19`$.
A comparison of the surface densities of the EROs in our sample with those in the Thompson et al. (1999) survey can be directly done after taking into account the different filters used in the two surveys. For the $`K^{}`$ filter used by Thompson et al. (1999), $`Ks`$ $``$ $`K^{}`$ \- 0.2 (adopting $`HK=1`$), and therefore their limit at $`K^{}19`$ corresponds to $`Ks18.8`$ which is the shallower limit of our survey, while their $`RK^{}`$ color is bluer than our $`RKs`$ by about 0.1 magnitudes for the redder objects (Thompson, private communication). At the level of $`Ks18.8`$ we find a density of $`0.042\pm 0.008`$ arcmin<sup>-2</sup> for EROs with $`RKs6`$, to be compared with the value of 0.039$`\pm `$0.016 that they find (these errors are poissonian). Thus, the two surface densities are in excellent agreement with each other. We also verified that the average $`RKs`$ color of all our objects with $`17.8<Ks<18.8`$ ($`RKs=3.70\pm 0.03`$, determined with a Kaplan-Meier estimator), is in good agreement with the average $`RK^{}=3.73\pm 0.04`$ in $`18<K^{}<19`$ by Thompson et al. (2000).
## 5 The angular correlation functions
Statistical measurements of the clustering of faint galaxies are important for studying the evolution of galaxies and the formation of structures in the Universe. In fact, the amplitude of clustering in 2D space is a useful probe of the underlying 3D structure (e.g. Connolly et al. connolly (1998), Efstathiou et al. 1991, Magliocchetti & Maddox 1999). The clustering of galaxies on the sky has been studied extensively especially in the optical, but also in the near-infrared (e.g. Roche et al. 1998 and 1999, Postman et al. 1998, Baugh et al. 1996b). Our survey, as noted before, is the widest at the limits of $`Ks19`$. It is therefore interesting to estimate the clustering of our sample of galaxies.
### 5.1 Calculation technique
The angular two-point correlation function $`w(\theta )`$ is defined as the excess probability (over a poissonian distribution) of finding galaxies separated by the apparent distance $`\theta `$:
$$dP=N^2[1+w(\theta )]d\mathrm{\Omega }_1d\mathrm{\Omega }_2$$
(1)
where N is the mean density per steradian (Groth & Peebles, 1977).
Several methods for estimating $`w(\theta )`$ from a set of object positions have been proposed and used, but the most bias–free and suitable for faint galaxies samples resulted to be the Landy & Szalay technique (Landy & Szalay 1993, see also Kerscher et al. 2000). This technique (adopted for the calculations in this paper) consists in deriving the counts of objects binned in logarithmic distance intervals, for the data-data sample $`[DD]`$, the data-random sample $`[DR]`$ and the random–random sample $`[RR]`$. These counts have to be normalized, i.e. divided for the total number of couples in each of the 3 samples. From them we can estimate $`w_\mathrm{b}(\theta )`$ as:
$$w_\mathrm{b}(\theta )=\frac{[DD]2[DR]+[RR]}{[RR]}$$
(2)
which is biased to lower values with respect to the real correlation function $`w(\theta )`$:
$$w(\theta )=w_\mathrm{b}(\theta )+\sigma ^2$$
(3)
where $`\sigma ^2`$ is the ”integral constraint” (Groth & Peebles, 1977):
$$\sigma ^2=\frac{1}{\mathrm{\Omega }^2}w(\theta )𝑑\mathrm{\Omega }_1𝑑\mathrm{\Omega }_2$$
(4)
Assuming that the angular correlation function $`w(\theta )`$ can be described by a power law of the form $`w(\theta )=A\theta ^\delta `$, then, following Roche et al. (1999), we can extimate the ratio between $`\sigma ^2`$ and the amplitude $`A`$ using the random–random sample:
$$C=\frac{\sigma ^2}{A}=\frac{N_{\mathrm{rr}}(\theta )\theta ^\delta }{N_{\mathrm{rr}}(\theta )}$$
(5)
The amplitude of the real two-point correlation function $`w(\theta )`$ can then be estimated by fitting to the measured $`w_\mathrm{b}(\theta )`$ the function:
$$w_\mathrm{b}(\theta )=A(\theta ^\delta C)$$
(6)
The errors can be estimated, following Baugh et al. (1996b), as:
$$\delta w_\mathrm{b}(\theta )=2\sqrt{(1+w_\mathrm{b}(\theta ))/DD)}$$
(7)
where $`DD`$ is the non normalized histogram of $`[DD]`$. Eq. (7) is equivalent to assuming 2$`\sigma `$ poissonian errors for the correlations, and it gives estimates that are comparable to the errors obtained with the bootstrap technique (Ling, Frenk & Barrow 1986). This is necessary because it is known that, as the counts in the different bins are not completely independent, assuming the 1$`\sigma `$ poissonian errors would result in an underestimate of the true variance of the global parameters of the angular correlation (see Mo et al. 1992).
In case of the presence of a randomly distributed spurious component among the analyzed sample of objects (an example of this case would be a residual stellar component among the galaxy sample), the resulting amplitudes are apparently reduced by a factor $`(1f)^2`$, where $`f`$ is the fraction of the randomly distributed component (see e.g. Roche et al. 1999), and the corresponding correction should be applied.
The random samples used in our analysis were obtained using the pseudo–random number generator routine of the C Library function drand48. Random samples with up to 200 000 objects were used. Typically the number of objects in the random samples were a factor of 100–200 larger than the number of observed objects. The random sample was generated with the same geometrical constraints as the data sample, avoiding for instance to place objects in the regions excluded around the brightest stars.
### 5.2 The clustering of the $`K`$-selected field galaxies
In our analysis a fixed slope of $`\delta =`$ 0.8 was assumed, as this is consistent with the typical slopes measured in both faint and bright surveys (e.g. Baugh et al. 1996b, Roche et al. 1996, Maddox et al. 1990), and because it gives us the possibility to directly compare our results with the published ones that are typically obtained adopting such a slope. The factor $`C`$ was estimated (with Eq. (5)) for both the whole and the deeper areas, turning out to be 4.55 and 5.16 respectively (the angles are expressed in degrees, if not differently stated). In Fig. 6 the observed, bias corrected, two-point correlation functions $`w(\theta )`$ are shown; the bins have a constant logarithmic width ($`\mathrm{\Delta }\mathrm{log}\theta =0.2`$), with the bin centers ranging from 3.6$`\mathrm{}`$ to 15$`\mathrm{}`$.
We clearly detect a positive correlation signal for our sample with an angular dependence broadly consistent with the adopted slope $`\delta =`$ 0.8, even if the measurements show some deviations, in particular for the brightest samples. A few cluster candidates are present in our survey. These possible clusters include galaxies with $`RKs4.5`$, and are therefore expected to be at $`z\stackrel{<}{}0.6`$. A detailed analysis of the cluster candidates will be given in a forthcoming paper. For the purpose of the present work, we tested that the measured clustering amplitudes are stable in case of removal of the galaxies of the most evident cluster from the sample. However, the presence of such clusters, most of which happen to be in the shallower area, could partly explain the observed deviations from the fitted $`w(\theta )=A\theta ^{0.8}`$ power laws for the three brightest samples.
The derived clustering amplitudes are presented in Table 4. The amplitude errors are obtained from the fit assuming Eq. (7). No correction for the stellar contamination was applied. In Fig. 7 the clustering amplitudes of our samples are compared with other published measurements. The data are shown together with a number of PLE models with different clustering evolutions, which are described in detail in Roche et al. (1998, 1999). Our measurements are in good agreement with both the models and the previous estimates of Roche et al. (1999), except for the point with limiting magnitude $`Ks=18.0`$, which is however the most uncertain of our data points.
As a check it was verified that the correlations of the stellar sample are consistent with zero, within the measured errors, at all the scales. This is a confirmation that the stars are homogeneously distributed on the field (as they should be) and the seeing variations across the area did not cause a detectable bias in our classification.
### 5.3 The clustering of the extremely red objects
The large sample of EROs derived from our survey allowed us for the first time to estimate their clustering properties. Even a simple visual inspection of the sky distribution of the objects with $`RKs5`$ (see Fig. 8) shows that the EROs have a very inhomogeneous distribution.
The results of the quantitative analysis of the clustering are shown in Fig. 9, where the observed, bias–corrected angular correlations $`w(\theta )`$ of the objects with $`RKs`$$``$5 are displayed. A strong clustering is indeed present at all the scales that could be measured, and its amplitudes (Table 5) are about an order of magnitude higher than the ones of the field population at the same $`Ks`$ limits. The correlations are very well fitted by a $`\delta =0.8`$ power law. No attempt was made to correct the amplitudes for the stellar contamination (see Sect. 5.1), and we stress that such corrections would increase them. Adopting the errors derived from the fits, our detections are significant at more than 7$`\sigma `$ level for the samples with $`RKs5`$ and $`Ks`$ limits equal to or fainter than $`Ks=18.5`$.
The amplitudes shown in Fig. 5 suggest a possible trend of decreasing strength of the clustering for fainter EROs: the $`Ks`$19.2 EROs are less clustered than the ones with $`Ks`$18.5 and the difference is significant at $`2.7\sigma `$, based on the derived errors. The significance of this effect would, however, decrease if the contamination by randomly distributed stars increases towards the limit of our survey. Although difficult to be quantified, this is likely to happen because, differently from the brightest EROs, only a small fraction of the EROs fainter than $`Ks`$ = 18.5 could be morphologically classified (see Fig. 5).
Defining redder thresholds drastically reduces the number of EROs and it is not possible to estimate with sufficient accuracy how the amplitudes change for objects with even redder $`RKs`$ colors. We could only verify that the sample of EROs with $`RKs5.3`$ has clustering amplitudes consistent with those of the $`RKs5`$ samples (see Table 5). To measure the clustering of the $`RKs`$6 EROs, an area at least 10 times larger than ours (i.e. $``$2 square degrees) at $`K=19`$ would be needed, assuming that their clustering amplitudes are similar to those of the EROs with $`RKs5`$.
Finally, it was studied if and how the clustering amplitude changes as a function of $`RKs`$ for the $`Ks`$18.8 sample (see Fig. 10). A clear increase of $`A`$ with $`RKs`$ is present for colors $`RKs`$$``$3.5, while the $`RKs`$3 sample has an amplitude that is consistent with that of the whole sample of field galaxies. The variation of $`A`$ can be described with a power law in the range of $`3RKs5.7`$. Previous efforts to disentangle the clustering properties of the red and blue populations in faint $`K`$–selected samples probably failed because the ERO population was not sufficiently sampled. For instance, Kummel & Wagner (2000) did not find significant differences in the clustering of objects with color bluer or redder than $`RKs`$=3.49 for their $`K<`$17 sample. This is not surprising since at K$`<`$17 the ERO population is almost absent (see Table 3 and Fig. 3).
To check for the stability of these results, possible systematics that could produce a bias in our work were analyzed. First af all, as the clustering of our $`Ks`$–selected galaxies is in good agreement with the literature data (Fig. 7), we can exclude the presence of measurable biases coming from the selection of the sample.
Regarding the color measuraments, since EROs are the tail of objects in the $`RKs`$ color distribution, systematic variations of the photometric zeropoints across the area could have the effect of creating artificial ERO overdensities and voids. To exclude this possibility we verified that the blue tail of the $`RKs`$ distribution is homogeneously spread across our survey, with a very low clustering amplitude. In case of zeropoints variations these should produce the same effect in both the tails of the color distribution. Moreover, to test the reality of the large void of EROs clearly seen in the bottom region of our survey (see Fig. 8), the $`RKs`$ color distribution of the galaxies inside and outside this large void were compared by means of a Kolmogorov-Smirnov test, selecting only the galaxies with $`RKs`$4 in both regions. The probability that the two distributions are extracted from the same population is 43%. Thus, the two regions are fully consistent with each other with respect to the color distribution of the blue population. This, together with the fact that no underdensity is present in this region when only the bluer galaxies are considered, shows that the void of EROs should be considered a real feature.
All these tests strongly suggest that the inhomogeneous ERO distribution is a real effect.
## 6 Main implications
### 6.1 On the nature of EROs
The strong clustering signal that we find to increase with the $`RKs`$ threshold and to reach very high values for the EROs is potentially capable of giving insight about the nature of these objects.
The main possible source of objects that may contribute to the ERO population, as discussed in the introduction, are old passively evolving $`z\stackrel{>}{}1`$ ellipticals, dust-reddened starburst galaxies and, in case of unresolved objects, low-mass stars or brown dwarfs. Our field being at high galactic latitude ($`b57\mathrm{°}`$), stars are expected to have no clustering, and to be homogeneously distributed, and certainly not to give the strong clustering signal detected. As for the starburst galaxies, it must be noted that in such galaxies the red colors are mainly driven by the amount of dust extinction and not by the redshift, as in the case of ellipticals (see Fig. 13), and therefore a wide redshift distribution is expected which should dilute their intrinsic clustering. Moreover, it is known that the IRAS–selected galaxies (which are typically star–forming galaxies) have very low intrinsic clustering (e.g. Fisher et al. 1994). We can therefore reasonably conclude that the observed signal is due to the clustering of high redshift ellipticals. This is also suggested by studies of the local universe which have shown that early-type galaxies are much more clustered than late-type galaxies (e.g. Guzzo et al. 1997, Willmer et al. 1998). In this regard the results plotted in Fig. 10 could be qualitatively explained by noting that in selecting redder samples the fraction of early type galaxies increases (the color of local ellipticals is just around $`RKs\stackrel{<}{}3`$) and by assuming that these galaxies are intrinsically more clustered, while such plot would be difficult to understand if mainly driven by the strongly reddened starburst galaxies. These considerations strongly suggest that EROs are mainly composed by $`z\stackrel{>}{}1`$ ellipticals, confirming the previous indications that had been found on this issue.
As the elliptical galaxies are the dominant population of galaxy clusters, we investigated the possibility that the detected clustering of EROs could be the result of a few massive clusters at $`z1`$ present in our field. For example, in the region inside the circle in Fig. 8, a large ERO overdensity is found, that one could suspect to be due to a high-$`z`$ cluster of galaxies. However, Fig. 11 shows that there is no clear color-magnitude sequence among the $`RKs>5`$ objects inside that region, suggesting that they do not all belong to a single cluster. In case of a cluster, even at high–$`z`$, a well defined color-magnitude sequence is in fact generally observed (e.g. Stanford et al. 1998). In the last years a few examples of massive $`z\stackrel{>}{}`$1 clusters of galaxies have been discovered (e.g. Stanford et al. 1997, Rosati et al. 1999), with X ray luminosity of $`10^{44}`$ erg s<sup>-1</sup>. A crude estimate of the number of structures of this sort that could be observed in our survey can be derived by calculating the number of high–$`z`$ clusters with L$`{}_{\mathrm{X}}{}^{}>10^{44}`$ erg s<sup>-1</sup> expected in the volume we are sampling. From the X ray luminosity function of such structures at $`z1`$ (Rosati et al. 2000) we estimate that the expected number of massive clusters in our field in the redshift interval $`0.9<z<2`$ is only $`0.1`$ (for $`\mathrm{\Omega }_0=1`$).
Moreover, the detection of the ERO positive correlation, following a $`\delta =0.8`$ power law on all the scales from 10$`\mathrm{}`$ to 15$`\mathrm{}`$ (corresponding to $``$8 Mpc at $`z1`$) suggests that the clustering signal does not come from a few possible clusters detected in our field, but rather from the whole large scale structure traced by the elliptical galaxies.
### 6.2 Fluctuations of the ERO number density
Our results on the clustering of EROs have important consequences on the problem of estimating the density of high-$`z`$ ellipticals (see Sect. 1).
The existence of an ERO angular correlation with $`\delta =0.8`$ and with a high amplitude implies significant surface density variations around the mean value even for relatively large areas. In the presence of a correlation with amplitude A, the rms fluctuations of the counts around the mean value $`\overline{n}`$ is (see for example Roche et al. 1999):
$$\sigma _{true}^2=\overline{n}(1+\overline{n}AC)$$
(8)
The factor $`C`$ is the same as in Eq. (5) and, by applying Eq. (5) for several areas, it was found that it can be approximated as:
$$C=58\mathrm{Area}^{0.4}$$
(9)
if the area is expressed in arcmin<sup>2</sup> and $`\delta =0.8`$. The validity of such an approximation has been tested for square regions and for areas not larger than the ones of our survey. With Eq. (8) and (9) the expected variations of the ERO number counts can be calculated, once their clustering amplitude is known.
To verify the consistency of this picture, we derived the distribution of the number of EROs (with $`RKs`$$``$5 and $`Ks`$18.8, i.e. those in Fig. 8) that can be recovered in our area by sampling it with a field of view of 5$`\mathrm{}\times `$5$`\mathrm{}`$, which is the typical field of view of a near-infrared imager such as SOFI. In Fig. 12 the observed frequencies of the number of EROs recovered in this counts-in-cell analysis is plotted. As the mean expected number of EROs is about 10, the poissonian fluctuations would be $`\sigma _{\mathrm{poisson}}`$3.2, while fluctuations with $`\sigma =`$5.4 are actually observed. Applying Eq. (8), the measured clustering amplitude $`A=0.013`$ implies $`\sigma =5.55`$, in excellent agreement with the measured $`\sigma `$ value.
We also note that the distribution of the numbers of EROs in Fig. 12 is not only asymmetric, but also very broad, ranging from $`N`$=0 to $`N`$=30. In 29% of the cases the number of EROs recovered is $`N`$5, corresponding to a surface density half of the real one, while only in 19% of the cases the observed number is $`N`$15. This shows that, on average, it is more probable to underestimate the real surface density of these objects. This is a clear property of the sky distribution that we observe, as the voids extend on a large fraction of the surveyed area. These results show how strong the effects of the field-to-field variations are in the estimate of the sky surface density of EROs. In this respect, it should be noted here that all previous estimates of the number density of high–$`z`$ ellipticals were based on surveys made with small fields of view, typically ranging from 1 arcmin<sup>2</sup> in the case of the NICMOS HDF-S (Benitez et al. 1999) to 60 arcmin<sup>2</sup> in the case of Barger et al. (2000).
### 6.3 Implications for the evolution of elliptical galaxies
The selection of galaxies with colors $`RKs>5`$ can be used to search for elliptical galaxies at $`z>0.9`$ (see Fig. 13), and to study their evolution by comparing their observed surface densities with those expected from PLE or hierarchical models of massive galaxy evolution. In this respect, very discrepant results have been obtained so far, making the formation of spheroids one of the most controversial problems of galaxy evolution (see the Introduction).
Our results on the ERO clustering clearly show that for such a comparison to be reliable, both a wide field survey (resulting in a large number of EROs) and a consistent estimate of their surface density fluctuations are necessary before reaching solid conclusions on the evolution of elliptical galaxies.
In this section, with the main aim to show the effect of the increased uncertanties due to the clustering, a preliminary comparison is presented between the sky density of EROs observed in our survey (Table 3) and the predictions of an extreme PLE model similar to that used by Zepf (1997). In this model, ellipticals formed at $`z_\mathrm{f}`$=5 and their star formation rate ($`SFR`$) is characterized by an exponentially decaying burst with $`SFRexp(t/\tau )`$, with $`\tau =0.1`$ Gyr. Adopting the Markze et al. (1994) local luminosity function of ellipticals, and the Bruzual & Charlot (1997) models with solar metallicity and Salpeter IMF, the expected surface densities of passively evolving ellipticals with $`RKs6`$ was calculated for different limiting $`Ks`$ magnitudes.
Fig. 14 shows the comparison between the expected and the observed densities of EROs with $`RKs6`$ (such a color threshold should select passively evolving galaxies at $`z\stackrel{>}{}1.3`$). For each data point we show three different error bars, which are actually the region of confidence in the poissonian case (at 1$`\sigma `$) and in the true (i.e. clustering corrected) case (at 1$`\sigma `$ and 2$`\sigma `$). Such confidence regions have been estimated, following the prescriptions of Eq. (8), by finding the range of values for the true average counts $`\overline{n}`$ for which the observed $`N`$ would represent a deviation of the required number of $`\sigma `$ from the real density. In other words such ranges are defined from the two solutions of the equation:
$$\alpha ^2=\frac{(\overline{n}N)^2}{\overline{n}(1+\overline{n}AC)}$$
(10)
where $`\alpha `$ is the number of $`\sigma `$ considered. The amplitudes of the angular correlation function used for the $`RKs6`$ EROs are those derived for the $`RKs5`$ EROs, which is likely to be a conservative assumption as the amplitudes of redder samples should be higher, as suggested by Fig. 10.
Fig. 14 shows that the observed EROs densities are indeed lower than the predictions of this particular PLE model. However, even for the most deviant point, this PLE model can be rejected at only the 2.5$`\sigma `$ and 2.3$`\sigma `$ level for $`\mathrm{\Omega }_0=1`$ and $`\mathrm{\Omega }_0=0.1`$ respectively, if we use the “correct” error bars. Note that the different points plotted in that figure are not statistically independent because they are partially obtained with the same objects (they are cumulative values).
It should be recalled here that the observed ERO densities plotted in Fig. 14 are an overestimate of the true density of the ellipticals because of the contamination by dusty starbursts (see Cimatti et al. 1998, 1999; Dey et al. 1999; Smail et al. 1999) and by field low-mass stars. The fraction of dusty starbursts in complete ERO samples is not known yet, as discussed in the introduction, but our results show that they should not be the dominant population. For instance, assuming that the fraction of dusty starbursts and low-mass stars is 20% and 10% of the ERO population respectively, this would decrease the observed densities plotted in Fig. 14 accordingly, but it would also increase by a factor of 2 the clustering amplitudes of the high-$`z`$ ellipticals (see Sect. 5.1), and hence the error bars related to those points. As a consequence, the statistical significance of the difference between data and model in this case would be only at the 2$`\sigma `$ level.
It is relevant to mention that the predictions of the PLE models depend very strongly on many parameters that have to be adopted a priori such as $`H_0`$, $`\mathrm{\Omega }_0`$, the local LF of ellipticals (uncertain by up to a factor of 2), the redshift of formation $`z_\mathrm{f}`$, the history of star formation, the metallicity, the IMF, the spectral synthesis models. For instance, even just a decrease of $`z_\mathrm{f}`$, or a small residual star formation at z $``$ 1.5 (Menanteau et al. 98, Jimenez et al. 99), would decrease the predicted numbers of EROs making them more consistent with our data. We therefore conclude that it seems premature to reject even extreme PLE models at a high level of statistical significance on the basis of these data.
A preliminary comparison of our results can be made with some aspects of the hierarchical models of galaxy formation. First of all, our findings could qualitatively fit into the predictions of such models, where high-$`z`$ ellipticals should be very clustered (Kauffmann et al. 1999) because they are expected to be linked to the most massive dark matter haloes which are strongly clustered at high–$`z`$. The indication (marginally significant at $`2.7\sigma `$ level) of a decrease of the clustering amplitude of the EROs with the $`Ks`$ magnitude (see Sect. 5.3), if mainly due to the mass of the galaxies, could also fit well in this framework because smaller objects should be connected to smaller dark matter haloes which are expected to be less correlated. On the other hand, our results seem to conflict with the predictions made by Kauffmann & Charlot (1998) on the fraction of $`K`$-selected galaxies with $`K19`$. In fact, the fraction of galaxies observed to have color $`RKs5.3`$ (which corresponds to the selection of $`z\stackrel{>}{}1`$ ellipticals) is about 7% of the total in our survey (see Table 3), to be compared with the 2–3% of $`z>1`$ galaxies with K$``$19 expected in the Kauffmann & Charlot (1998) hierarchical model. This result on the fraction of $`z\stackrel{>}{}1`$ galaxies in our $`K`$-selected sample broadly agree with the finding of Eisenhardt et al. (2000).
## 7 Summary
The main results of this work are:
* We have presented a survey which covers 701 arcmin<sup>2</sup> and is 85% complete to $`Ks18.8`$ over the whole area and to $`Ks19.2`$ over 447.5 arcmin<sup>2</sup>; the R-band limit is $`R26.2`$ at the 3$`\sigma `$ level.
* The observed galaxy counts are derived over the largest area so far published in the range of $`18Ks19.2`$. Such counts are in excellent agreement with other published data.
* The median $`RKs`$ color of field galaxies increases by 0.5 mags from $`Ks=16.5`$ to $`Ks=18`$, and it remains constant to $`Ks=19.2`$.
* A sample of 398 EROs has been selected. This sample is the largest published to date and is characterized by an area larger by about four times than previous surveys. The ERO counts and surface densities have been derived for several color thresholds and $`Ks`$ limiting magnitudes. In particular, we find $`0.67\pm 0.03`$ (poissonian) EROs arcmin<sup>-2</sup> with $`RKs5`$ and $`0.10\pm 0.01`$ EROs arcmin<sup>-2</sup> with $`RKs6`$ at $`Ks19.2`$.
* The surface density of EROs with $`RKs7`$ has been estimated for the first time to be of the order of $`0.01`$ arcmin<sup>-2</sup> at $`Ks19`$.
* The angular correlation function of field galaxies, fitted with a fixed slope $`\delta =0.8`$, has an amplitude $`A(1\mathrm{°})0.0015`$ at $`18.5Ks19.2`$, in agreement with previous measurements.
* For the first time, we detected the clustering of EROs, with an amplitude $`A(1\mathrm{°})0.015`$ for the objects with $`RKs5`$, in the range $`18.5Ks19.2`$) which is about a factor of ten higher than that of field galaxies. The ERO two point correlations are very well fitted by a $`\delta =0.8`$ power law.
* The clustering amplitude of the galaxies increases with the $`RKs`$ color threshold following the relation $`logA0.43(RKs)`$, for $`3RKs5.7`$ at $`Ks18.8`$.
* The strong clustering of EROs is shown to be a direct evidence that a large fraction of these objects are indeed high–$`z`$ ellipticals. Our result is therefore the first detection of the large scale structure traced by the elliptical galaxies at $`z1`$.
* The ERO clustering explains the conflicting results obtained so far on the density of high-$`z`$ ellipticals in terms of strong field-to-field variations affecting the surveys based on small fields of view (e.g. $`5\times 5`$ arcmin).
* Taking into account the clustering of EROs, even the predictions of extreme PLE models for the comoving density of high–$`z`$ ellipticals cannot be rejected at much more than 2$`\sigma `$ significance level.
###### Acknowledgements.
We would like to thank Nathan Roche for providing his models in digital form, Gustavo Bruzual and Stephane Charlot for their synthetic stellar population models. We also thank Leonardo Vanzi for his assistance during the NTT observations and the anonymous referee for useful comments. LP acknowledges the support of CNAA during the realization of this project.
|
warning/0005/gr-qc0005008.html
|
ar5iv
|
text
|
# Wormhole with Quantum Throat
## I Introduction
By definition a wormhole (WH) is a bridge connecting two asymptotically flat regions. Usually such construction is a classical object and should satisfy to Einstein equations (see for more detailed introduction to this field). The topology of the 4D WH is $`R_1\times R_2\times S^2`$, where $`R_1`$ is the time, $`R_2`$ is the radial coordinate and $`S^2`$ is the cross section of the WH. A space region near to the minimal cross section $`S_{min}^2`$ is called as a throat. The basic problem for existing of the WH is concentrated on the throat: in this region a matter violates the so-called null energy condition. In this paper we offer a model of the WH in which the throat is a set of quantum handles (wormholes) in the presence of the strong electric field. These quantum handles can be considered as quantum WH’s of a spacetime foam with separated mouthes (see Fig.1). We remind that the hypothesized spacetime foam , is a set of quantum wormholes (handles) appearing in the spacetime on the Planck scale level ($`l_{Pl}10^{33}cm`$). For the macroscopic observer these quantum fluctuations are smoothed and we have an ordinary smooth manifold with the metric submitting to Einstein equations.
The presence of the spacetime foam with virtual wormholes (VWH) which have separated mouthes gives a physical effect: VWH’s can entrap a part of the electric flux lines and in spite of the fact that at the infinity $`|e|/m>1`$ we will have a non-singular spacetime ($`e`$ and $`m`$ are the charge and mass registered at the infinity).
The mechanism for creating the WH with quantum throat can be offered by the following way:
1. The VWH’s connect two regions with the strong electric field by such a way that the electric flux lines leak through VWH’s from one region to another one , see Fig.2.
2. Each VWH is the solutions of the 5D Einstein equations with $`G_{5t}0`$.
3. As usually the $`G_{5t}`$ metric component can be considered as the 4D electric field which joins the electric flux lines of the above-mentioned two 4D regions.
4. For this model the whole spacetime is 5D one and outside of the throat the $`G_{55}`$ is non-dynamical variable and we have the 5D Kaluza-Klein theory in its initial interpretation ($`G_{55}=1`$ and 5D gravity is equivalent to Einstein-Maxwell theory), inside of the throat the $`G_{55}`$ is dynamical variable and we have the ordinary Kaluza-Klein gravity with 4D metric + electromagnetic and scalar fields. Splitting off the 5D dimension takes place on the event horizon. The matching conditions of the 4D and 5D fields on the event horizon is discussed in Ref.. As this composite WH is the classical solution of the 5D Kaluza-Klein equations a probability for the quantum creation of such VWH is not zero.
## II Qualitative description of the model
The model of the VWH is presented on the Fig.(2).
On the Fig.(3) are presented the two parts of the WH with the quantum throat.
On the Fig.3A is presented the part of the WH with incoming force lines of the electric field and respectively on the Fig.3B the part with outcoming force lines. For the 4D observer each mouth is like to moving (+/-) electric charge. But this movement is stochastic one and as a consequence we have $`magnetic`$ $`field0`$ where $``$ is the stochastical averaging.
In Ref. is shown that such composite WH has a spin-like structure. The cause of this is following. The metric describing the 5D throat is
$`ds^2`$ $`=`$ $`\eta _{AB}\omega ^A\omega ^B=`$ (2)
$`{\displaystyle \frac{r_0^2}{\mathrm{\Delta }(r)}}(d\chi \omega (r)dt)^2+\mathrm{\Delta }(r)dt^2dr^2a(r)\left(d\theta ^2+\mathrm{sin}\theta ^2d\phi ^2\right),`$
$`a`$ $`=`$ $`r_0^2+r^2,\mathrm{\Delta }=\pm {\displaystyle \frac{2r_0}{q}}{\displaystyle \frac{r^2+r_0^2}{r^2r_0^2}},\omega =\pm {\displaystyle \frac{4r_0^2}{q}}{\displaystyle \frac{r}{r^2r_0^2}}.`$ (3)
where $`\chi `$ is the 5<sup>th</sup> extra coordinate; $`\eta _{AB}=(\pm ,,,,)`$, $`A,B=0,1,2,3,5`$; $`r,\theta ,\phi `$ are the $`3D`$ polar coordinates; here $`r_0>0`$ and $`q`$ are some constants. We see that the signs of the $`\eta _{55}`$ and $`\eta _{00}`$ are not defined. We remark that this 5D metric is located behind the event horizon therefore the 4D observer is not able to determine the signs of the $`\eta _{55}`$ and $`\eta _{00}`$ . Moreover this 5D metric (2) can fluctuate between these two possibilities. Hence the external 4D observer is forced to describe such composite WH by means of a spinor.
This allows us to do the following assumption:
###### Assumption 1
The stochastical polarized spacetime foam approximately can be determinated by some effective field: a spinor field $`\psi `$.
In this case a density $`ϵ`$ of the VWH is
$$ϵ=|\psi |^2=\stackrel{~}{\psi }\psi $$
(4)
where ($`\stackrel{~}{}`$) means the transposition and a density of an effective electric charge $`\rho `$ (the mouthes of VWH entrapped the electric force lines) is
$$\rho =e\stackrel{~}{\psi }\psi $$
(5)
where $`e`$ is some charge.
## III Exact description of the model
As the hypothesized spacetime foam is a consequence of quantum gravity all our fields (metric $`g_{\mu \nu }`$, electromagnetic filed $`A_\mu `$ and spinor field $`\psi `$) in this model should be quantized fields. The interaction between these fields is very strong and we can not use the Feynmann diagram technique. For the quantization of this model we will use the Heisenberg quantization method which he applied for the non-linear spinor field . The essence of this method consists in that the classical fields $`f(x^\mu )`$ in field equations are excanched on the field operators $`f(x^\mu )\widehat{f}(x^\mu )`$. In this case we have differential equations for operators. Certainly it is not clear what is a solution of such differential equations. In fact Heisenberg has shown that differential equations for the operator field is equivalent to some infinite set of differential equations for Green functions (for the small coupling constant this is Dayson-Schwinger equations system).
Following this way we write differential equations for the gravitational + electromagnetic fields in the presence of the spacetime foam $`(\psi )`$ as follows
$`\widehat{R}_{\mu \nu }{\displaystyle \frac{1}{2}}\widehat{g}_{\mu \nu }\widehat{R}`$ $`=`$ $`\widehat{T}_{\mu \nu },`$ (6)
$`\left(i\widehat{\gamma }^\mu _\mu +e\widehat{A}_\mu {\displaystyle \frac{i}{4}}\widehat{\omega }_{\overline{a}\overline{b}\mu }\widehat{\gamma }^\mu \widehat{\gamma }^{[\overline{a}}\widehat{\gamma }^{\overline{b}]}m\right)\widehat{\psi }`$ $`=`$ $`0,`$ (7)
$`D_\nu \widehat{F}^{\mu \nu }`$ $`=`$ $`4\pi e\left(\widehat{\overline{\psi }}\widehat{\gamma }^\mu \widehat{\psi }\right)`$ (8)
where $`\widehat{R}_{\mu \nu }`$ is the operator of Ricci tensor; $`\widehat{T}_{\mu \nu }`$ is the operator of the sum of the energy-momentum tensor the spinor and Maxwell fields; $`\widehat{g}_{\mu \nu }`$ is the operator of the gravitational field; $`\widehat{\gamma }^\mu =\widehat{h}_{\overline{a}}^\mu \gamma ^{\overline{a}}`$ is the operator of the Dirac matrices; $`\widehat{g}^{\mu \nu }=(1/2)(\widehat{\gamma }^\mu \widehat{\gamma }^\nu +\widehat{\gamma }^\nu \widehat{\gamma }^\mu )`$, $`[]`$ means the antisymmetrization; $`\gamma ^{\overline{a}}`$ ($`\overline{a}`$ is the vier-bein index) is the usual $`\gamma `$-matrices
$$\gamma ^{\overline{0}}=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right);\gamma ^{\overline{i}}=\left(\begin{array}{cc}0& \sigma ^i\\ \sigma ^i& 0\end{array}\right);i=1,2,3$$
(9)
where $`\sigma ^i`$ are the Pauli matrices; $`\widehat{h}_{\overline{a}}^\mu `$ is the vier-bein operator; Greek indexes are the spacetime indexes; Latin indexes with the bar are vier-bein indexes; $`\widehat{F}_{\mu \nu }=_\mu \widehat{A}_\nu _\nu \widehat{A}_\mu `$ is the operator of the Maxwell tensor of the electromagnetic field; $`\widehat{A}_\mu `$ is the operator of the potential; $`m`$ and $`e`$ are some constants. Certainly this equation system is hopelessly complicated, and are impossible to find exact solutions.
We will consider equations for average values $`g_{\mu \nu }`$, $`A_\mu `$, $`\psi `$ and so on. In the first approximation we suppose that
$$\widehat{f}(x)f(\widehat{x}),$$
(10)
where $`\widehat{f}`$ can be $`\widehat{R}_{\mu \nu }`$, $`\widehat{\omega }_{\overline{a}\overline{b}\mu }`$ and so on; $`\widehat{x}`$ can be $`\widehat{g}_{\mu \nu }`$, $`\widehat{\psi }`$, $`\widehat{A}_\mu `$ and so on. In this case we have the classical system of the Einstein-Dirac-Maxwell equations, i.e. now the system (6)-(8) is without ($`\widehat{}`$).
Fot our model we use the following ansatz: for the spherically symmetric metric
$$ds^2=e^{2\nu (r)}\mathrm{\Delta }(r)dt^2\frac{dr^2}{\mathrm{\Delta }(r)}r^2\left(d\theta ^2+\mathrm{sin}^2d\phi ^2\right),$$
(11)
for the electromagnetic potential
$$A_\mu =(\varphi ,0,0,0),$$
(12)
for the spinor field
$$\stackrel{~}{\psi }=e^{i\omega t}\frac{e^{\nu /2}}{r\mathrm{\Delta }^{1/4}}(f,0,ig\mathrm{cos}\theta ,ig\mathrm{sin}\theta e^{i\phi }).$$
(13)
The following is very important for us: the ansatz (13) for the spinor field $`\psi `$ has the $`T_{t\phi }`$ component of the energy-momentum tensor and the $`J^\phi =4\pi e(\overline{\psi }\gamma ^\phi \psi )`$ component of the current. Let we remind that $`\psi `$ determines the stochastical gas of the VWH’s which can not have a preferred direction in the spacetime. This means that after substitution expression (11)-(13) into field equations they should be averaged by the spin direction of the ansatz (13)<sup>*</sup><sup>*</sup>*another words the averaging $``$ in the expression (10) is not only quantum but stochastical, too.. After this averaging we have $`T_{t\phi }=0`$ and $`J^\phi =0`$ and we have the following equations system describing our spherically symmetric spacetime
$`f^{}\sqrt{\mathrm{\Delta }}`$ $`=`$ $`{\displaystyle \frac{f}{r}}g\left(\left(\omega e\varphi \right){\displaystyle \frac{e^\nu }{\sqrt{\mathrm{\Delta }}}}+m\right),`$ (14)
$`g^{}\sqrt{\mathrm{\Delta }}`$ $`=`$ $`f\left(\left(\omega e\varphi \right){\displaystyle \frac{e^\nu }{\sqrt{\mathrm{\Delta }}}}m\right){\displaystyle \frac{g}{r}},`$ (15)
$`r\mathrm{\Delta }^{}`$ $`=`$ $`1\mathrm{\Delta }\kappa {\displaystyle \frac{e^{2\nu }}{\mathrm{\Delta }}}\left(\omega e\varphi \right)\left(f^2+g^2\right)r^2e^{2\nu }\varphi _{}^{}{}_{}{}^{2}`$ (16)
$`r\mathrm{\Delta }\nu ^{}`$ $`=`$ $`\kappa {\displaystyle \frac{e^{2\nu }}{\mathrm{\Delta }}}\left(\omega e\varphi \right)\left(f^2+g^2\right)\kappa {\displaystyle \frac{e^\nu }{r\sqrt{\mathrm{\Delta }}}}fg{\displaystyle \frac{\kappa }{2}}m{\displaystyle \frac{e^\nu }{\sqrt{\mathrm{\Delta }}}}\left(f^2g^2\right),`$ (17)
$`r^2\mathrm{\Delta }\varphi ^{\prime \prime }`$ $`=`$ $`8\pi e\left(f^2+g^2\right)\left(2r\mathrm{\Delta }r^2\mathrm{\Delta }\nu ^{}\right)\varphi ^{}`$ (18)
where $`\kappa `$ is some constant. This equations system was investigated in and result is the following. A particle-like solution exists which has the following expansions near $`r=0`$
$`f(r)`$ $`=`$ $`f_1r+𝒪(r^2),g(r)=𝒪(r^2),`$ (19)
$`\mathrm{\Delta }(r)`$ $`=`$ $`1+𝒪(r^2),\nu (r)=𝒪(r^2),\varphi (r)=𝒪(r^2)`$ (20)
and the following asymptotical behaviour
$`\mathrm{\Delta }(r)`$ $``$ $`1{\displaystyle \frac{2m_{\mathrm{}}}{r}}+{\displaystyle \frac{(2e_{\mathrm{}})^2}{r^2}},\nu (r)const,`$ (21)
$`\varphi (r)`$ $``$ $`{\displaystyle \frac{2e_{\mathrm{}}}{r}},`$ (22)
$`f`$ $``$ $`f_0e^{\alpha r},gg_0e^{\alpha r},{\displaystyle \frac{f_0}{g_0}}=\sqrt{{\displaystyle \frac{m_{\mathrm{}}+\omega }{m_{\mathrm{}}\omega }}},\alpha ^2=m_{\mathrm{}}^2\omega ^2.`$ (23)
where $`m_{\mathrm{}}`$ is the mass for the observer at infinity and $`2e_{\mathrm{}}`$ is the charge of this solution. Our interpretation of this solution is presented on the Fig.(4).
The solution exists for $`(|e_{\mathrm{}}|/m_{\mathrm{}})>1`$ and $`(|e_{\mathrm{}}|/m_{\mathrm{}})<1`$it depends on the mass $`m`$ but for us is essential the first case with $`(|e_{\mathrm{}}|/m_{\mathrm{}})>1`$. In this case the classical Einstein-Maxwell theory leads to the “naked” singularity. The presence of the spacetime foam drastically changes this result: the appearance of the VWH’s can prevent the formation of the “naked” singularuty in the Reissner-Nordström solution with $`|e|/m>1`$.
## IV Conclusions
Thus our model is based on the following assumptions:
* in basic the quantum (virtual) WHs of the spacetime foam have the 5D throat.
* each quantum WH has a spin-like structure,
* the spacetime foam effectively can be described with the help of a spinor field.
As the consequence we have the result that the strong electric field separates the VWH’s of the spacetime foam by such a way that they can prevent a singularity in the Reissner-Nordström solution (with $`|e|/m>1`$) on account of the formation of VWHs.
In the spirit of the Einstein idea that the right-hand of gravitational equations should be zero we can note that this model of the WH with quantum throat is the vacuum model since the gauge fields can be considered as components of the metric in some multidimensional Kaluza-Klein gravity.
## V Acknowledgment
I am grateful for financial support from the Georg Forster Research Fellowship from the Alexander von Humboldt Foundation and H.-J. Schmidt for an invitation to Potsdam University.
|
warning/0005/math0005041.html
|
ar5iv
|
text
|
# Polar Varieties and Efficient Real Elimination1
## 1 Introduction
The core of this paper consists in the exhibition of a system of canonical equations which describe locally the generic polar varieties of a given semialgebraic complete intersection manifold $`S_0`$ contained in the real $`n`$–dimensional affine space $`\mathrm{IR}^n`$. This purely mathematical description of the polar varieties allows the design of a new type of efficient algorithm (with intrinsic complexity bounds), which computes, in case that $`S_0`$ is smooth and compact, at least one representative point for each connected component of $`S_0`$ (the algorithm returns each such point in a suitable symbolic codification). This new algorithm (and, in particular, its complexity) is the main practical outcome of the present paper. Let us now briefly describe our results.
Suppose that the real variety $`S_0`$ is compact and given by polynomial equations of the following form:
$$f_1(X_1,\mathrm{},X_n)=\mathrm{}=f_p(X_1,\mathrm{},X_n)=0,$$
where $`p,n\mathrm{IN},pn`$ and $`f_1,\mathrm{},f_p`$ belong to the polynomial ring $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ in the indeterminates $`X_1,\mathrm{},X_n`$ over the rational numbers Q . Let $`d`$ be a given natural number and assume that for $`1kp`$ the total degree $`\mathrm{deg}f_k`$ of the polynomial $`f_k`$ is bounded by $`d`$. Moreover, we suppose that the polynomials $`f_1,\mathrm{},f_p`$ form a regular sequence in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ and that they are given by a division-free arithmetic circuit of size $`L`$ that evaluates them in any given point of the real (or complex) $`n`$–dimensional affine space $`\mathrm{IR}^n`$ (or $`\text{C}\text{ }^n`$). Further, we assume that the Jacobian $`J(f_1,\mathrm{},f_p)`$ of the equation system $`f_1=\mathrm{}=f_p=0`$ has maximal rank in any point of $`S_0`$ (thus, implicitly, we assume that $`S_0`$ is smooth). Let $`W_0:=V(f_1,\mathrm{},f_p)`$ denote the (complex) algebraic variety defined by the polynomials $`f_1,\mathrm{},f_p`$ in the affine space $`\text{C}\text{ }^n`$. We denote the singular locus of $`W_0`$ by $`SingW_0`$.
Moreover, let us suppose that the variables $`X_1,\mathrm{},X_n`$ are in generic position with respect to the equation system $`f_1,\mathrm{},f_p`$. For $`1inp`$ let $`W_i`$ be the $`i`$–th formal (complex) polar variety associated with $`W_0`$ (and the variables $`X_{p+i},\mathrm{},X_n`$).
Further, let us denote the real counterpart of $`W_i`$ by $`S_i:=W_i\mathrm{IR}^n`$. We call $`S_i`$ the $`i`$–th formal real polar variety associated with the real semialgebraic variety $`S_0`$ (and the variables $`X_{p+i},\mathrm{},X_n`$) . It turns out that the (locally) closed sets $`W_iSingW_0`$ (resp. $`S_i`$) are either empty or complex (resp. real) smooth manifolds of dimension $`n(p+i)`$. Moreover, for $`1inp`$, one sees easily that
$$\stackrel{~}{W_i}:=\overline{W_iSingW_0}$$
is the $`i`$–th polar variety (in the usual sense) associated with $`W_0`$ and the variables $`X_{p+i},\mathrm{},X_n`$ (here, $`\overline{W_iSingW_0}`$ denotes the Q -Zariski closure in $`\text{C}\text{ }^n`$ of the quasi–affine variety $`W_iSingW_0`$). For a precise definition of the notion of formal polar varieties and of polar varieties in the usual sense we refer to Section 2.
Suppose that the real variety $`S_0`$ is non–empty and satisfies our assumptions. In Theorem 10 of this paper we show that every real polar variety $`S_i=W_i\mathrm{IR}^n,\mathrm{\hspace{0.33em}1}inp,`$ is a non–empty, smooth manifold of dimension $`npi`$ containing at least one point of each connected component of the real variety $`S_0`$. In particular, the real variety $`S_{np}`$ is a finite set containing at least one representative point of each connected component of $`S_0`$.
Under the same assumptions we show in Theorem 8 that for $`\mathrm{\hspace{0.33em}1}inp`$ the quasi–affine variety $`W_iSingW_0`$ is a locally complete intersection that satisfies the Jacobian criterion. More precisely, the quasi–affine variety $`W_iSingW_0`$ is a smooth manifold of codimension $`p+i`$ that can be described locally by certain regular sequences consisting of the polynomials $`f_1,\mathrm{},f_p`$ and $`i`$ many well–determined $`p`$–minors of the Jacobian $`J(f_1,\mathrm{},f_p)`$ of the $`f_1,\mathrm{},f_p`$. In particular, the quasi–affine variety $`W_{np}SingW_0`$ is zero-dimensional, whence $`\stackrel{~}{W}_{np}=W_{np}SingW_0`$. Thus $`\stackrel{~}{W}_{np}`$ is a zero-dimensional complex variety that contains a representative point of each connected component of the real variety $`S_0`$.
The practical outcome of Theorem 8 and Theorem 10 consists in the design of an efficient algorithm (with intrinsic complexity bounds), which adapts the elimination procedure for complex algebraic varieties developed in and to the real case. Under the additional assumption that for any $`1kp`$, the intermediate ideal $`(f_1,\mathrm{},f_k)`$ generated by $`f_1,\mathrm{},f_k`$ in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ is radical, we shall apply this procedure to the $`p\left(\genfrac{}{}{0pt}{}{n}{p1}\right)`$ well–determined equation systems of Theorem 8, which describe the zero-dimensional algebraic variety $`\stackrel{~}{W}_{np}=W_{np}SingW_0`$ locally. In order to find at least one representative point for every connected component of the real variety $`S_0`$, we have just to run the procedure of and on all these equation systems. Counting arithmetic operations in Q at unit costs, this can be done in sequential time
$$\left(\genfrac{}{}{0pt}{}{n}{p1}\right)L(nd\delta )^{O(1)},$$
where $`\delta `$ is the following geometric invariant of the regular sequence $`f_1,\mathrm{},f_p`$:
$$\delta :=\mathrm{max}\{\mathrm{max}\{\mathrm{deg}\overline{V(f_1,\mathrm{},f_k)SingW_0}|1kp\},$$
$$\mathrm{max}\{\mathrm{deg}\stackrel{~}{W_i}|1inp\}\}$$
(here, $`\mathrm{deg}\overline{V(f_1,\mathrm{},f_k)W_0}`$ and $`\mathrm{deg}\stackrel{~}{W_i}`$ denote the geometric degree in the sense of of the corresponding algebraic varieties).
This is the content of Theorem 11 below. For any $`1kp`$ and any $`1inp`$ the quantity $`\delta `$ bounds the degree of the algebraic variety $`\overline{V(f_1,\mathrm{},f_k)SingW_0}`$ and of the $`i`$–th polar variety $`\stackrel{~}{W_i}=\overline{W_iSingW_0}`$.
In and the quantity $`\mathrm{max}\{\mathrm{deg}V(f_1,\mathrm{},f_i)|1ip\}`$ is called the geometric degree (of the complex interpretation) of the equation system $`f_1,\mathrm{},f_p`$. In analogy to this terminology, we shall call $`\delta `$ the geometric degree of the real interpretation of the equation system $`f_1,\mathrm{},f_p`$. In view of the complexity result above we shall understand the parameter $`\delta `$ as an intrinsic measure for the size of the real interpretation of the given polynomial equation system. Nevertheless, the word ”intrinsic” should be interpreted with some caution in this context: observe that the complexity parameter $`\delta `$ depends rather on the equations $`f_1,\mathrm{},f_p`$ and their order than just on the variety $`\overline{W_0SingW_0}`$.
In order to make our complexity result more transparent we are going now to exhibit, in terms of extrinsic parameters, some estimations for the intrinsic system degree $`\delta `$.
Let us write $`d_1:=\mathrm{deg}f_1,\mathrm{},d_p:=\mathrm{deg}f_p`$ and let $`D:=d_1\mathrm{}d_p`$ denote the classical Bézout number of the polynomial system $`f_1,\mathrm{},f_p`$. Then we have the following degree estimations for the complex algebraic variety $`W_0=V(f_1,\mathrm{},f_p)`$
$$\mathrm{deg}\overline{S_0}\mathrm{deg}W_0Dd^p$$
($`\overline{S_0}`$ denotes again the Q -Zariski closure in $`\text{C}\text{ }^n`$ of the real variety $`S_0`$ ).
On the other hand, we conclude from Theorem 8 that, for every $`i,\mathrm{\hspace{0.33em}1}inp,`$ the polar variety $`\stackrel{~}{W_i}`$ is defined by the initial system $`f_1,\mathrm{},f_p`$ and certain $`p`$–minors of the Jacobian $`J(f_1,\mathrm{},f_p)`$. Let us denote the maximum degree of these $`p`$–minors by $`c_i`$. It turns out that for, any $`1inp`$, the polar variety $`\stackrel{~}{W_i}`$ is a codimension one subvariety of $`\stackrel{~}{W}_{i1}`$. Now one sees easily that the quantity $`D_i:=Dc_1\mathrm{}c_i`$ represents a reasonable ”Bézout number” of the variety $`\stackrel{~}{W_i}`$ and that this Bézout number satisfies the estimate $`\mathrm{deg}\stackrel{~}{W_i}D_i`$. Putting all this together, we deduce the following estimate for the intrinsic system degree $`\delta `$:
$$\delta D_{np}=Dc_1\mathrm{}c_{np}.$$
Observing that for any $`i,\mathrm{\hspace{0.33em}1}inp,`$ the inequality $`c_id_1+\mathrm{}+d_pp`$ holds, we find the estimations:
$$\delta D(d_1+\mathrm{}+d_pp)^{np}d^p(pdp)^{np}<p^{np}d^n.$$
In conclusion, our new real algorithm has a time complexity that is, in worst case, polynomial in the ”Bézout number” $`Dc_1\mathrm{}c_{np}`$ of the zero–dimensional polar variety $`\stackrel{~}{W}_{np}`$.
Our complexity bound $`\left(\genfrac{}{}{0pt}{}{n}{p1}\right)L(nd\delta )^{O(1)}`$ depends on the intrinsic (geometric, semantic) parameter $`\delta `$ and on the extrinsic (algebraic) parameters $`d`$ and $`n`$ in a polynomial manner, and it depends on the syntactic parameter $`L`$ only linearly. In this sense one may consider our complexity bound as intrinsic. Our real algorithm promises therefore to be practically applicable to special equation systems with low value for the intrinsic parameter $`\delta `$.
On the other hand, even in worst case our algorithm improves upon the known $`d^{O(n)}`$–time procedures for the algorithmic problem under consideration, also in their most efficient versions , (see also , , , , , , , , , ). However, this distinction does not become apparent when we measure complexities simply in terms of $`d`$ and $`n`$ (all mentioned algorithms have worst–case complexities of type $`d^{O(n)}`$), but it becomes clearly visible when we use the ”Bézout number” just introduced as complexity parameter. Only our new algorithm is polynomial in this quantity. On the other hand, we are only able to reach our goal of algorithmic efficiency by means of a strict limitation to a purely geometric point of view. For the moment there is no hope that any of the standard questions of real algebra (e.g. finding generators for the real radical of a polynomial ideal or the formulation of an effective real Nullstellensatz) can be solved within the complexity framework of this paper (compare , and ).
In conclusion we may say that this paper establishes a new connection between the algorithmic complexity of finding a representative set of real solutions of a given polynomial equation system and the geometry of the (complex) algebraic variety defined by this system. However, there is a price to pay for that: this connection becomes only visible if we restrict ourselves to reduced complete intersection systems that define smooth, compact real varieties.
Our (algorithmic and mathematical) methods and results represent a non–obvious generalization of the main outcome of , where an intrinsic type algorithm was designed for the problem of finding at least one representative point in each connected component of a real, compact hypersurface given by an $`n`$–variate, smooth polynomial equation $`f`$ of degree $`d2`$ with rational coefficients (such that $`f`$ represents a regular equation of that hypersurface). This is the particular case of codimension $`p=1`$ of the present paper, and our setting leads to the complexity bound $`L(nd\delta )^{O(1)}`$ proved in .
## 2 Polar Varieties
### 2.1 Notations, Notions and General Assumptions
Let $`X_1,\mathrm{},X_n`$ be indeterminates (or variables) over the rational numbers Q and let polynomials $`f_1,\mathrm{},f_p\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ with $`1pn`$ be given. Let $`\text{C}\text{ }^n`$ and $`\mathrm{IR}^n`$ denote the $`n`$–dimensional affine space over the complex and the real numbers, respectively. We think $`\text{C}\text{ }^n`$ to be equipped with the Q -Zariski topology, whereas, on $`\mathrm{IR}^n`$, we consider the strong topology. For any subset $`U\text{C}\text{ }^n`$ we denote its Q -Zariski–closure by $`\overline{U}`$. By $`X:=(X_1,\mathrm{},X_n)`$ we denote the vector of variables $`X_1,\mathrm{},X_n`$ and by $`x:=(x_1,\mathrm{},x_n)`$ any point of the affine space $`\text{C}\text{ }^n`$ or $`\mathrm{IR}^n`$. We suppose that the polynomials $`f_1,\mathrm{},f_p`$ form a reduced regular sequence in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ (here ”reduced” means that for any $`1kp`$ the ideal $`(f_1,\mathrm{},f_k)`$ is radical). The Jacobian of these polynomials is denoted by
$$J(f_1,\mathrm{},f_p):=\left[\frac{f_k}{X_j}\right]_{\genfrac{}{}{0pt}{}{1kp}{1jn}}.$$
For any point $`x\text{C}\text{ }^n`$ we write
$$J(f_1,\mathrm{},f_p)(x):=\left[\frac{f_k}{X_j}(x)\right]_{\genfrac{}{}{0pt}{}{1kp}{1jn}}$$
for the Jacobian of the polynomials $`f_1,\mathrm{},f_p`$ at $`x`$.
The common complex zeros of the polynomials $`f_1,\mathrm{},f_p`$ form an affine, Q -definable subvariety of $`\text{C}\text{ }^n`$, which we denote by
$$W_0:=V(f_1,\mathrm{},f_p):=\{x\text{C}\text{ }^n|f_1(x)=\mathrm{}=f_p(x)=0\}.$$
A point $`xW_0=V(f_1,\mathrm{},f_p)`$ is said to be non–singular or smooth (in $`W_0`$) if the rank of the Jacobian of $`f_1,\mathrm{},f_p`$ in $`x`$ is $`p`$. Otherwise $`x`$ is called a singular point of $`W_0`$. By $`SingW_0`$ we denote the set of all singular points of $`W_0`$. Since we suppose that the ideal $`(f_1,\mathrm{},f_p)`$ is radical, our notion of a smooth point coincides with the usual one for algebraic varieties by Jacobi’s criterion.
###### Remark 1
If $`xW_0`$ is smooth, then the hypersurfaces defined by the polynomials $`f_1,\mathrm{},f_p`$ intersect transversally at the point$`x`$.
###### Definition 2
For every $`i,\mathrm{\hspace{0.33em}1}inp,`$ let $`\mathrm{\Delta }_i`$ denote the set of all common complex zeros of all $`p`$-minors of the Jacobian $`J(f_1,\mathrm{},f_p)`$ corresponding to the columns $`\{1,\mathrm{},p+i1\}`$. In other words, $`\mathrm{\Delta }_i`$ is the determinantal variety defined by all p-minors of the submatrix $`J_1^{p+i1}(f_1,\mathrm{},f_p)`$ determined by the columns $`\{1,\mathrm{},p+i1\}`$ of the Jacobian $`J(f_1,\mathrm{},f_p)`$.
We introduce the affine variety
$$W_i:=W_0\mathrm{\Delta }_i$$
associated with the linear subspace of $`\text{C}\text{ }^n`$, namely
$$X^{p+i1}:=\{x\text{C}\text{ }^n|X_{p+i}(x)=\mathrm{}=X_n(x)=0\}$$
and call $`W_i`$ the $`i`$–th formal polar variety of $`W_0`$.
By
$$\stackrel{~}{W_i}:=\overline{W_iSingW_0}$$
we denote the $`i`$–th polar variety (in the usual sense) of the variety $`W_0`$.
###### Remark 3
* Our definition of polar and formal polar variety depends rather on the regular sequence $`f_1,\mathrm{},f_p`$ than on the algebraic variety $`W_0`$. The ad hoc term ”formal polar variety” is only used in this paper for the purpose of clarification of our subsequent mathematical arguments.
* The index $`i`$ reflects the expected codimension of the polar variety $`\stackrel{~}{W_i}`$ in $`W_0`$. With respect to the ambient space $`\text{C}\text{ }^n`$, the expected codimension of $`\stackrel{~}{W_i}`$ is $`p+i`$ (see Theorem 8 below for a precise statement).
* According to our notation, the common zeros of all $`p`$–minors of the Jacobian $`J(f_1,\mathrm{},f_p)`$ form the determinantal variety $`\mathrm{\Delta }_{np+1}`$. Obviously, we have $`SingW_0=W_0\mathrm{\Delta }_{np+1}=W_{np+1}`$.
* The formal polar varieties $`W_i,\mathrm{\hspace{0.33em}1}inp,`$ constitue a decreasing sequence. In particular, we have
$$W_0W_1\mathrm{}W_i\mathrm{}W_{np}W_{np+1}=SingW_0.$$
The concept of polar variety goes back to J.–V. Poncelet. Its development has a long history: Let us mention among others the contributions of F. Severi, J. A. Todd, S. Kleiman, R. Piene, D. T. Lê, B. Teissier, J.–P. Henry, M. Merle … (see e.g. and the references quoted there).
### 2.2 Local Description of the Determinantal Varieties
In this subsection we develop a succinct local description of the determinantal varieties $`\mathrm{\Delta }_i,\mathrm{\hspace{0.33em}1}inp`$. The following general Exchange Lemma will be our main tool for this description (this lemma is used in a similar form in ). It describes an exchange relation between certain minors of a given matrix .
Let $`A`$ be a given $`(p\times n)`$-matrix with entries $`a_{ij}`$ from an arbitrary commutative ring. Let $`l`$ and $`k`$ be any natural numbers with $`ln`$ and $`k\mathrm{min}\{p,l\}`$. Furthermore, let $`I_k:=(i_1,\mathrm{},i_k)`$ be an ordered sequence of $`k`$ different elements from the finite set of natural numbers $`\{1,\mathrm{},l\}`$ and let $`M_A\left(I_k\right):=M_A(i_1,\mathrm{},i_k)`$ denote the $`k`$-minor of the matrix $`A`$ built up by the first $`k`$ rows and the columns $`i_1,\mathrm{},i_k`$. If it is clear by the context what is the matrix $`A`$, we shall just write $`M(i_1,\mathrm{},i_k):=M\left(I_k\right):=M_A\left(I_k\right)`$.
###### Lemma 4 (Exchange Lemma)
As before let a matrix $`A`$ and natural numbers $`l`$ and $`k`$ be given, as well as two intersecting index sets $`I_k=(i_1,\mathrm{},i_k)`$ and $`I_{k1}=(j_1,\mathrm{},j_{k1})`$. Then, for suitable numbers $`\epsilon _j\{1,1\}`$ with $`jI_kI_{k1}`$ we have the following identity:
$$()M\left(I_{k1}\right)M\left(I_k\right)=\underset{jI_kI_{k1}}{}\epsilon _jM\left(I_k\{j\}\right)M\left(I_{k1}\{j\}\right).$$
Proof
Consider the following $`((2k1)\times (2k1))`$-matrix $`L`$ with entries from the given matrix $`A`$:
$$L:=\left[\begin{array}{cc}& L_1(I_k)\\ O& \mathrm{}\\ & L_{k1}(I_k)\\ & \\ & \\ L_1(I_{k1})& L_1(I_k)\\ \mathrm{}& \mathrm{}\\ L_k(I_{k1})& L_k(I_k)\end{array}\right].$$
Here, for any $`1jk,L_j\left(I_k\right)`$ denotes the row vector of length $`k`$ that we obtain selecting, from the $`j`$–th row of the matrix $`A`$, the $`k`$ elements placed in the columns $`I_k=(i_1,\mathrm{},i_k)`$. Similarly, $`L_j\left(I_{k1}\right)`$ is obtained from the $`j`$–th row of $`A`$ selecting the $`k1`$ elements placed in the columns $`I_{k1}=(j_1,\mathrm{},j_{k1})`$.
Now it is not difficult to verify the identity $`()`$ by calculating the determinant $`detL`$ of the quadratic matrix $`L`$ via Laplace expansion in two different ways. First, by expansion of $`detL`$ according to the first $`k1`$ columns of $`L`$, we obtain the left–hand side of $`()`$, disregarding the sign. Expansion of $`detL`$ according to the first $`k1`$ rows of $`L`$ leads to the right–hand side of $`()`$. This implies the identity $`()`$ for an appropriate choice of the signs $`\epsilon _j`$, with $`jI_kI_{k1}`$. $`\mathrm{}`$
Let $`m\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ denote the $`(p1)`$–minor of the Jacobian $`J(f_1,\mathrm{},f_p)`$ given by the first $`(p1)`$ rows and columns, i.e., let
$$m:=det\left[\frac{f_k}{X_j}\right]_{\genfrac{}{}{0pt}{}{1kp1}{1jp1}}.$$
We consider the determinantal variety $`\mathrm{\Delta }_i`$ outside of the hypersurface
$$V(m):=\{x\text{C}\text{ }^n|m(x)=0\}$$
and denote this localization by $`(\mathrm{\Delta }_i)_m`$ , i.e., we set
$$(\mathrm{\Delta }_i)_m:=\mathrm{\Delta }_iV(m).$$
From now on , for $`1i_1\mathrm{}i_pn`$, let us denote by
$$M(i_1,\mathrm{},i_p)$$
the polynomial in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ defined as the $`p`$–minor of the Jacobian $`J(f_1,\mathrm{},f_p)`$ built up by its $`p`$ rows and the columns $`i_1,\mathrm{},i_p`$. As before, we denote by
$$M(i_1,\mathrm{},i_p)(x)$$
the specialization of $`M(i_1,\mathrm{},i_p)`$ in a given point $`x\text{C}\text{ }^n`$.
###### Proposition 5
Let $`1inp`$ be arbitrarily fixed, and let $`m`$ be the $`(p1)`$– minor defined above. Then the determinantal variety $`\mathrm{\Delta }_i`$ is locally (i.e., outside of the hypersurface $`V(m)`$) described by the $`i`$ polynomials
$$M(1,\mathrm{},p1,p),M(1,\mathrm{},p1,p+1),\mathrm{},M(1,\mathrm{},p1,p+i1).$$
In other words, we have
$$(\mathrm{\Delta }_i)_m:=\{x\text{C}\text{ }^n|m(x)0,M(1,\mathrm{},p1,s)(x)=0,s\{p,\mathrm{},p+i1\}\},$$
where $`M(1,\mathrm{},p1,s)`$ denotes, as above, the $`p`$–minor of the Jacobian $`J(f_1,\mathrm{},f_p)`$ built up by the first $`p1`$ columns and the $`s`$–th column.
Proof
It suffices to show that
$$(\mathrm{\Delta }_i)_m\{x\text{C}\text{ }^n|m(x)0,M(1,\mathrm{},p1,s)=0,s\{p,\mathrm{},p+i1\}\}$$
holds.
Let $`x^{}\text{C}\text{ }^n`$ be any point satisfying the conditions $`m(x^{})0`$ and $`M(1,\mathrm{},p1,s)(x^{})=0`$ for every $`s\{p,\mathrm{},p+i1\}`$. We have to verify that
$$M(i_1,\mathrm{},i_p)(x^{})=0$$
holds for all ordered $`p`$–tuples $`(i_1,\mathrm{},i_p)`$ of elements of $`\{1,\mathrm{},p+i1\}`$.
Applying the Exchange Lemma to $`m=M(1,\mathrm{},p1)`$ and $`M(i_1,\mathrm{},i_p)`$, we deduce the identity
$$m(x^{})M(i_1,\mathrm{},i_p)(x^{})=$$
$$=\underset{j\{i_1,\mathrm{},i_p\}\{1,\mathrm{},p1\}}{}\epsilon _jM\left(\{i_1,\mathrm{},i_p\}\{j\}\right)(x^{})M(1,\mathrm{},p1,j)(x^{})$$
for suitable numbers $`\epsilon _j\{1,1\}`$ with $`j\{i_1,\mathrm{},i_p\}\{1,\mathrm{},p1\}`$. By assumption we have $`m(x^{})0`$ and $`M(1,\mathrm{},p1,j)(x^{})=0\text{for all}j\{p,\mathrm{},p+i1\}`$. This implies that $`x^{}`$ belongs to the set $`(\mathrm{\Delta }_i)_m`$.
$`\mathrm{}`$
###### Notation 6
In the sequel we shall simply write $`M_j`$ for the $`p`$– minor $`M(1,\mathrm{},p1,j)`$ given by the first $`p1`$ columns of $`J(f_1,\mathrm{},f_p)`$ and the column $`j\{p,\mathrm{},n\}`$ .
###### Remark 7
* Proposition 5 implies that the codimension of $`\mathrm{\Delta }_i`$ outside of the hypersurface $`V(m)`$ is at most $`i`$.
* Proposition 5 holds also for the determinantal variety $`\mathrm{\Delta }_{np+1}`$ that defines the singular locus $`SingW_0=W_{np+1}`$ of the variety $`W_0`$. Hence, for any point $`x^{}\text{C}\text{ }^n`$ satisfying the condition $`m(x^{})0`$ and the $`np+1`$ equations
$$M_j(x^{})=0,j\{p,\mathrm{},n\},$$
the Jacobian $`J(f_1,\mathrm{},f_p)(x^{})`$ becomes singular.
* Replacing the previously chosen $`(p1)`$-minor $`m`$ by any other $`(p1)`$-minor of the Jacobian $`J(f_1,\mathrm{},f_p)`$, the statement of Proposition 5 remains true mutatis mutandis.
### 2.3 Local Description of the Formal Polar Varieties
The aim of this subsection is to show the following fact:
Let the variables $`X_1,\mathrm{},X_n`$ be in generic position with respect to the polynomials $`f_1,\mathrm{},f_p`$, and let $`\stackrel{~}{m}`$ be any $`(p1)`$–minor of the Jacobian $`J(f_1,\mathrm{},f_p)`$. In this subsection we are going to show that any formal polar variety $`W_i,\mathrm{\hspace{0.33em}1}inp,`$ is a smooth complete intersection variety outside of the closed set $`SingW_0V(\stackrel{~}{m})`$. Moreover, we shall exhibit a reduced regular sequence describing this variety outside of $`SingW_0V(\stackrel{~}{m})`$.
As in the previous subsection, let $`m\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ denote the $`(p1)`$–minor of the Jacobian $`J(f_1,\mathrm{},f_p)`$ built up by the first $`(p1)`$ rows and columns.
Let $`Y_1,\mathrm{},Y_n`$ be new variables and let $`Y:=(Y_1,\mathrm{},Y_n)`$. For any linear coordinate transformation $`X=AY`$, with $`A`$ being a regular $`(n\times n)`$–matrix, we define the polynomials
$$G_1(Y):=f_1(AY),\mathrm{},G_p(Y):=f_p(AY).$$
The Jacobian of $`G_1,\mathrm{},G_p`$ has the form
$$J(G_1,\mathrm{},G_p):=\left[\frac{G_k}{Y_j}\right]_{\genfrac{}{}{0pt}{}{1kp}{1jn}}=J(f_1,\mathrm{},f_p)A.$$
Using a similar notation as before, we denote by
$$\stackrel{~}{M}(i_1,\mathrm{},i_p)$$
the $`p`$–minor of the new Jacobian $`J(G_1,\mathrm{},G_p)`$ that corresponds to the columns $`1i_1<\mathrm{}<i_pn`$.
Moreover, we denote by $`\stackrel{~}{M}_j`$ the $`p`$– minor $`\stackrel{~}{M}(1,\mathrm{},p1,j)`$ determined by the fixed first $`p1`$ columns of $`J(G_1,\mathrm{},G_p)`$ and the column $`j\{p,\mathrm{},n\}`$.
For $`pr,tn`$ let $`Z_{r,t}`$ be a new indeterminate. Using the following regular $`(np+1)\times (np+1)`$–parameter matrix
$$Z:=\left[\begin{array}{cccccccc}1\hfill & 0\hfill & & 0\hfill & & & \mathrm{}& 0\hfill \\ Z_{p+1,p}\hfill & 1\hfill & & & & & & \\ \mathrm{}\hfill & \mathrm{}\hfill & \mathrm{}& & \text{O}\hfill & & & \mathrm{}\hfill \\ Z_{p+i1,p}\hfill & Z_{p+i1,p+1}\hfill & \mathrm{}& 1\hfill & & & & \\ Z_{p+i,p}\hfill & Z_{p+i,p+1}\hfill & \mathrm{}& Z_{p+i,p+i1}\hfill & 1\hfill & & & \\ \mathrm{}\hfill & \mathrm{}\hfill & & \mathrm{}\hfill & \mathrm{}\hfill & \mathrm{}\hfill & & 0\hfill \\ Z_{n,p}\hfill & Z_{n,p+1}\hfill & \mathrm{}& Z_{n,p+i1}\hfill & Z_{n,p+i}\hfill & Z_{n,p+i+1}\hfill & \mathrm{}& 1\hfill \end{array}\right],$$
we construct an $`(n\times n)`$–coordinate transformation matrix $`A:=A(Z)`$, which will enable us to prove the statement at the beginning of this subsection.
For the moment, let us fix an index $`1inp`$. We consider the formal polar variety $`W_i`$ outside of the hypersurface $`V(m)`$. Corresponding to our choice of $`i`$, the matrix $`Z`$ may be subdivided into submatrices as follows:
$$Z=\left[\begin{array}{cc}Z_1^{(i)}\hfill & O_{i,npi+1}\\ Z^{(i)}\hfill & Z_2^{(i)}\end{array}\right].$$
Here the matrix $`Z^{(i)}`$ is defined as
$$Z^{(i)}:=\left[\begin{array}{ccc}Z_{p+i,p}& \mathrm{}& Z_{p+i,p+i1}\\ \mathrm{}& \mathrm{}& \mathrm{}\\ Z_{np}& \mathrm{}& Z_{n,p+i1}\end{array}\right],$$
and $`Z_1^{(i)}`$ and $`Z_2^{(i)}`$ denote the quadratic lower triangular matrices bordering $`Z^{(i)}`$ in $`Z`$, and $`O_{i,npi+1}`$ is the $`i\times (npi+1)`$ zero matrix. Let
$$A:=A(Z):=\left[\begin{array}{ccc}I_{p1}& O_{p1,i}& O_{p1,npi+1}\\ O_{i,p1}& Z_1^{(i)}& O_{i,npi+1}\\ O_{npi+1,p1}& Z^{(i)}& Z_2^{(i)}\end{array}\right].$$
Here the submatrices $`I_r`$ and $`O_{r,s}`$ are unit or zero matrices, respectively, of corresponding size, and $`Z^{(i)},Z_1^{(i)},`$ and $`Z_2^{(i)}`$ are the submatrices of the parameter matrix $`Z`$ introduced before. Thus, $`A`$ is a regular, parameter dependent $`(n\times n)`$–coordinate transformation matrix.
Like the matrix $`Z`$, the matrix $`A`$ contains
$$s:=\frac{(np)(np+1)}{2}$$
parameters $`Z_{r,t}`$ which we may specialize into any point $`z`$ of the affine space $`\text{C}\text{ }^s`$. For such a point $`z\text{C}\text{ }^s`$ we denote the corresponding specialized matrices by $`A(z)`$, $`Z_1^{(i)}(z),Z_2^{(i)}(z)`$ and $`Z^{(i)}(z)`$.
We consider now the coordinate transformation given by $`X=AY`$ with $`A=A(Z)`$ and calculate the Jacobian $`J(G_1,\mathrm{},G_p)`$ with respect to the new polynomials $`G_1,\mathrm{},G_p`$ . Recall that the coordinate transformation matrix $`A`$ depends on our previous choice of the index $`1inp`$.
According to the structure of the coordinate transformation matrix $`A=A(Z)`$ we subdivide the Jacobian $`J(f_1,\mathrm{},f_p)`$ into three submatrices
$$J(f_1,\mathrm{},f_p)=\left[\begin{array}{ccc}U& V& W\end{array}\right],$$
with
$$U:=\left[\frac{f_k}{X_j}\right]_{\genfrac{}{}{0pt}{}{1kp}{1jp1}},V:=\left[\frac{f_k}{X_j}\right]_{\genfrac{}{}{0pt}{}{1kp}{pjp+i1}},W:=\left[\frac{f_k}{X_j}\right]_{\genfrac{}{}{0pt}{}{1kp}{p+ijn}}.$$
From the identity $`J(G_1,\mathrm{},G_p)=J(f_1,\mathrm{},f_p)A`$ we deduce that our new Jacobian is of the form:
$$J(G_1,\mathrm{},G_p)=\left[\frac{G_k}{Y_j}\right]_{\genfrac{}{}{0pt}{}{1kp}{1jn}}=\left[\begin{array}{ccc}U& VZ_1^{(i)}+WZ^{(i)}& WZ_2^{(i)}\end{array}\right].$$
We are interested in a local description of the $`i`$–th formal polar variety $`W_i=W_0\mathrm{\Delta }_i`$ outside of the hypersurface $`V(m)`$, where $`m`$ is the fixed upper left $`(p1)`$– minor of the Jacobian $`J(f_1,\mathrm{},f_p)`$ (and also of its submatrix $`U`$). Since the coordinate transformation $`X=AY`$ leaves the submatrix $`U`$ unchanged, the $`(p1)`$–minor $`m`$ remains fixed under this transformation. From Proposition 5 we know that the localized determinantal variety $`(\mathrm{\Delta }_i)_m`$ is described by the $`i`$ equations
$$M_p=0,\mathrm{},M_{p+i1}=0,$$
and by the condition $`m0`$. The $`p`$– minors $`M_p,\mathrm{},M_{p+i1}`$ defining these equations are built up by the submatrix $`\left[UV\right]`$ of the Jacobian $`J(f_1,\mathrm{},f_p)`$. Under the coordinate transformation $`A(Z)`$ the matrix $`\left[UV\right]`$ is changed into the submatrix
$$\left[UVZ_1^{(i)}+WZ^{(i)}\right]$$
of the Jacobian $`J(G_1,\mathrm{},G_p)`$ and the $`p`$–minors $`M_p,\mathrm{},M_{p+i1}`$ are changed into the $`p`$– minors
$$\stackrel{~}{M}_p,\mathrm{},\stackrel{~}{M}_{p+i1}$$
of the matrix $`\left[UVZ_1^{(i)}+WZ^{(i)}\right]`$. This implies the matrix identity
$$()[\stackrel{~}{M}_p,\mathrm{},\stackrel{~}{M}_{p+i1}]=[M_p,\mathrm{},M_{p+i1}]Z_1^{(i)}+[M_{p+i},\mathrm{},M_n]Z^{(i)}.$$
For the previously chosen index $`1inp`$, the coordinate transformation $`X=A(Z)Y`$ induces the following morphism of affine spaces:
$$\mathrm{\Phi }_i:\text{C}\text{ }^n\times \text{C}\text{ }^s\text{C}\text{ }^p\times \text{C}\text{ }^i,$$
defined by
$$(x,z)\mathrm{\Phi }_i(x,z):=(f_1(x),\mathrm{},f_p(x),\stackrel{~}{M}_p(x,z),\mathrm{},\stackrel{~}{M}_{p+i1}(x,z)).$$
Consider an arbitrary point $`z\text{C}\text{ }^s`$. We denote by $`\mathrm{\Delta }_i^z`$ the determinantal subvariety of $`\text{C}\text{ }^n`$ defined by all $`p`$–minors of the matrix $`\left[UVZ_1^{(i)}(z)+WZ^{(i)}(z)\right]`$ (which is a submatrix of the new Jacobian obtained by specializing the coefficients of the polynomials $`G_1,\mathrm{},G_p`$ into the point $`z\text{C}\text{ }^s`$). Writing $`W_i^z:=W_0\mathrm{\Delta }_i^z`$, one sees immediately that the zero fiber $`\mathrm{\Phi }_i^1(0)`$ of the morphism $`\mathrm{\Phi }_i`$ contains the set
$$(W_i^z)_m:=W_0(\mathrm{\Delta }_i^z)_m.$$
In other words, for any arbitrarily chosen point $`z\text{C}\text{ }^s`$, the zero fiber $`\mathrm{\Phi }_i^1(0)`$ of the morphism $`\mathrm{\Phi }_i`$ contains the transformed formal polar variety $`W_i^z`$, localized in the hypersurface $`V(m)`$ and expressed in the old coordinates.
We are going now to analyze the rank of the Jacobian of the morphism $`\mathrm{\Phi }_i`$ in an arbitrary point $`(x,z)\text{C}\text{ }^n\times \text{C}\text{ }^s`$ with $`x(W_i^z)_m`$. Using the subdivision of the parameter matrix $`Z`$ into the parts $`Z^{(i)},Z_1^{(i)}`$ and $`Z_2^{(i)}`$, the Jacobian $`J(\mathrm{\Phi }_i)`$ of the morphism $`\mathrm{\Phi }_i`$ can be written symbolically as
$$J(\mathrm{\Phi }_i)=\left[\frac{\mathrm{\Phi }_i}{X}\frac{\mathrm{\Phi }_i}{Z^{(i)}}\frac{\mathrm{\Phi }_i}{Z_1^{(i)}}\frac{\mathrm{\Phi }_i}{Z_2^{(i)}}\right].$$
We have
$$\left[\frac{\mathrm{\Phi }_i}{X}\frac{\mathrm{\Phi }_i}{Z^{(i)}}\right]=$$
$$=\left[\begin{array}{cccc}J(f_1,\mathrm{},f_p)& O_{p,npi+1}& \mathrm{}& O_{p,npi+1}\\ & [\frac{\stackrel{~}{M}_p}{Z_{p+i,p}},\mathrm{},\frac{\stackrel{~}{M}_p}{Z_{np}}]& \mathrm{}& O_{1,npi+1}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ & O_{1,npi+1}& \mathrm{}& [\frac{\stackrel{~}{M}_{p+i1}}{Z_{p+i,p+i1}},\mathrm{},\frac{\stackrel{~}{M}_{p+i1}}{Z_{n,p+i1}}]\end{array}\right],$$
where the columns correspond to the partial derivatives of $`\mathrm{\Phi }_i`$ with respect to the variables
$$X_1,\mathrm{},X_n,Z_{p+i,p},\mathrm{},Z_{n,p},\mathrm{},Z_{p+i,p+i1},\mathrm{},Z_{n,p+i1}$$
(in this order). The entries $`O_{r,t}`$ denote here zero matrices of corresponding size and the row matrices labeled by ”$``$” represent the partial derivatives with respect to the variables $`X_1,\mathrm{},X_n`$ of the minors $`\stackrel{~}{M}_p,\mathrm{},\stackrel{~}{M}_{p+i1}`$. These row matrices will be irrelevant for our considerations.
Furthermore, the third submatrix $`\left[{\displaystyle \frac{\mathrm{\Phi }_i}{Z_1^{(i)}}}\right]`$ of $`J(\mathrm{\Phi }_i)`$ can be written as
$$\left[\begin{array}{cccc}O_{p,i1}& O_{p,i2}& \mathrm{}& 0\\ [\frac{\stackrel{~}{M}_p}{Z_{p+1,p}},\mathrm{},\frac{\stackrel{~}{M}_p}{Z_{p+i1,p}}]& O_{1,i2}& \mathrm{}& 0\\ O_{1,i1}& [\frac{\stackrel{~}{M}_{p+1}}{Z_{p+2,p+1}},\mathrm{},\frac{\stackrel{~}{M}_p}{Z_{p+i1,p+1}}]& \mathrm{}& 0\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ O_{1,i1}& O_{1,i2}& \mathrm{}& \left[\frac{\stackrel{~}{M}_{p+i1}}{Z_{p+i1,p+i2}}\right]\\ O_{1,i1}& O_{1,i2}& \mathrm{}& 0\end{array}\right],$$
and the last submatrix $`\left[{\displaystyle \frac{\mathrm{\Phi }_i}{Z_2^{(i)}}}\right]`$ of $`J(\mathrm{\Phi }_i)`$ is a zero matrix since the $`p`$– minors $`\stackrel{~}{M}_p,\mathrm{},\stackrel{~}{M}_{p+i1}`$ are indepedent of the parameters $`Z_{r,t}`$ occurring in the submatrix $`Z_2^{(i)}`$ of the coordinate transformation matrix $`A(Z)`$.
Therefore, the Jacobian $`J(\mathrm{\Phi }_i)`$ is of full rank $`p+i`$ wherever the submatrix
$$\stackrel{~}{J}(\mathrm{\Phi }_i):=\left[\frac{\mathrm{\Phi }_i}{X}\frac{\mathrm{\Phi }_i}{Z^{(i)}}\right]$$
is of full rank $`p+i`$. On the other hand, considering the $`i`$ row matrices contained in $`\stackrel{~}{J}(\mathrm{\Phi }_i)`$ for $`pjp+i1`$
$$[\frac{\stackrel{~}{M}_j}{Z_{p+i,j}},\mathrm{},\frac{\stackrel{~}{M}_j}{Z_{n,j}}],$$
we see that the representation $`()`$ of the transformed $`p`$–minors $`\stackrel{~}{M}_j`$ implies the identity
$$[\frac{\stackrel{~}{M}_j}{Z_{p+i,j}},\mathrm{},\frac{\stackrel{~}{M}_j}{Z_{n,j}}]=[M_{p+i},\mathrm{},M_n].$$
Thus, we obtain the representation
$$\stackrel{~}{J}(\mathrm{\Phi }_i)=\left[\begin{array}{cccc}J(f_1,\mathrm{},f_p)& O_{p,npi+1}& \mathrm{}& O_{p,npi+1}\\ & [M_{p+i},\mathrm{},M_n]& \mathrm{}& O_{1,npi+1}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ & O_{1,npi+1}& \mathrm{}& [M_{p+i},\mathrm{},M_n]\end{array}\right].$$
Since all entries of the submatrix $`\stackrel{~}{J}(\mathrm{\Phi }_i)`$ of the Jacobian $`J(\mathrm{\Phi }_i)`$ belong to the polynomial ring $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$, we see that the rank of the matrix $`J(\mathrm{\Phi }_i)`$ in a given point $`(x,z)\text{C}\text{ }^n\times \text{C}\text{ }^s`$ with $`x(W_i^z)_m`$ depends only on the choice of $`x`$. According to our localization outside of the hypersurface $`V(m)`$, let us consider an arbitrary smooth point $`\stackrel{~}{x}`$ of $`W_0=V(f_1,\mathrm{},f_p)`$ satisfying the condition $`m(\stackrel{~}{x})0`$. Suppose that the submatrix $`\stackrel{~}{J}(\mathrm{\Phi }_i)(\stackrel{~}{x})`$ is not of full rank, i.e., that
$$rk\stackrel{~}{J}(\mathrm{\Phi }_i)(\stackrel{~}{x})<p+i$$
holds. This latter inequality is valid if and only if all $`p`$-minors $`M_{p+i},\mathrm{},M_n`$ of the Jacobian $`J(f_1,\mathrm{},f_p)`$ vanish at $`\stackrel{~}{x}`$. Let $`\stackrel{~}{z}\text{C}\text{ }^s`$ be any parameter point such that the pair $`(\stackrel{~}{x},\stackrel{~}{z})`$ belongs to the fiber $`\mathrm{\Phi }_i^1(0)`$ of the morphism $`\mathrm{\Phi }_i`$. Since the $`p`$-minors $`\stackrel{~}{M}_p,\mathrm{},\stackrel{~}{M}_{p+i1}`$ of the transformed Jacobian $`J(G_1,\mathrm{},G_p)`$ must vanish at $`(\stackrel{~}{x},\stackrel{~}{z}),`$ we deduce from $`()`$ that
$$[0,\mathrm{},0]=[M_p(\stackrel{~}{x}),\mathrm{},M_{p+i1}(\stackrel{~}{x})]Z_1^{(i)}(\stackrel{~}{z})$$
holds (here $`Z_1^{(i)}(\stackrel{~}{z})`$ denotes again the matrix obtained by specializing the entries of $`Z_1^{(i)}`$ into the corresponding coordinates of the point $`\stackrel{~}{z}\text{C}\text{ }^s`$). Because of the lower triangular form of the regular matrix $`Z_1^{(i)}`$, the latter matrix equation holds if and only if the conditions
$$M_{p+i1}(\stackrel{~}{x})=\mathrm{}=M_p(\stackrel{~}{x})=0.$$
are satisfyied. Therefore, our assumptions on $`\stackrel{~}{x}`$ and $`\stackrel{~}{z}`$ imply $`m(\stackrel{~}{x})0`$ and $`M_p(\stackrel{~}{x})=\mathrm{}=M_n(\stackrel{~}{x})=0.`$ However, by Remark 7 this means that the Jacobian $`J(f_1,\mathrm{},f_p)(\stackrel{~}{x})`$ is singular. Hence, $`\stackrel{~}{x}`$ is not a smooth point of $`W_0`$, i.e., $`\stackrel{~}{x}SingW_0`$, which contradicts our assumption on $`\stackrel{~}{x}`$.
Now, suppose that we are given a point $`(\overline{x},z)\text{C}\text{ }^n\times \text{C}\text{ }^s`$ that belongs to the fiber $`\mathrm{\Phi }_i^1(0)`$. Then $`\overline{x}`$ belongs to $`W_0`$. Further, suppose that $`\overline{x}`$ is a smooth point of $`W_0`$ outside of the hypersurface $`V(m)`$. Let us consider the Zariski–open neighbourhood $`\stackrel{~}{U}`$ of $`\overline{x}`$ consisting of all points $`x\text{C}\text{ }^n`$ with $`m(x)0`$ and $`rkJ(f_1,\mathrm{},f_p)=p`$, i.e., we consider
$$\stackrel{~}{U}:=\text{C}\text{ }^n\left(SingW_0V(m)\right).$$
We are going to show that the restricted morphism
$$\mathrm{\Phi }_i:\stackrel{~}{U}\times \text{C}\text{ }^s\text{C}\text{ }^p\times \text{C}\text{ }^i$$
is transversal to the origin $`0\text{C}\text{ }^p\times \text{C}\text{ }^i`$.
In order to see this, consider an arbitrary point $`(x,z)`$ of $`\stackrel{~}{U}\times \text{C}\text{ }^s`$ that satisfies the equation $`\mathrm{\Phi }_i(x,z)=0`$. Thus, $`x`$ belongs to $`\stackrel{~}{U}W_0`$ and is, therefore, a smooth point of $`W_0`$, which is outside of the hypersurface $`V(m)`$. By the preceding considerations on the rank of the Jacobian $`J(\mathrm{\Phi }_i)`$ it is clear that $`J(\mathrm{\Phi }_i)`$ has the maximal rank $`p+i`$ at $`(x,z)`$. This means that $`(x,z)`$ is a regular point of $`\mathrm{\Phi }_i`$. Since $`(x,z)`$ was an arbitrary point of $`\mathrm{\Phi }_i^1(0)(\stackrel{~}{U}\times \text{C}\text{ }^s)`$, the claimed transversality has been shown.
Now, applying the Weak–Transversality–Theorem of Thom–Sard (see e.g. ) to the diagram
$$\begin{array}{ccc}\mathrm{\Phi }_i^1(0)(\stackrel{~}{U}\times \text{C}^s)\hfill & & \text{C}^n\times \text{C}^s\\ & & \\ & & \text{C}^s\end{array}$$
one concludes that there is a residual dense set $`\mathrm{\Omega }_i`$ of parameters $`z\text{C}\text{ }^s`$ for which transversality holds. This implies that, for every fixed $`z\mathrm{\Omega }_i`$, the transformed and localized formal polar variety
$$W_i^z\left(SingW_0V(m)\right)$$
is either empty or a smooth variety of codimension $`p+i`$. This variety can be described locally by the polynomials
$$()f_1(X),\mathrm{},f_p(X),\stackrel{~}{M}_p(X,z),\mathrm{},\stackrel{~}{M}_{p+i1}(X,z)$$
that form a regular sequence outside of $`SingW_0V(m)`$. Up to now, our considerations concerned only the change of coordinates for an arbitrarily fixed $`1inp`$. However, $`\mathrm{\Omega }:=_{i=1}^{np}\mathrm{\Omega }_i`$ is a dense residual parameter set in $`\text{C}\text{ }^s`$ from which we can choose a simultaneous change of coordinates for all $`1inp`$. For every choice $`z\mathrm{\Omega }`$ and $`1inp`$ the transformed formal polar variety $`W_i^z`$ is, outside of the closed set $`SingW_0V(m)`$, a smooth complete intersection variety described by the (local) regular sequence $`()`$. One sees now easily that the affine space $`\mathrm{IR}^s`$ contains a non–empty residual dense set of parameters $`z`$ such that the conclusions above apply to the coordinate transformation $`X=A(z)Y`$. Moreover, $`z`$ can be chosen from $`\text{Q}\text{ }^s`$.
Taking into account Proposition 5 and Remark 7, we deduce the following result from our argumentation:
###### Theorem 8
Let $`W_0=V(f_1,\mathrm{},f_p)`$ be a reduced complete intersection variety given by polynomials $`f_1,\mathrm{},f_p`$ in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ and suppose that the variables $`X_1,\mathrm{},X_n`$ are in generic position with respect to $`f_1,\mathrm{},f_p`$. Further, let $`m`$ be the upper left $`(p1)`$–minor of the Jacobian $`J(f_1,\mathrm{},f_p)`$. Then, every formal polar variety $`W_i,\mathrm{\hspace{0.33em}1}inp,`$ localized with respect to the closed set $`SingW_0V(m)`$, is either empty or a smooth variety of codimension $`p+i`$ that can be described by the equations
$$f_1,\mathrm{},f_p,M_p,\mathrm{},M_{p+i1},$$
where $`M_j,pjp+i1,`$ is the $`p`$– minor of the Jacobian $`J(f_1,\mathrm{},f_p)`$ given by the columns $`1,\mathrm{},p1,j`$. Then the polynomials
$$f_1,\mathrm{},f_p,M_p,\mathrm{},M_{p+i1}$$
form a regular sequence outside of $`SingW_0V(m)`$ .
###### Remark 9
Taking into account that the argumentation on the localization with respect to the fixed $`(p1)`$-minor $`m`$ remains valid mutatis mutandis for any other $`(p1)`$-minor $`\stackrel{~}{m}`$ of the Jacobian $`J(f_1,\mathrm{},f_p)`$, Theorem 8 can be restated for any fixed $`(p1)`$–minor just by reordering of columns and rows of the Jacobian $`J(f_1,\mathrm{},f_p)`$.
### 2.4 Existence of Real Points in the Polar Varieties
Let $`f_1,\mathrm{},f_p\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ be a reduced regular sequence and let again $`W_0:=V(f_1,\mathrm{},f_p)`$ be the affine variety defined by $`f_1,\mathrm{},f_p`$. Consider the real variety $`S_0:=W_0\mathrm{IR}^n`$ and suppose that
* $`S_0`$ is nonempty and bounded (and hence compact),
* the Jacobian $`J(f_1,\mathrm{},f_p)(x)`$ is of maximal rank in all points $`x`$ of $`S_0`$ (i.e., $`S_0`$ is a smooth subvariety of $`\mathrm{IR}^n`$ given by the reduced regular sequence $`f_1,\mathrm{},f_p`$),
* the variables $`X_1,\mathrm{},X_n`$ are in generic position with respect to the polynomials $`f_1,\mathrm{},f_p`$.
Further, let $`C`$ be any connected component of the compact set $`S_0`$, and let $`b:=(a_1,\mathrm{},a_{p1},a_p,\mathrm{},a_{n1},a_n)C`$ be a locally maximal point of the last coordinate $`X_n`$ in the non–empty compact set $`CS_0`$. Without loss of generality we may assume that the upper left $`(p1)`$–minor $`m`$ of the Jacobian $`J(f_1,\mathrm{},f_p)`$ does not vanish in $`b`$ (by our assumptions there must be a $`(p1)`$–minor of $`J(f_1,\mathrm{},f_p)`$ not vanishing at $`b`$). In any local parametrization of $`S_0`$ at $`b`$ the variable $`X_n`$ cannot be an independent variable, since $`X_n`$ attains a local maximum in $`b`$ ($`a_n`$ is this local maximum). Hence, without loss of generality we may assume that the local parametrization of $`S_0`$ in $`b`$ has the following form: there exists an open set $`𝒰\mathrm{IR}^{np}`$ containing the point $`a:=(a_p,\mathrm{},a_{n1})`$, and a continuously differentiable function
$$\phi :𝒰\mathrm{IR}^p,\phi :=(\phi _1,\mathrm{},\phi _{p1},\phi _n)$$
such that
$$x_1=\phi _1(x_p,\mathrm{},x_{n1}),\mathrm{},x_{p1}=\phi _{p1}(x_p,\mathrm{},x_{n1}),$$
$$x_n=\phi _n(x_p,\mathrm{},x_{n1})$$
holds for any $`x=(x_p,\mathrm{},x_{n1})𝒰`$. With respect to this local parametrization, the polynomials $`f_k,\mathrm{\hspace{0.33em}1}kp`$, induce real valued functions of the form:
$$\stackrel{~}{f_k}(X_p,\mathrm{},X_{n1}):=$$
$$f_k(\phi _1(X_p,\mathrm{},X_{n1}),\mathrm{},\phi _{p1}(X_p,\mathrm{},X_{n1}),$$
$$X_p,\mathrm{},X_{n1},\phi _n(X_p,\mathrm{},X_{n1})).$$
For every $`1kp,`$ and every $`pjn1,`$ one has the identity
$$\frac{\stackrel{~}{f}_k}{X_j}=\frac{f_k}{X_j}+\frac{f_k}{X_1}\frac{\phi _1}{X_j}+\mathrm{}+\frac{f_k}{X_{p1}}\frac{\phi _{p1}}{X_j}+\frac{f_k}{X_n}\frac{\phi _n}{X_j}=0$$
(1)
in the open set $`𝒰`$.
Considering the $`(p\times p)`$–matrix
$$B:=\left[\begin{array}{cc}\frac{f_1}{X_1}\mathrm{}\hfill & \frac{f_1}{X_{p1}}\frac{f_1}{X_n}\hfill \\ & \\ \frac{f_p}{X_1}\mathrm{}\hfill & \frac{f_p}{X_{p1}}\frac{f_p}{X_n}\hfill \end{array}\right],$$
and observing that $`B`$ is regular in $`𝒰`$, we obtain from (1) that
$$detB(x)\left[\begin{array}{c}\frac{\phi _1}{X_j}\\ \mathrm{}\\ \frac{\phi _{p1}}{X_j}\\ \frac{\phi _n}{X_j}\end{array}\right]=(AdjB)(x)\left[\begin{array}{c}\frac{f_1}{X_j}(x)\\ \mathrm{}\\ \frac{f_{p1}}{X_j}(x)\\ \frac{f_p}{X_j}(x)\end{array}\right]$$
(2)
holds for any $`x𝒰`$ (here $`AdjB`$ denotes the adjoint matrix of the matrix $`B`$). As $`b`$ is a locally maximal point of $`X_n`$, we have that
$$\frac{\phi _n}{X_j}(a)=0$$
holds for every $`pjn1`$. Thus, equation (2) implies
$$B(n,1)(b)\frac{f_1}{X_j}(b)+\mathrm{}+B(n,p)(b)\frac{f_p}{X_j}(b)=0$$
(3)
for every $`pjn1`$ (here for $`1kp`$ we denote the entry of the adjoint matrix $`AdjB`$ at the cross point of the $`k`$–th column and the last row by $`B(n,k)`$). Taking into account the particular form of the matrix $`B`$, the equation system (3) means that
$$det\left[\begin{array}{cc}\frac{f_1}{X_1}(b)\mathrm{}\hfill & \frac{f_1}{X_{p1}}(b)\frac{f_1}{X_j}(b)\hfill \\ & \\ \frac{f_p}{X_1}(b)\mathrm{}\hfill & \frac{f_p}{X_{p1}}(b)\frac{f_p}{X_j}(b)\hfill \end{array}\right]=\mathrm{\hspace{0.33em}0}$$
(4)
holds for every $`pjn1`$. Using our notations for the $`p`$–minors of the Jacobian $`J(f_1,\mathrm{},f_p)`$, we reinterprete now the equations (4) as
$$M_p(b)=\mathrm{}=M_{n1}(b)=0.$$
Since $`m(b)0`$ holds by assumption, Proposition 5 implies that $`b`$ belongs to the localized determinantal variety $`(\mathrm{\Delta }_{np})_m`$. Therefore, we have $`bW_0(\mathrm{\Delta }_{np})_m`$, i.e., the last formal polar variety $`W_{np}`$ contains the point $`b`$. On the other hand, $`b`$ is a nonsingular point of $`W_0`$ and belongs therefore to $`\stackrel{~}{W}_{np}=\overline{W_{np}SingW_0}`$. Thus $`\stackrel{~}{W}_{np}`$ is a non–empty set of dimension zero that contains the real point $`b`$ of the arbitrarily chosen connected component $`C`$ of the real variety $`S_0`$. In particular, $`b\stackrel{~}{W}_{np}\mathrm{IR}^nW_i\mathrm{IR}^n=S_i`$ holds for any $`1inp`$.
These considerations imply the following result:
###### Theorem 10
Let $`W_0:=V(f_1,\mathrm{},f_p)`$ be as in Theorem 8. If the real variety $`S_0:=W_0\mathrm{IR}^n`$ is non–empty, bounded and smooth, and if the variables $`X_1,\mathrm{},X_n`$ are in generic position with respect to $`f_1,\mathrm{},f_p`$, then every real formal polar variety $`S_i=W_i\mathrm{IR}^n,\mathrm{\hspace{0.33em}1}inp,`$ is a non–empty, smooth manifold of dimension $`n(p+i)`$ and contains at least one representative point of each connected component of the real variety $`S_0`$.
## 3 Real Equation Solving
The geometric results of Section 2 allow us to design a new efficient procedure that finds at least one representative point in each connected component of a given smooth, compact, real complete intersection variety.
This procedure will be formulated in the algorithmic (complexity) model of (division-free) arithmetic circuits and networks (arithmetic-boolean circuits) over the rational numbers Q .
Roughly speaking, a division-free arithmetic circuit $`\beta `$ over Q is an algorithmic device that supports a step by step evaluation of certain (output) polynomials belonging to $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$, say $`f_1,\mathrm{},f_p`$. Each step of $`\beta `$ corresponds either to an input from $`X_1,\mathrm{},X_n`$, to a constant (circuit parameter) from Q or to an arithmetic operation (addition/subtraction or multiplication). We represent the circuit $`\beta `$ by a labelled directed acyclic graph (dag). The size of this dag measures the sequential time requirements of the evaluation of the output polynomials $`f_1,\mathrm{},f_p`$ performed by the circuit $`\beta `$.
A (division-free) arithmetic network over Q is nothing else but an arithmetic circuit that additionally contains decision gates comparing rational values or checking their equality, and selector gates depending on these decision gates.
Arithmetic circuits and networks represent non–uniform algorithms, and the complexity of executing a single arithmetic operation is always counted at unit cost. Nevertheless, by means of well known standard procedures our algorithms will always be transposable to the uniform random bit model and they will be practically implementable as well. All this can be done in the spirit of the general asymptotic complexity bounds stated in Theorem 11 below.
Let us also remark that the depth of an arithmetic circuit (or network) measures the parallel time of its evaluation, whereas its size allows an alternative interpretation as ”number of processors”. In this context we would like to emphasize the particular importance of counting only nonscalar arithmetic operations (i.e.,only essential multiplications), taking Q -linear operations (in particular, additions/subtractions) for cost–free. This leads to the notion of nonscalar size and depth of a given arithmetic circuit or network $`\beta `$. It can be easily seen that the nonscalar size determines essentially the total size of $`\beta `$ (which takes into account all operations) and that the nonscalar depth dominates the logarithms of degree and height of the intermediate results of $`\beta `$.
An arithmetic circuit (or network) becomes a sequential algorithm when we play a so–called pebble game on it. By means of pebble games we are able to introduce a natural space measure in our algorithmic model and along with this, a new, more subtle sequential time measure. If we play a pebble game on a given arithmetic circuit, we obtain a so–called straight line program (slp). In the same way we obtain a computation tree from a given arithmetic network. For more details on our complexity model we refer to , , , , , and especially to (where also the implementation aspect is treated).
In the next Theorem 11 we are going to consider families of polynomials $`f_1,\mathrm{},f_p`$ belonging to $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$, for which we arrange the following assumptions and notations:
* $`f_1,\mathrm{},f_p`$ form a regular sequence in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$,
* for every $`1kp`$ the ideal $`(f_1,\mathrm{},f_k)`$ generated by $`f_1,\mathrm{},f_k`$ in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ is radical and defines a subvariety of $`\text{C}\text{ }^n`$ of dimension $`nk`$ that we denote by $`V_k:=V(f_1,\mathrm{},f_k)`$.
* the variables $`X_1,\mathrm{},X_n`$ are in generic position with respect to the polynomials $`f_1,\mathrm{},f_p`$.
Let $`W_0:=\{x\text{C}\text{ }^n|f_1(x)=\mathrm{}=f_p(x)=0\}`$ and denote by $`SingW_0`$ the singular locus of $`W_0`$. For $`1inp`$ let $`W_i`$ be the $`i`$–th formal polar variety associated with $`W_0`$ and the variables $`X_{p+i},\mathrm{},X_n`$, and let $`\stackrel{~}{W_i}:=\overline{W_iSingW_0}`$ be the $`i`$–th polar variety of $`W_0`$ in the usual sense (see Section 2 for precise definitions). Further, for $`1kp`$ we shall write $`\stackrel{~}{V}_k:=\overline{V_kSingW_0}`$. We call
$$\delta :=\mathrm{max}\{\mathrm{max}\{\mathrm{deg}\stackrel{~}{V}_k|1kp\},\mathrm{max}\{\mathrm{deg}\stackrel{~}{W_i}|1inp\}\}$$
the degree (of the real interpretation) of the polynomial equation system $`f_1,\mathrm{},f_p`$. Finally, let us make the following assumption:
* the specialized Jacobian $`J(f_1,\mathrm{},f_p)(x)`$ has maximal rank in any point $`x`$ of $`S_0:=W_0\mathrm{IR}^n=\{x\mathrm{IR}^n|f_1(x)=\mathrm{}=f_p(x)=0\}`$ and $`S_0`$ is a bounded semialgebraic set (hence, $`S_0`$ is empty or a smooth, compact real manifold of dimension $`np`$; see Section 2 for details).
###### Theorem 11
Let $`n,p,d,\delta ,L`$ and $`\mathrm{}`$ be natural numbers with $`d2`$ and $`pn`$. There exists an arithmetic network $`𝒩`$ over Q of size $`\left(\genfrac{}{}{0pt}{}{n}{p1}\right)L(nd\delta )^{O(1)}`$ and nonscalar depth $`O(n(\mathrm{log}nd+\mathrm{})\mathrm{log}\delta )`$ with the following property: Let $`f_1,\mathrm{},f_p`$ be a family of $`n`$–variate polynomials of a degree at most $`d`$ and assume that $`f_1,\mathrm{},f_p`$ are given by a division–free arithmetic circuit $`\beta `$ in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ of size $`L`$ and nonscalar depth $`\mathrm{}`$. Suppose that the polynomials $`f_1,\mathrm{},f_p`$ satisfy the conditions (i), (ii), (iii) and (iv) above. Further, suppose that the degree of the real interpretation of the polynomial system $`f_1,\mathrm{},f_p`$ is bounded by $`\delta `$ (let us now freely use the notations just introduced before).
The algorithm represented by the arithmetic network $`𝒩`$ starts from the circuit $`\beta `$ as input and decides first whether the complex variety $`\stackrel{~}{W}_{np}`$ is empty. If this is not the case, then $`\stackrel{~}{W}_{np}`$ is a zero–dimensional complex variety and the network $`𝒩`$ produces an arithmetic circuit in Q of asymptotically the same size and nonscalar depth as $`𝒩`$, which represents the coefficients of $`n+1`$ univariate polynomials $`q,p_1,\mathrm{},p_n\text{Q}\text{ }[X_n]`$ satisfying the following conditions:
$$\mathrm{deg}q=\mathrm{\#}\stackrel{~}{W}_{np},$$
$$\mathrm{max}\{\mathrm{deg}p_k|1kn\}<\mathrm{deg}q,$$
$$\stackrel{~}{W}_{np}=\{(p_1(u),\mathrm{},p_n(u))|u\text{C}\text{ },q(u)=0\}.$$
Moreover, the algorithm represented by the arithmetic network $`𝒩`$ decides whether the set $`\stackrel{~}{W}_{np}\mathrm{IR}^n`$ is empty. In this case we conclude $`S_0=W_0\mathrm{IR}^n=\mathrm{}`$. Otherwise, the network $`𝒩`$ produces at most $`\mathrm{\#}\stackrel{~}{W}_{np}\delta `$ sign sequences belonging to the set $`\{1,0,1\}`$ such that these sign sequences encode the real zeros of the polynomial $`q`$ ”à la Thom” (). In this way, namely by means of the Thom encoding of the real zeros of $`q`$ and by means of the polynomials $`p_1,\mathrm{},p_n`$, the arithmetic network $`𝒩`$ describes the finite, non–empty set
$$\stackrel{~}{W}_{np}\mathrm{IR}^n=\{(p_1(u),\mathrm{},p_n(u))|u\mathrm{IR},q(u)=0\},$$
which contains at least one representative point for each connected component of the real variety
$$S_0=\{x\mathrm{IR}^n|f_1(x)=\mathrm{}=f_p(x)=0\}.$$
Proof
We shall freely use the notations of Section 2. Any selection of indices $`1i_1<\mathrm{}<i_pn`$ and $`1j,kp`$ determines a $`p`$– minor $`M(i_1,\mathrm{},i_p)`$ and a $`(p1)`$– minor $`m(i_1,\mathrm{},i_p;j,k)`$ of the Jacobian $`J(f_1,\mathrm{},f_p)`$ in the following way: $`M(i_1,\mathrm{},i_p)`$ is the determinant of the $`(p\times p)`$– submatrix of $`J(f_1,\mathrm{},f_p)`$ with columns $`i_1,\mathrm{},i_p`$, and $`m(i_1,\mathrm{},i_p;j,k)`$ is the determinant of the matrix obtained from the former one deleting the row number $`j`$ and the column number $`i_k`$. There are $`p^2\left(\genfrac{}{}{0pt}{}{n}{p}\right)`$ such possible selections. Let us fix one of them, say $`i_1:=1,\mathrm{},i_p:=p;j:=p,k:=p`$. Then, using the notations of Section 2, we have $`m(i_1,\mathrm{},i_p;j,k)=m,M(i_1,\mathrm{},i_p)=M_p`$. Let us abbreviate $`g:=mM_p`$. From our assumptions on $`f_1,\mathrm{},f_p`$ and Theorem 8 and Theorem 10 of Section 2 we deduce the following facts: For any $`1inp`$ the polynomials $`f_1,\mathrm{},f_p,M_p,\mathrm{},M_{p+i1}`$ have degree at most $`pd`$. They generate the trivial ideal or form a regular sequence in the localized Q -algebra $`\text{Q}\text{ }[X_1,\mathrm{},X_n]_g`$. In either case the ideal generated by $`f_1,\mathrm{},f_p,M_p,\mathrm{},M_{p+i1}`$ in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]_g`$ is radical and defines a complex variety that is empty or of degree
$$\mathrm{deg}(\overline{W_iV(g)})\mathrm{deg}(\overline{W_iSingW_0})=\mathrm{deg}\stackrel{~}{W_i}\delta .$$
Moreover, by assumption, the polynomials $`f_1,\mathrm{},f_p`$ form a regular sequence in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]_g`$ and for each $`1kp`$ the ideal generated by $`f_1,\mathrm{},f_k`$ in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]_g`$ is radical and defines a complex variety of degree
$$\mathrm{deg}\overline{(V_kV(g))}\mathrm{deg}\stackrel{~}{V}_k\delta .$$
One sees easily that the polynomials $`f_1,\mathrm{},f_p,M_p,\mathrm{},M_{n1}`$ and $`g`$ can be evaluated by a division–free arithmetic circuit of size $`O(L+n^5)`$ and nonscalar depth $`O(\mathrm{log}n+\mathrm{})`$. Applying now, for each $`1inp`$, the algorithm underlying , Proposition 18 in its rational version , Theorem 19 to the system
$$f_1=0,\mathrm{},f_p=0,M_p=0,\mathrm{},M_{p+i1}=0,g0$$
we are able to check whether the particular system
$$f_1=0,\mathrm{},f_p=0,M_p=0,\mathrm{},M_{n1}=0,g0$$
has a solution in $`\text{C}\text{ }^n`$. If this is the case, then this system defines a zero–dimensional algebraic set, namely $`W_{np}V(g)`$, and the algorithm produces an arithmetic circuit $`\overline{\gamma }`$ in Q that represents the coefficients of $`n+1`$ univariate polynomials $`\overline{q},\overline{p}_1,\mathrm{},\overline{p}_n\text{Q}\text{ }[X_n]`$ satisfying the following conditions:
$$\mathrm{deg}\overline{q}=\mathrm{\#}(W_{np}V(g)),$$
$$\mathrm{max}\{\mathrm{deg}\overline{p}_k|1kn\}<\mathrm{deg}\overline{q},$$
$$W_{np}V(g)=\{(\overline{p}_1(u),\mathrm{},\overline{p}_n(u))|u\text{C}\text{ },\overline{q}(u)=0\}.$$
The algorithm is represented by an arithmetic network of size $`L(nd\delta )^{O(1)}`$ and nonscalar depth $`O(n(\mathrm{log}nd+\mathrm{})\mathrm{log}\delta )`$, and the circuit $`\overline{\gamma }`$ has asymptotically the same size and nonscalar depth. Running this procedure for each selection $`1i_1<\mathrm{}<i_pn`$ and $`1j,kp`$ we obtain an arithmetic network $`𝒩_0`$ of size $`p^2\left(\genfrac{}{}{0pt}{}{n}{p}\right)L(nd\delta )^{O(1)}=\left(\genfrac{}{}{0pt}{}{n}{p1}\right)L(nd\delta )^{O(1)}`$ and nonscalar depth $`O(n(\mathrm{log}nd+\mathrm{})\mathrm{log}\delta )`$, which decides whether $`\stackrel{~}{W}_{np}=W_{np}SingW_0`$ is empty. Suppose that this is not the case. Then $`𝒩_0`$ describes locally the variety $`\stackrel{~}{W}_{np}`$, which is now zero-dimensional. Each local description of $`\stackrel{~}{W}_{np}`$ contains an arithmetic circuit representation of the coefficients of the minimal polynomial of the variable $`X_n`$ with respect to the corresponding local piece of $`\stackrel{~}{W}_{np}`$ . Moreover, one easily obtains the same type of information for any linear form $`X_i+X_n`$ and any variable $`X_i`$ with $`1i<n`$. One multiplies now all minimal polynomials of the variable $`X_n`$ obtained in this way. Making this product squarefree (see e.g , Lemma 12) one obtains the polynomial $`q`$ of the statement of Theorem 11. Doing the same thing for the minimal polynomials of each linear form $`X_i+X_n`$ and each variable $`X_i`$ with $`1i<n`$, yields by means of , Lemma 26, the polynomials $`p_1,\mathrm{},p_n`$ of the statement of Theorem 11. All this can be done by means of an arithmetic network $`𝒩_1`$, which extends $`𝒩_0`$ and has asympotically the same size and nonscalar depth. The desired arithmetic network $`𝒩`$ is now obtained from $`𝒩_1`$ in the same way as in the proof , Theorem 8, namely as follows: applying the main algorithm of or and adding suitable comparison gates for rational numbers, we extend $`𝒩_1`$ to a new arithmetical network $`𝒩`$ of asymptotically the same size and depth, such that $`𝒩`$ decides whether the univariate polynomial $`q`$ has a real zero. If this is the case, the network $`𝒩`$ enumerates the existing real zeros of $`q`$, encoding them ”à la Thom” (). If $`q`$ has no real zero we conclude $`S_0=\mathrm{}`$. Otherwise, the network $`𝒩`$ encodes all real zeros of $`q`$ by means of $`\mathrm{\#}\stackrel{~}{W}_{np}\delta `$ sign sequences belonging to the set $`\{1,0,1\}`$. This encoding and the polynomials $`p_1,\mathrm{},p_n`$ describe now the set $`\stackrel{~}{W}_{np}\mathrm{IR}^n=\{(p_1(u),\mathrm{},p_n(u))|u\mathrm{IR},q(u)=0\}`$ that contains a representative point for each connected component of $`S_0`$. $`\mathrm{}`$
###### Remark 12
1. Using the refined algorithmic techniques of or it is not too difficult to see that for inputs $`f_1,\mathrm{},f_p`$ represented by straight–line programs of length $`T`$ and space $`S`$ the arithmetic network $`𝒩`$ can be converted into an algebraic computation tree which solves the algorithmic problem of Theorem 11 in time $`O((Tdn^2+n^5)\delta ^3\mathrm{log}^3\delta \mathrm{log}^2\mathrm{log}\delta )`$ and space $`O(Sdn\delta ^2)`$.
2. The smooth, compact hypersurface case (with $`p:=1`$) of Theorem 11 corresponds exactly to , Theorem 8.
3. Let $`J(f_1,\mathrm{},f_p)^T`$ denote the transposed matrix of the Jacobian $`J(f_1,\mathrm{},f_p)`$ of the polynomials $`f_1,\mathrm{},f_p`$ in the statement of Theorem 11 and let
$$𝒟:=detJ(f_1,\mathrm{},f_p)J(f_1,\mathrm{},f_p)^T.$$
From the well–known Cauchy–Binet formula one deduces easily that, with the notations of Section 2, the identity
$$𝒟=\underset{1i_1<\mathrm{}<i_pn}{}\stackrel{2}{det}M(i_1,\mathrm{},i_p)$$
holds. Replacing now, in the statement and the proof of Theorem 11 for $`1inp`$, the polar variety $`\stackrel{~}{W}_i`$ by $`\widehat{W}_i:=\overline{W_iV(𝒟)}`$ and the parameter $`\delta `$ by
$$\widehat{\delta }:=\mathrm{max}\{\mathrm{max}\{\mathrm{deg}\stackrel{~}{V}_k|1kp\},\mathrm{max}\{\mathrm{deg}\widehat{W_i}|1inp\}\}$$
one obtains a somewhat improved complexity result, since $`\widehat{\delta }\delta `$ holds.
Let us now suppose that the polynomials $`f_1,\mathrm{},f_p\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ satisfy the conditions (i), (ii), (iii), (iv) above. Unfortunately, the complexity parameter $`\delta `$ of Theorem 11 is strongly related to the complex degrees of the polar varieties $`\stackrel{~}{W}_1,\mathrm{},\stackrel{~}{W}_{np}`$ of $`W_0=\{x\text{C}\text{ }^n|f_1(x)=\mathrm{}=f_p(x)=0\}`$ and not to their real degrees. Under some additional algorithmic assumptions, which we are going to explain below, we may replace the complexity parameter $`\delta `$ by a smaller one that measures only the real degrees of the polar varieties $`\stackrel{~}{W}_1,\mathrm{},\stackrel{~}{W}_{np}`$. We shall call this new complexity parameter the real degree of the equation system $`f_1,\mathrm{},f_p`$ and denote it by $`\delta ^{}`$.
Let $`1kp`$ and let us consider the decomposition of the intermediate variety $`\stackrel{~}{V}_k`$ into irreducible components with respect to the Q -Zariski topology of $`\text{C}\text{ }^n`$ say $`\stackrel{~}{V}_k=C_1\mathrm{}C_s`$. We call an irreducible component $`C_r,\mathrm{\hspace{0.33em}1}rs`$, real if $`C_r\mathrm{IR}^n`$ contains a smooth point of $`C_r`$. The union of all real irreducible components of $`\stackrel{~}{V}_k`$ is called the real part of $`\stackrel{~}{V}_k`$ and denoted by $`V_k^{}`$. We call $`\mathrm{deg}V_k^{}`$ the real degree of the intermediate variety $`\stackrel{~}{V}_k`$. Similarly, we introduce for every $`1inp`$ the real part $`W_i^{}`$ of the polar variety $`\stackrel{~}{W}_i`$ and its real degree $`\mathrm{deg}W_i^{}`$. Finally, we define the real degree of the equation system $`f_1,\mathrm{},f_p`$ as
$$\delta ^{}:=\mathrm{max}\{\mathrm{max}\{\mathrm{deg}V_k^{}|1kp\},\mathrm{max}\{\mathrm{deg}W_i^{}|1inp\}\}.$$
Now, we are going to restate the main outcome of Theorem 11 in terms of the new complexity parameter $`\delta ^{}`$. For this purpose we have to include the following two subroutines in our algorithmic model:
* the first subroutine we need is a factorization algorithm for univariate polynomials over Q . In the bit complexity model the problem of factorizing univariate polynomials over Q is known to be of polynomial time complexity , whereas in the arithmetic model we are considering here this question is more intricate . In the extended complexity model we are going to consider here, the cost of factorizing a univariate polynomial of degree $`D`$ over Q (given by its coefficients) is accounted as $`D^{O(1)}`$.
* the second subroutine allows us to discard non-real irreducible components of the occurring complex polar varieties. This second subroutine starts from a straight-line program for a single polynomial in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ as input and decides whether this polynomial has a real zero (however, without actually exhibiting it if there is one). Again this subroutine is taken into account at polynomial cost.
We call an arithmetic network over Q extended if it contains extra nodes corresponding to the first and second subroutine.
Modifying our algorithmic model in this way, we are able to formulate the following complexity result, which generalizes , Theorem 12 and improves the complexity outcome of our previous Theorem 11.
###### Remark 13
Let $`n,p,d,\delta ^{},L`$ and $`\mathrm{}`$ be natural numbers with $`d2`$ and $`pn`$. There exists an extended arithmetic network $`𝒩^{}`$ over Q of size $`\left(\genfrac{}{}{0pt}{}{n}{p1}\right)L(nd\delta ^{})^{O(1)}`$ with the following property: Let $`f_1,\mathrm{},f_p`$ be a family of $`n`$–variate polynomials of a degree at most $`d`$ and assume that $`f_1,\mathrm{},f_p`$ are given by a division–free arithmetic circuit $`\beta `$ in $`\text{Q}\text{ }[X_1,\mathrm{},X_n]`$ of size $`L`$. Suppose that the polynomials $`f_1,\mathrm{},f_p`$ satisfy the conditions (i), (ii), (iii), and (iv) contained in the formulation of Theorem 11. Let us now freely use the notations introduced in the present section. Assume that the real variety $`S_0=\{x\mathrm{IR}^n|f_1(x)=\mathrm{}=f_p(x)=0\}`$ is not empty and that the real degree of the polynomial system $`f_1,\mathrm{},f_p`$ is bounded by $`\delta ^{}`$. The algorithm represented by the arithmetic network $`𝒩^{}`$ starts from the circuit $`\beta `$ as input and decides first whether the complex variety $`W_{np}^{}`$ is empty. If this is not the case, then $`W_{np}^{}`$ is a zero–dimensional complex variety and the network $`𝒩^{}`$ produces an arithmetic circuit in Q of asymptotically the same size as $`𝒩^{}`$,which represents the coefficients of $`n+1`$ univariate polynomials $`q^{},p_1^{},\mathrm{},p_n^{}\text{Q}\text{ }[X_n]`$ satisfying the conditions
$$\mathrm{deg}q^{}=\mathrm{\#}W_{np}^{},$$
$$\mathrm{max}\{\mathrm{deg}p_k^{}|1kn\}<\mathrm{deg}q^{},$$
$$W_{np}^{}=\{(p_1^{}(u),\mathrm{},p_n^{}(u))|u\text{C}\text{ },q^{}(u)=0\}.$$
Each over Q irreducible component of the complex variety $`W_{np}^{}`$ contains at least one real point characterized by an irreducible factor of the polynomial $`q^{}`$. The algorithm represented by the network $`𝒩^{}`$ returns all these points in a codification ”à la Thom”. Moreover, the non–empty set $`W_{np}^{}\mathrm{IR}^n`$ contains at least one representative point for each connected component of the real variety $`S_0`$.
The proof of this remark is a straight–forward adaptation of the arguments of the proof of , Theorem 12 (which treats only the hypersurface case with $`p:=1`$) to the arguments of Theorem 11 above. Therefore, we omit this proof.
Let us finally observe that the practical relevance of the complexity outcome of Remark 6 is highly hypothetical, because it depends on the strong assumption that extended arithmetical networks are realizable by performant, programmable algorithms. Nevertheless, by means of Remark 6, we wish to underline the importance of the search for efficient procedures that realize the first and second subroutine introduced as extra nodes in our complexity model of extended arithmetic networks.
|
warning/0005/astro-ph0005316.html
|
ar5iv
|
text
|
# Testing theories that predict time variation of fundamental constants
## 1 Introduction
Time variation of fundamental constants has plenty of theoretical and experimental research since the large number hypothesis (LNH) proposed by Dirac (1937). The great predictive power of the LNH, induced a large number of research papers and suggested new sources of variation. Among them, the attempt to unify all fundamental interactions resulted in the development of multidimensional theories like Kaluza-Klein (Kaluza, 1921; Klein, 1926; Chodos and Detweiler, 1980; Marciano, 1984) and superstring ones (Damour and Polyakov, 1994) which predict not only energy dependence of the fundamental constants but also dependence of their low-energy limits on cosmological time. In such theories, the temporal variation of fundamental constants is related with the variation of the extra compact dimensions.
Following a different path of research, Beckenstein (1982) proposed a theoretical framework to study the fine structure constant variability based on very general assumptions: covariance, gauge invariance, causality and time-reversal invariance of electromagnetism , as well as the idea that the Planck-Wheeler length $`\left(10^{33}cm\right)`$ is the shortest scale allowable in any theory.
Different versions of the theories mentioned above predict different time behaviours for the fundamental constants. Thus, experimental bounds on the variation of fundamental constants are an important tool to check the validity of such theories (Marciano, 1984; Chodos and Detweiler, 1980; Beckenstein, 1982).
The experimental research can be grouped into astronomical and local methods. The latter ones include geophysical methods such as the natural nuclear reactor that operated about $`\mathrm{1.8\; 10}^9`$ in Oklo, Gabon (Damour and Dyson, 1996), the analysis of natural long-lived $`\beta `$ decayers in geological minerals and meteorites (Sisterna and Vucetich, 1990) and laboratory measurements such as comparisons of rates between clocks with different atomic number (Prestage, Toelker and Maleki, 1995). The astronomical methods are based mainly in the analysis of spectra form high-redshift quasar absorption systems (Drinkwater et al., 1998; Webb et al, 1999, 2001; Murphy et al., 2001a; Cowie and Songaila, 1995; Bahcall, Sargent and Schmidt, 1967). Besides, other constraints can be derived from primordial nucleosynthesis (Bernstein, Brown and Feinberg, 1988) and the Cosmic Microwave Background (CMB) fluctuation spectrum (Battye, Crittenden and Weller, 2001; Avelino et al, 2000; Landau, Harari and Zaldarriaga, 2001).
Although, most of the previous mentioned experimental data gave null results, (Webb et al, 1999), reported a significantly different measurement of the time variation of the fine structure constant, which was confirmed recently (Webb et al, 2001; Murphy et al., 2001a). This suggests an examination of the available experimental results in the context of typical theories predicting time variation of fundamental constants.
Thus, in this work, we consider several astronomical and local bounds on time variation of fundamental constants in the framework of two Kaluza-Klein-like late time solutions (Marciano, 1984; Bailin and Love, 1987) and some particular cases of Beckenstein theory (Beckenstein, 1982). In particular we put bounds on the free parameters of the different models, the size of the extra dimensions in the first case, and the parameters $`l`$ and $`\gamma `$ of Beckenstein’s theory. Besides, the consistency of experimental data with a given family of theories can be checked.
The paper is organized as follows: In section II we describe briefly the models we want to test, in section III we describe the experimental constraints, we will use to check our models, in section IV we present our results and briefly discuss our conclusions.
## 2 Theoretical models predicting time variation of fundamental constants
### 2.1 Kaluza-Klein-like models
The basic idea of Kaluza-Klein theories is to enlarge space-time to $`4+D`$ dimensions in such a way that the D extra spatial dimensions form a very small compact manifold with mean radius $`R_{KK}`$.
So, the metric in $`4+D`$ dimensions can be written :
$$dS^2=dt^2r^2\left(t\right)g_{mn}R_{KK}^2\left(t\right)g_{uv}$$
(1)
where $`g_{mn}`$ is the metric of an $`S^3`$of unit radius , $`r\left(t\right)`$ is the scale factor of the ordinary space, $`g_{uv}`$ is the metric of an $`S^D`$ of unit radius and $`R_{KK}\left(t\right)`$ is the scale factor of the internal space.
In Kaluza-Klein theories, gauge fields of the Standard Model of Fundamental Interactions are related to the $`g_{\mu \nu }`$ elements that connect the internal dimensions with the usual $`3+1`$ space-time. The gauge coupling constants are related to the “internal” scale of the extra dimensions through one or more scalar fields (Weinberg, 1983).
In some models, the “internal” dimensions are small compared to the large “ordinary” dimensions. However, at the Planck time, the characteristic size of both internal and external dimensions are likely to be the same. The cosmological evolution which determines the way in which the extra dimensions are compactified depends on how many extra dimensions are taken and on the energy-momentum tensor considered: radiation, monopoles, cosmological constant, etc.
The generalized Einstein equations can be written as follows (Kolb and Turner, 1990):
$$R_{MN}=8\pi \stackrel{~}{G}\left[T_{MN}\frac{1}{D+2}g_{MN}T_P^P\frac{1}{D+2}\frac{\stackrel{~}{\mathrm{\Lambda }}}{8\pi \stackrel{~}{G}}g_{MN}\right]$$
(2)
where $`\stackrel{~}{G}`$ is the gravitational constant in $`4+D`$ dimensions and $`\stackrel{~}{\mathrm{\Lambda }}`$ is a cosmological constant in $`4+D`$ dimensions.
The evolution of the extra dimensions with cosmological time is related with the time variation of fundamental constants through the equation (Kaluza, 1921; Klein, 1926; Marciano, 1984; Weinberg, 1983):
$$\alpha _i\left(M_{KK}\right)=\frac{K_iG}{R_{KK}^2}=K_iGM_{KK}^2$$
(3)
where $`\alpha _i\left(M_{KK}\right)`$,i=1,2,3 are the coupling constants of $`U(1)`$, $`SU(2)`$ and $`SU(3)`$ for a typical energy $`R_{KK}=\frac{1}{M_{KK}}`$. We assume as usual, the existence of a GUT energy scale $`\mathrm{\Lambda }_{GUT}`$ beyond which all these constants merge in only one $`\alpha _i`$. The $`K_i`$ are numbers that depend on the $`D`$ dimensional topology.
The expressions for the gauge coupling constants at different energies are related through the group renormalization equation (Marciano, 1984):
$$\alpha _i^1\left(E_1\right)=\alpha _i^1\left(E_2\right)\frac{1}{\pi }\underset{j}{}C_{ij}\left[\mathrm{ln}\left(\frac{E_2}{m_j}\right)+\theta \left(E_1m_j\right)\mathrm{ln}\left(\frac{m_j}{E_1}\right)\right]$$
(4)
So, we can find the low-energy limit for the gauge coupling constants using eq.(4) twice:
$`E_1`$ $`=`$ $`\mathrm{\Lambda }_{GUT}E_2=M_{KK}`$ (5)
$`E_1`$ $`=`$ $`M_WE_2=\mathrm{\Lambda }_{GUT}`$
Inserting eq.(3) we obtain:
$$\alpha _1^1\left(M_W\right)=\frac{KG}{R_{KK}^2}\frac{76}{6\pi }\mathrm{ln}\left(\frac{R_{KK}^1}{\mathrm{\Lambda }_{GUT}}\right)+\frac{2}{\pi }\mathrm{ln}\left(\frac{\mathrm{\Lambda }_{GUT}}{M_W}\right)$$
(6)
$$\alpha _2^1\left(M_W\right)=\frac{KG}{R_{KK}^2}\frac{76}{6\pi }\mathrm{ln}\left(\frac{R_{KK}^1}{\mathrm{\Lambda }_{GUT}}\right)\frac{5}{3\pi }\mathrm{ln}\left(\frac{\mathrm{\Lambda }_{GUT}}{M_W}\right)$$
(7)
$$\alpha _3^1\left(M_W\right)=\frac{KG}{R_{KK}^2}\frac{76}{6\pi }\mathrm{ln}\left(\frac{R_{KK}^1}{\mathrm{\Lambda }_{GUT}}\right)\frac{7}{2\pi }\mathrm{ln}\left(\frac{\mathrm{\Lambda }_{GUT}}{M_W}\right)$$
(8)
In this way we get expressions for the gauge coupling constants depending on $`R_{KK}`$ and $`\mathrm{\Lambda }_{GUT}`$. In order to compare equations (6), (7) and (8) with experimental and observational values, we still should calculate the adjustment for energies $`1`$ GeV. However, since this adjustment is very small, we will not consider it.
The gauge coupling constants are related with the fine structure constant $`\alpha `$, the QCD energy scale $`\mathrm{\Lambda }_{QCD}`$ and the Fermi coupling constant $`G_F`$ through the following equations:
$$\alpha ^1\left(E\right)=\frac{5}{2}\alpha _1^1\left(E\right)+\alpha _2^1\left(E\right)$$
(9)
$$\mathrm{\Lambda }_{QCD}=E\mathrm{exp}\left[\frac{2\pi }{7}\alpha _3^1\left(E\right)\right]$$
(10)
$$G_F=\frac{\pi \alpha _2\left(M_W\right)}{\sqrt{2}M_W^2}$$
(11)
It has been shown that Kaluza-Klein equation are either non-integrable, or their solutions lack of physical interest (Helmi and Vucetich, 1995). However, several non-exact solutions of eq.(2) have been analized in the literature (see Bailin and Love (1987); Kolb and Turner (1990) and references therein).
For the purposes of this paper, though, we are interested in typical late time solutions since the data we work with belong to times not earlier than nucleosynthesis. Thus, we consider models where the scale factor of the Universe behaves as in a flat Robertson-Walker space-time with and without cosmological constant and the radius of the internal dimensions behaves as the following schematic solutions motivated in Marciano (1984); Bailin and Love (1987):
$$R_{KK}\left(t\right)R_0+\mathrm{\Delta }R\left(1\mathrm{cos}\left[\omega \left(tt_0\right)\right]\right)$$
(12)
$$R_{KK}\left(t\right)R_0+\mathrm{\Delta }R\left[\frac{t_0}{t}\right]^{3/4}$$
(13)
where $`R_0=R_{KK}\left(t_{Planck}\right)R_{Planck}`$. We expect that typical solutions of Kaluza-Klein cosmologies behave asymptotically like eqs.(12) and (13) with $`\mathrm{\Delta }R<<R_0`$ and $`\omega `$ depending on the details of the model. We will refer to solution 12 as generic model 1 and to solution 13 as generic model 2. Generic model 1 is similar in shape to the variation in $`\alpha `$ reported by Webb et al (2001). Indeed, it predicts a null variation of the fine structure constant today and a greater variation in the past.
Thus, the free parameter in all Kaluza-Klein-like models will be : $`\frac{\mathrm{\Delta }R}{R_0}`$ and we will take as usual $`\mathrm{\Lambda }_{GUT}=10^{16}GeV`$. Table 3 shows the cosmological model and the values of $`\omega `$ considered for each particular model.
### 2.2 Beckenstein models
As we have mentioned above, Beckenstein (1982) proposed a framework for the fine structure constant $`\alpha `$ variability based on very general assumptions such us: covariance, gauge invariance, causality and time-reversal invariance of electromagnetism , as well as the idea that the Planck-Wheeler length $`\left(10^{33}cm\right)`$ is the shortest scale allowable in any theory.
He obtained the following equation for the temporal variation of $`\alpha `$:
$$\left(\frac{a^3\dot{\epsilon }}{\epsilon }\right)^.=a\left(t\right)^3\varsigma \left(\frac{l^2}{\mathrm{}c}\right)\rho _mc^4$$
(14)
where $`\epsilon =\left(\frac{\alpha }{\alpha _{today}}\right)^{\frac{1}{2}}`$, $`l`$ is a length scale of the theory, $`\rho _m`$ is the total rest mass density of matter, $`a\left(t\right)`$ is the expansion scale factor and $`\varsigma `$ is a dimensionless parameter which measures the fraction of mass in the form of Coulomb energy of an average nucleon, compared to the free proton mass (Beckenstein (1982) assumed that $`\varsigma `$ is constant and equal to $`1.3\times 10^2`$).
In an expanding Universe where $`\rho _m=\frac{3H_0^2}{8\pi G}\left[\frac{a\left(t_0\right)}{a\left(t\right)}\right]^3`$, we obtain:
$$\frac{\dot{\epsilon }}{\epsilon }=\varsigma \left(\frac{l^2c^3}{\mathrm{}}\right)\rho _m\left(tt_c\right)$$
(15)
where $`t_c`$ is an integration constant. We consider a flat model with cosmological constant where the scale factor varies as:
$$a\left(t\right)=a\left(t_0\right)\left(\frac{\mathrm{\Omega }_m}{\mathrm{\Omega }_\mathrm{\Lambda }}\right)^{\frac{1}{3}}\left[\mathrm{sinh}\left(\frac{3}{2}\mathrm{\Omega }_\mathrm{\Lambda }^{1/2}H_0t\right)\right]^{\frac{2}{3}}$$
(16)
Integrating eq.(15), we obtain the time variation of the fine structure constant as follows:
$$\frac{\mathrm{\Delta }\alpha }{\alpha }=\frac{3\varsigma }{8\pi }\left(H_0t_0^1\right)^2\left(\frac{l}{L_p}\right)^2\left[\begin{array}{c}\beta \mathrm{coth}\beta \frac{t}{t_0}\beta \mathrm{coth}\left(\beta \frac{t}{t_0}\right)+\mathrm{ln}\left(\frac{\mathrm{sinh}\left(\beta \frac{t}{t_0}\right)}{\mathrm{sinh}\left(\beta \right)}\right)\\ +\gamma \left(\beta \mathrm{coth}\left(\beta \frac{t}{t_0}\right)\beta \mathrm{coth}\beta \right)\end{array}\right]$$
(17)
with
$$\mathrm{coth}\beta =\mathrm{\Omega }_\mathrm{\Lambda }^{\frac{1}{2}}$$
where $`L_p=\left(\frac{G\mathrm{}}{c^3}\right)^{\frac{1}{2}}.`$ In all cases the integration constant is such that $`\epsilon \left(t_0\right)=1`$ and $`\mathrm{\Omega }_m+\mathrm{\Omega }_\mathrm{\Lambda }=1`$
Table 4 shows the cosmological parameters for the models we use to test this theory. The free parameters in this models are $`L=\frac{l}{L_p}`$ and $`\gamma `$.
## 3 Bounds from astronomical and geophysical data
In this section, we make a critical discussion of the rather heterogeneous data set we use to test our models.
### 3.1 The Oklo Phenomenon
One of the most stringent limits on time variation of fundamental constants follows from an analysis of isotope ratios of $`{}_{}{}^{149}\mathrm{Sm}/^{147}\mathrm{Sm}`$ in the natural uranium fission reactor that operated $`1.8\times 10^9`$ yr ago at the present day site of the Oklo mine in Gabon, Africa (Schlyakter, 1976; Damour and Dyson, 1996). From an analysis of nuclear and geochemical data, the operating conditions of the reactor could be reconstructed and the thermal neutron capture cross sections of several nuclear species measured. In particular, a shift in the lowest lying resonance level in $`{}_{}{}^{149}\mathrm{Sm}:\mathrm{\Delta }=E_r^{149(\mathrm{Oklo})}E_r^{149(\mathrm{now})}`$ can be derived from a shift in the neutron capture cross section of the same nucleus (Schlyakter, 1976; Damour and Dyson, 1996). We know that we can translate the shift in $`\mathrm{\Delta }`$ into a bound on a possible difference between the values of $`\alpha `$ and $`G_F`$ during the Oklo phenomenon and their value now. Damour and Dyson (1996) derived bounds on $`\alpha `$ and $`G_F`$ separately; here we consider both variations at the same time as follows:
$$\mathrm{\Delta }=\alpha \frac{E_r}{\alpha }\frac{\mathrm{\Delta }\alpha }{\alpha }+G_F\frac{E_r}{G_F}\frac{\mathrm{\Delta }G_F}{G_F}$$
(18)
where $`\mathrm{\Delta }\alpha =\alpha ^{Oklo}\alpha ^{now}`$ and $`\mathrm{\Delta }G_F=G_F^{Oklo}G_F^{now}`$. The value of $`\mathrm{\Delta }`$ is shown in Table 1. Finally, using the values of $`\mathrm{\Delta },\alpha \frac{E_r}{\alpha },G_F\frac{E_r}{G_F}`$ from Damour and Dyson (1996), we can relate $`\mathrm{\Delta }`$ with $`\frac{\mathrm{\Delta }\alpha }{\alpha }`$ and $`\frac{\mathrm{\Delta }G_F}{G_F}`$ (see first entry in Table 2).
### 3.2 Long-lived $`\beta `$ decayers
The half-life of long-lived $`\beta `$ decayers such $`{}_{}{}^{187}\mathrm{Re},^{40}\mathrm{K},^{87}\mathrm{Rb}`$ has been determined either in laboratory measurements or by comparison with the age of meteorites, as found from $`\alpha `$ decay radioactivity analysis. Sisterna and Vucetich (1990) have derived a relation between the shift in the half-life of three long lived $`\beta `$ decayers and a possible variation between the values of the fundamental constants $`\alpha ,\mathrm{\Lambda }_{QCD}`$ and $`G_F`$ at the age of the meteorites and their value now (see entries 2,3 and 4 of Table 2).
The values of $`\frac{\mathrm{\Delta }\lambda }{\lambda }`$ for $`{}_{}{}^{187}\mathrm{Re}`$ , $`{}_{}{}^{40}\mathrm{K}`$, $`{}_{}{}^{87}\mathrm{Rb}`$ are respectively shown in entries 2, 3, and 4 in Table 1 where $`\mathrm{\Delta }=\frac{\mathrm{\Delta }\lambda }{\lambda }`$ and $`\mathrm{\Delta }\lambda =\lambda (t=5.535\times 10^9)\lambda \left(t=t_0=1.0035\times 10^{10}\right).`$
### 3.3 Laboratory experiments
The best limit on $`\alpha `$ variation, comes from a laboratory experiment (Prestage, Toelker and Maleki, 1995); it is a limit on a present day variation of $`\alpha `$. The experiment is based on a comparison of rates between clocks based on hyperfine transitions in atoms with different atomic number . H-maser and Hg+ clocks have a different dependence on $`\alpha `$ since their relativistic contributions are of order $`\left(\alpha Z\right)^2`$. The result of a 140 day clock day comparison between an ultrastable frequency standard based on Hg+ ions confined to a linear ion trap and a cavity tuned H maser (Prestage, Toelker and Maleki, 1995) is shown in Table 1 where $`\mathrm{\Delta }=\frac{\mathrm{\Delta }\alpha }{\alpha }`$.
### 3.4 Quasar absorption systems
Quasar absorption systems present ideal laboratories to search for any temporal variation in the fundamental constants. The continuum spectrum of a quasar was formed at an epoch corresponding to the redshift $`z`$ of main emission details specified by the relationship $`\lambda _{obs}=\lambda _{lab}\left(1+z\right)`$. Quasar spectra of high redshift show the absorption resonance lines of the alkaline ions like CIV, MgII, FeII, SiIV and others, corresponding to the $`S_{1/2}P_{3/2}\left(\lambda _1\right)`$ and $`S_{1/2}P_{1/2}\left(\lambda _2\right)`$ transitions. The relative magnitude of the fine splitting of the corresponding resonance lines is proportional to the square of the fine structure constant $`\alpha `$ to lowest order in $`\alpha `$.
$$\frac{\mathrm{\Delta }\lambda }{\lambda }=\frac{\lambda _1\lambda _2}{\lambda }\alpha ^2$$
(19)
Therefore, any change in $`\alpha `$ will result in a corresponding change in $`\mathrm{\Delta }\lambda `$ in the separation of the doublets of the quasar as follows:
$$\frac{\mathrm{\Delta }\alpha }{\alpha }=\frac{1}{2}\left[\frac{\left(\frac{\mathrm{\Delta }\lambda }{\lambda }\right)_z}{\left(\frac{\mathrm{\Delta }\lambda }{\lambda }\right)_{now}}1\right]$$
Cowie and Songaila (1995), Varshalovich, Panchuk and Ivanchik (1996) and Murphy et al. (2001b) have applied this method to SiIV doublet absorption lines systems at different redshifts ($`2.5<z<3.33`$) to find the values shown in entries 6 to 10 of table 1 where $`\mathrm{\Delta }=\frac{\mathrm{\Delta }\alpha }{\alpha }.`$
Webb et al (1999) have improved this method comparing transitions of different species, with widely differing atomic masses. As mentioned before, this is the only data consistent with a time varying fine structure constant. In turn, recent work (Webb et al, 2001; Murphy et al., 2001a) including new optical data confirms their previous results. The values of $`\frac{\mathrm{\Delta }\alpha }{\alpha }`$ at redshift $`z=1.2`$, $`z=2.7`$ and $`z=2.5`$ are respectively shown in entries 11, 12 and 13 of Table 1.
Moreover, the ratio of frequencies of the hyperfine 21 cm absorption transition of neutral hydrogen $`\nu _a`$ to an optical resonance transition $`\nu _b`$ is proportional to $`x=\alpha ^2g_p\frac{me}{mp}`$ where $`g_p`$ is the proton $`g`$ factor. Thus, a change of this quantity will result in a difference in the redshift measured from 21 cm and optical absorption lines as follows:
$$\frac{\mathrm{\Delta }x}{x}=\frac{z_{opt}z_{21}}{\left(1+z\right)}$$
(20)
So, combining the measurements of optical and radio redshift, a bound on $`x`$ can be obtained.
The upper bounds on $`x`$ obtained by Cowie and Songaila (1995) at redshift $`z=1.776`$ are shown in Table 1 where $`\mathrm{\Delta }=\frac{\mathrm{\Delta }x}{x}`$. The relationship between $`\frac{\mathrm{\Delta }x}{x}`$ and the variation of $`\alpha `$, $`G_F`$ and $`\mathrm{\Lambda }_{QCD}`$ is shown in table 2. Other bounds on $`x`$ were obtained by Wolfe and Davis (1979) at redshift $`z=0.69`$ (entry 15 of Table 1) and Wolfe, Brown and Roberts (1976) at redshift $`z=0.52`$ (entry 16 of Table 1)
On the other hand, the ratio of the rotational transition frequencies of diatomic molecules such as CO to the 21 cm hyperfine transition in hydrogen is proportional to $`y=g_p\alpha ^2`$. Thus, any variation in $`y`$ would be observed as a difference in the redshifts measured from 21 cm and molecular transition lines:
$$\frac{\mathrm{\Delta }y}{y}=\frac{z_{mol}z_{21}}{\left(1+z\right)}$$
(21)
Murphy et al. (2001c) have placed upper limit on $`y`$ at redshift $`z=0.25`$ and at redshift $`z=0.68`$. The observed values are shown in entries 17 and 18 of Table 1, where $`\mathrm{\Delta }=\frac{\mathrm{\Delta }y}{y}`$. Entries 17 and 18 of Table 2 relate $`\frac{\mathrm{\Delta }y}{y}`$ with the variation of $`\alpha `$.
Finally, observations of molecular hydrogen in quasar absorption systems can be used to set bounds on the evolution of $`\mu =\frac{m_e}{m_p}`$. The most stringent bounds established by Pothekin et al (1998) are shown in entry 19 of Table 2.
### 3.5 Nucleosynthesis
Primordial nucleosynthesis also provides a bound on the variation of fundamental constants. A didactical analysis of <sup>4</sup>He production can be found in Bernstein, Brown and Feinberg (1988). At the conclusion of the big-bang nucleosynthesis the <sup>4</sup>He mass fraction of the total baryonic mass is given by (Bernstein, Brown and Feinberg, 1988):
$$Y=2\mathrm{exp}[\frac{t_c}{\tau }]X\left(t_F\right)$$
(22)
where $`t_c`$ is the neutron capture time, $`\tau `$ is the neutron mean life and $`X\left(t_F\right)`$ is ratio of the neutron to total baryon number at the time where the baryons become uncoupled from the leptons (freeze-out time).
In appendix I, we derive the following expression for the change in the helium abundance $`\mathrm{\Delta }Y`$ brought about by changes in the fundamental constants:
$$\frac{\mathrm{\Delta }Y}{Y}=0.74\frac{\mathrm{\Delta }R_{KK}}{R_{KK}}+0.64\frac{\mathrm{\Delta }G_F}{G_F}+1.76\frac{\mathrm{\Delta }\alpha }{\alpha }0.3\frac{\mathrm{\Delta }\mathrm{\Lambda }_{QCD}}{\mathrm{\Lambda }_{QCD}}$$
(23)
### 3.6 Cosmic Microwave Background
Any variation of the fine structure constant $`\alpha `$ alters the physical conditions at recombination and therefore changes the cosmic microwave background (CMB) fluctuation spectrum. Moreover, the fluctuacion spectrum of CMB is sensitive to many cosmological parameters such as the density of barionic and dark matter, the Hubble constant and the index of primordial spectral fluctuations. Recently, different independent analysis (Battye, Crittenden and Weller, 2001; Avelino et al, 2000; Landau, Harari and Zaldarriaga, 2001) showed that the recent published data of Boomerang and Maxima are better fitted with a varying fine structure constant and a density of baryonic matter closer to nucleosynthesis bounds. The same authors established a bound on $`\alpha `$ variation at the epoch at which neutral hydrogen formed (see entry 21 in Table 1).
## 4 Results and Discussion
From the data rewiewed in the last section, we have performed a statistical analysis working on $`\chi ^2`$ function with MINUIT to compute the best-fit parameter values and uncertainties including correlations between parameters.
For the Kaluza-Klein like models, results within $`99\%`$ of confidence level $`\left(3\sigma \right)`$ are shown in table 3. For the models derived from Beckenstein’s proposal we obtain results with $`90\%`$ of confidence level (see table 4). The contours of the likelihood functions for Beckenstein’s models in regions of 90 % and 70 % of confidence level are shown in figures 1 and 2.
The values of the free parameters obtained are coincident within uncertainties for the Kaluza-Klein like models (table3) and for Beckenstein’s models (table 4). Besides, the values obtained are consistent with theoretical supposition $`\mathrm{\Delta }R<<R_0`$ for Kaluza-Klein like models, but they disagree with the supposition $`l>L_p`$ implied in Beckenstein’s framework.
Thus, the present available data set, considered within Bekenstein’s framework, is capable to rule out $`\alpha `$ variability, while the original paper had to recourse to Eötvös-like experiments to achieve the same result. Livio and Stiavelli (1998) have also analyzed $`\alpha `$ variation in the context of Bekenstein’s theory. Our results are in agreement with their analysis, even though they didn’t allow both free parameters of the theory: $`\frac{l}{L_p}`$ and $`\gamma `$ to vary independently.
However, it should be noted that Beckenstein’s framework is very similar to the dilatonic sector of string theory, and it has been pointed out that in the context of string theories (Bachas, 2000; Antoniadis and Pioline, 1999) there is no need for an universal relation between the Planck and the string scale.
Finally, our results are consistent with no time variation of fundamental constants over cosmological time in agreement most of the experimental results. Indeed, excluding the Webb et al. data points from our fits does not change significantly the values of the adjusted constants. Thus, this rather large class of theories cannot explain this discrepant result.
The authors whishes to thank Professor D. Harari for many interesting discussions. H. V. acknowledges economic support from grant G035-UNLP.
## Appendix A Appendix I
Following (Bernstein, Brown and Feinberg, 1988) and eq. 22, the change in the helium abundance is given by:
$$\frac{\mathrm{\Delta }Y}{Y}=\frac{t_c}{\tau }\left(\frac{\mathrm{\Delta }\tau }{\tau }\frac{\mathrm{\Delta }t_c}{t_c}\right)+\frac{\mathrm{\Delta }X\left(t_F\right)}{X\left(t_F\right)}$$
(A1)
where
$$\frac{\mathrm{\Delta }X\left(t_F\right)}{X\left(t_F\right)}=0.52\frac{\mathrm{\Delta }b}{b}$$
(A2)
and
$$b=255\left(\frac{45}{4\pi N}\right)^{1/2}\frac{M_{pl}}{\tau Q^2}$$
(A3)
$$Q=\mathrm{\Delta }m=m_nm_p$$
(A4)
where $`N`$ is the number of neutrino types.
Since $`\tau =Q^5G_F^2`$, we find for the ratio of neutron to total baryon number at the freeze-out time:
$$\frac{\mathrm{\Delta }X\left(t_F\right)}{X\left(t_F\right)}=0.52\left[\frac{\mathrm{\Delta }M_{pl}}{M_{pl}}2\frac{\mathrm{\Delta }G_F}{G_F}7\frac{\mathrm{\Delta }Q}{Q}\right]$$
(A5)
Next, also from (Bernstein, Brown and Feinberg, 1988) we take the following expression for the neutron time capture:
$$t_c=\left(\frac{45}{16\pi N}\right)^{1/2}\left(\frac{11}{4}\right)^{2/3}\frac{M_{pl}}{T_{\gamma ,c}^2}+t_0$$
(A6)
where $`t_0`$ is an integration constant, $`T_{\gamma ,c}`$ is the temperature of the photon at the neutron capture time. Thus, the last equation yields:
$$\frac{\mathrm{\Delta }t_c}{t_c}=\frac{\mathrm{\Delta }M_{pl}}{M_{pl}}2\frac{\mathrm{\Delta }T_{\gamma ,c}}{T_{\gamma ,c}}$$
(A7)
Writing $`T_{\gamma ,c}=\frac{\epsilon _D}{z_c}`$ with $`\epsilon _D=m_n+m_pm_D`$ and $`z_c=\frac{\epsilon _D}{T_{\gamma ,c}}`$ we obtain:
$$\frac{\mathrm{\Delta }T_{\gamma ,c}}{T_{\gamma ,c}}=\frac{\mathrm{\Delta }\epsilon _D}{\epsilon _D}\frac{\mathrm{\Delta }z_c}{z_c}=\frac{\mathrm{\Delta }\mathrm{\Lambda }_{QCD}}{\mathrm{\Lambda }_{QCD}}\frac{\mathrm{\Delta }z_c}{z_c}$$
(A8)
Since at the neutron capture time, the neutrons are essentially all converted into helium, we may identify the temperature $`T_{\gamma ,c}`$ at which neutrons are captured, or equivalently the redshift $`z_c=\frac{\epsilon _D}{T_{\gamma ,c}}`$ , by the condition:
$$\left(\frac{dX_D}{dz}\right)_{z=z_c}=0$$
(A9)
where $`X_D`$ is the ratio of deuterons to total baryon number.
From (Bernstein, Brown and Feinberg, 1988) it is easy to see that the last equation is equivalent to the following:
$$f\left(z_c\right)=\mathrm{ln}\left(C_0\right)+\frac{4}{3}\mathrm{ln}\left(\frac{\epsilon _D}{m_p}\right)+\mathrm{ln}\left(\frac{M_{pl}}{m_p}\right)+\frac{4}{3}\mathrm{ln}\left(\alpha \right)\frac{17}{6}\mathrm{ln}\left(z_c\right)+z_c5.11\frac{\alpha ^{\frac{1}{2}}z^{\frac{1}{3}}}{\left(\frac{\epsilon _D}{m_p}\right)^{\frac{1}{3}}}=0$$
(A10)
where $`C_0`$ is a constant and $`z_c=26`$.
Assuming:
$$\delta f=\left(\frac{f}{z}\right)_{z=z_c}^{\alpha _i=\alpha _{io}}\delta z+\left(\frac{f}{\alpha }\right)_{z=z_c}^{\alpha _i=\alpha _{io}}\delta \alpha +\left(\frac{f}{M_{pl}}\right)_{z=z_c}^{\alpha _i=\alpha _{io}}\delta M_{pl}+\left(\frac{f}{\epsilon _D}\right)_{z=z_c}^{\alpha _i=\alpha _{io}}\delta \epsilon _D=0$$
(A11)
where $`\alpha _i=\alpha _{io}`$ means $`\alpha =\alpha _{today}`$ and $`\mathrm{\Lambda }_{QCD}=\mathrm{\Lambda }_{QCDtoday}`$ we obtain the following expression:
$$\frac{\mathrm{\Delta }z_c}{z_c}=\left[\left(\frac{f}{\alpha }\frac{\alpha }{z}\right)_{z=z_c}^{\alpha _i=\alpha _{io}}\frac{\mathrm{\Delta }\alpha }{\alpha }+\left(\frac{f}{M_{pl}}\frac{M_{pl}}{z}\right)_{z=z_c}^{\alpha _i=\alpha _{io}}\frac{\mathrm{\Delta }M_{pl}}{M_{pl}}+\left(\frac{f}{\epsilon _D}\frac{\epsilon _D}{z}\right)_{z=z_c}^{\alpha _i=\alpha _{io}}\frac{\mathrm{\Delta }\epsilon _D}{\epsilon _D}\right]\left(\frac{f}{z}\right)^1$$
(A12)
Evaluating eq.(A12) yields:
$$\frac{\mathrm{\Delta }z_c}{z_c}=0.13\frac{\mathrm{\Delta }\alpha }{\alpha }+0.046\frac{\mathrm{\Delta }M_{pl}}{M_{pl}}+0.26\frac{\mathrm{\Delta }\mathrm{\Lambda }_{QCD}}{\mathrm{\Lambda }_{QCD}}$$
(A13)
Thus, from eqs. (A1), (A5), (A7), (A8), (A13) and as $`\frac{\mathrm{\Delta }Q}{Q}=\frac{\mathrm{\Delta }\alpha }{\alpha }`$, the final expression yields:
$$\frac{\mathrm{\Delta }Y}{Y}=0.74\frac{\mathrm{\Delta }R_{KK}}{R_{KK}}+0.64\frac{\mathrm{\Delta }G_F}{G_F}+1.76\frac{\mathrm{\Delta }\alpha }{\alpha }0.3\frac{\mathrm{\Delta }\mathrm{\Lambda }_{QCD}}{\mathrm{\Lambda }_{QCD}}$$
(A14)
where we have used the equality $`R_{KK}\left(t_{pl}\right)R_{pl}=\frac{1}{M_{pl}}`$
|
warning/0005/nlin0005047.html
|
ar5iv
|
text
|
# Wave Solutions of Evolution Equations and Hamiltonian Flows on Nonlinear Subvarieties of Generalized Jacobians 11footnote 1PACS numbers 03.40.Gc, 11.10.Ef, 68.10.-m, AMS Subject Classification 58F07, 70H99, 76B15
## 1 Introduction.
The quasi-periodic solutions of most classical integrable PDEs can be obtained using the inverse scattering transform (IST) (see, for example, Ablowitz and Segur , Newell and Ablowitz and Clarkson ). This is done by establishing a connection with an isospectral eigenvalue problem for an associated Schrödinger operator.
The solution of nonlinear evolution equations using algebraic-geometric techniques was initially developed to handle $`N`$-phase wave trains. This approach can be summarized as follows. By using the trace formula, families of quasi-periodic and soliton solutions are associated with Hamiltonian flows on finite dimensional phase spaces. These flows are described by using so called $`\mu `$-variable representations leading to an Abel–Jacobi mapping which include holomorphic and, in some cases, meromorphic differentials (see amongst others, Ablowitz and Ma , Dubrovin , Ercolani and Alber and Alber ). Then the mapping is inverted in terms of Riemann theta-functions and their singular limits. Many well-known nonlinear equations such as KdV, sine-Gordon, focusing and defocusing nonlinear Schrődinger equations, which describe a wide variety of important phenomena in physics, optics, biology and engineering, were studied by using this approach.
Recently special attention was given to the shallow water (SW) equation derived in Cammasa and Holm in the context of the Hamiltonian structure,
$$U_t+3UU_x=U_{xxt}+2U_xU_{xx}+UU_{xxx}2\kappa U_x,$$
(1.1)
and the Dym type equation (see Cewen , Hunter and Zheng and Alber et al. )
$$U_{xxt}+2U_xU_{xx}+UU_{xxx}2\kappa U_x=0,\kappa =\mathrm{const}.$$
(1.2)
Camassa and Holm described classes of $`n`$-peakon soliton-type solutions for an integrable (SW) equation (1.1). In particular, they obtained a system of completely integrable Hamiltonian equations for the locations of the “peaks” of the solution, the points at which its spatial derivative changes sign. In other words, each peakon solution can be associated with a mechanical system of moving particles. Calogero and Calogero and Francoise further extended the class of mechanical systems of this type.
The problem of describing complex traveling wave and quasi-periodic solutions of the equations (1.1) and (1.2) can be reduced to solving finite-dimensional Hamiltonian systems on symmetric products of hyperelliptic curves. Namely, according to Alber et al , such solutions can be represented in the following form
$$U(x,t)=\mu _1+\mathrm{}+\mu _gM,$$
where $`g`$ is a positive integer, $`M`$ is a constant and the evolution of the variables $`\mu `$ is given by the equations
$$\underset{i=1}{\overset{g}{}}\frac{\mu _i^k\mathrm{d}\mu _i}{2\sqrt{R(\mu _i)}}=\{\begin{array}{ccc}0\hfill & k=1,\mathrm{},g2,\hfill & \\ \mathrm{d}t\hfill & k=g1,\hfill & \\ \mathrm{d}x\hfill & k=g.\hfill & \end{array}$$
(1.3)
Here $`R(\mu )`$ is a polynomial of degree $`2g+2`$ (for shallow water equation (1.1)) or $`2g+1`$ (for the Dym type equation (1.2)). Also $`M=0`$ for the Dym type equation.
In contrast to the finite-dimensional reductions of such equations as KdV and sine-Gordon equations, system (1.3) contains a meromorphic differential, also the number of holomorphic differentials is less than the genus $`g`$ of the corresponding hyperelliptic curve: $`W^2=R(\mu )`$. This implies that the problem of inversion (1.3) can not be solved in terms of meromorphic functions of $`x`$ and $`t`$. Examples of such equations arise in several problems of mechanics. These were considered in Vanhaecke and Abenda and Fedorov , where a connection was established with the flows on nonlinear subvarieties of hyperelliptic Jacobian varieties, so-called $`\mathrm{𝑠𝑡𝑟𝑎𝑡𝑎}`$. In Alber et al. a whole class of $`N`$-component systems with poles was shown to be integrable by reducing them to similar nonstandard inversion problems which contained meromorphic differentials. Therefore $`N`$-component systems can be overdetermined, implying that the genus of the spectral curve can be higher then the number of $`\mu `$-variables.
$`N`$-component systems can be briefly described as follows. For the KdV equation, the spectral parameter appears linearly in the potential of the corresponding Schrödinger equation: $`V=u\lambda `$, in the context of the IST method. In contrast, Antonowicz and Fordy \[1987a,b, 1988, 1989\] and Antonowicz, Fordy and Liu investigated potentials with poles in the spectral parameter for what they refer to as energy dependent Schrödinger operators connected to certain systems of evolution equations. Specifically, they obtained multi-Hamiltonian structures for $`N`$-component integrable systems of equations related to the following isospectral eigenvalue problem:
$$L\psi =\left(\frac{^2}{x^2}+\frac{V}{K}\right)\psi =0,$$
(1.4)
$$K=\underset{j=0}{\overset{M}{}}k_j\lambda ^j;V=\underset{j=0}{\overset{N}{}}v_j(x,t)\lambda ^j,$$
(1.5)
where the $`k_j`$ are constants and the $`v_j(x,t)`$ are functions of the variable $`x`$, the parameter $`t`$ and the spectral parameter $`\lambda `$ is complex. This includes the coupled KdV and Dym systems.
In Alber et al \[1994, 1995, 2000a\], the presence of a pole in the potential was shown to be necessary for the existence of weak billiard solutions of nonlinear equations. Billiard solutions of nonlinear PDE’s have been related to finite-dimensional integrable dynamical systems with reflections including ellipsoidal Birkhoff billiards. It turned out that the existence of billiard solutions and the presence of monodromy effects is a specific feature of the whole class of $`N`$-component systems with poles (see Alber et al. ).
Quasi-periodic solutions of the Dym equation were studied in Dmitrieva \[1993a\] and Novikov by using a connection with KdV equation and introducing additional phase functions. Soliton solutions of the Dym type equation were studied in Dmitrieva \[1993b\]. Periodic solutions of the shallow water equation were discussed in McKean and Constantin . Beals et al. used Stieltjes theorem on continued fractions and the classical momemt problem for studing multi-peakon solutions of the (SW) equation.
The main goal of this paper is to describe explicit formulae in terms of theta-functions and their singular limits for the solutions to the shallow water equation (1.1). We also explain the role of the mysterious phase functions used by Dmitrieva \[1993a\] when studying SW equation and equation of the Dym type. This phase is present for the whole class of WKI hierarchy. (see Wadati et al. ).
In the present paper the traveling wave, soliton, peakon, cuspon and quasi-periodic solutions are considered. Usually in the case of integrable evolution equations quasi-periodic flows are liberalized on the Jacobi varieties. In this paper we show that in the case of N-component systems with poles the $`x`$\- and $`t`$-flows take place on nonlinear subvarieties (strata) of generalized (noncompact) Jacobians. For this reason, the term “liberalization” is no longer applicable here. This makes the above nonlinear equations quite different from such well known equations as KdV, sine-Gordon and nonlinear Schrődinger equations. For the sake of clarity, in this paper we start with solutions related to (hyper)elliptic curves of at most genus 2. The case of arbitrary genus is only notationally more complicated and we provide complete formulae. In addition, we give a complete classification of real bounded solutions $`U(x,t)`$ in the above cases and provide corresponding plots.
Notice that the complex geometry of the traveling wave solutions, cusp and peakon solutions was previously studied in Alber et al in connection with geodesic flows with reflections on Riemannian manifolds and in Li and Olver from the point of view of singularity analysis.
### The Contents of the Paper.
In Section 2 we demonstrate the main difference between the nonlinear SW equation and the KdV equation from the point of view of the algebraic-geometric approach by obtaining traveling wave solutions as a result of inverting elliptic integrals of the second and third kind. Here we express the amplitude $`U`$ and the phase $`X`$ as meromorphic functions of an auxiliary variable parameterizing the elliptic curve.
In Section 3 we apply different singular limits to the problem of inversion resulting in formulae for different periodic and solitary solutions. In particular, peakon solutions are obtained as limits of the traveling wave solutions and are related to various singularizations of the elliptic curves.
Section 4 provides explicit expressions in terms of theta-functions for the stationary quasi-periodic solutions. This is done by using new complex parametrizations on the related associated hyperelliptic curve of genus 2.
In Section 5 we find time-dependent solutions by integrating and inverting equations (1.3) in the genus 2 case. We show that these equations can be extended to a standard Abel–Jacobi mapping of a symmetric product of the hyperelliptic curve to its generalized Jacobian. The original system (1.3) then defines a mapping onto a 2-dimensional nonlinear stratum of the Jacobian, a generalized theta-divisor, where the dynamics actually takes place. By fixing $`t`$ in the expression for the solution in terms of theta-functions, we then recover the stationary quasi-periodic solutions obtained in Section 4.
Section 6 contains qualitative analysis of real bounded solutions for the case when the Weierstrass points of the spectral curve are real.
In a forthcoming paper we will consider different singular limits of the quasi-periodic solutions when the spectral curve becomes singular and its arithmetic genus drops to zero. The solutions are then expressed in terms of purely exponential tau-functions and, in the real bounded case, they describe, in particular, a quasi-periodic train of peakons tending to a periodic one at infinity.
## 2 Traveling Wave Solutions.
The main difference between the nonlinear SW or Dym equations and the KdV equation from the point of view of the algebraic-geometric approach can be demonstrated already on the level of traveling wave solutions.
The traveling wave solutions of the KdV equation are obtained by inversion of an elliptic integral of the first kind, i.e., the integral of a holomorphic differential, which results in a meromorphic function.
In contrast to this, after substituting $`U(x,t)=\lambda (xct)`$ into the Dym type equation (1.2) and using a simple transformation (see Alber et al ), we arrive at the problem of inversion of the following Abelian integral of the second kind
$$_{U_0}^U\sqrt{\frac{\lambda c}{\kappa (\lambda a_1)(\lambda a_2)}}𝑑\lambda =_{U_0}^U\frac{\lambda c}{\sqrt{\kappa }\sqrt{R_3(\lambda )}}𝑑\lambda =xct=X,$$
(2.1)
defined on the elliptic curve $`=\{W^2=R_3(\lambda )\}`$, where
$$R_3(\lambda )=(\lambda a_1)(\lambda a_2)(\lambda c)$$
and $`U_0`$ being a constant. Here we suppose that the roots of $`R_3(\lambda )`$ are distinct. The differential in (2.1) has a double pole at infinity $`\lambda =\mathrm{}`$ and a double zero at the Weierstrass point $`\lambda =c`$ on $``$. It follows that the complex inverse function $`U(X)`$ must have two independent periods on $``$, the periods of the differential along two nontrivial homology cycles on $``$. On the other hand, $`U(X)`$ blows up only at $`X=\mathrm{}`$. There are no meromorphic functions with such properties (see e.g. Markushevich ). Moreover, because of the double zero of the differential, the solution $`U(X)`$ has moving branch points of the form
$$Uc=O((XX_0)^{2/3}).$$
(2.2)
One can show that $`U(X)`$ has infinitely many such branch points.
To deal with this we describe the complex function $`U(X)`$ in a new parametric form by introducing a new complex variable $`u`$ which gives a parameterization of the curve $``$:
$$_\lambda ^{\mathrm{}}\frac{d\lambda }{2\sqrt{R_3(\lambda )}}=u.$$
(2.3)
Then, according to the theory of elliptic functions,
$$\lambda (u)=\mathrm{}(u)+\mathrm{\Delta }=\frac{d^2}{du^2}\mathrm{log}\theta _{11}(z)+\mathrm{const},z=2\pi iu/\omega _3$$
(2.4)
where $`\mathrm{}(u)`$ is the elliptic Weierstrass function with periods $`2\omega _1`$ and $`2\omega _3`$ depending on the coefficients of $`R_3(\lambda )`$ and $`\mathrm{\Delta }=(a_1+a_2+c)/3`$. The $`\theta _{11}(z)`$ is the quasi-periodic Riemann theta-function which vanishes at the points of the period lattice
$`\mathrm{\Lambda }=\{z=2\pi i+𝐁\}`$, $`𝐁=\pi i\omega _1/\omega _3`$ (see, e.g. Dubrovin ):
$$\theta _{11}(z)=\underset{M}{}\mathrm{exp}\left(\frac{1}{2}𝐁(M+1/2)^2+(M+1/2)(z+\pi i)\right).$$
Here we use two variables $`u,z`$ because of different normalizations of (quasi)-periods of $`\mathrm{}`$ and $`\theta `$. The curve $``$ can be identified with the factor $`/\mathrm{\Lambda }`$. Now the integral (2.1) can be transformed as follows
$$_{u_0}^u(\mathrm{}(u)+\mathrm{\Delta }c)𝑑u=\sqrt{\kappa }X+\mathrm{const},u_0=\mathrm{const},$$
(2.5)
which yields
$$\sqrt{\kappa }X+\mathrm{const}=\zeta (u)+(\mathrm{\Delta }c)u=\frac{d}{dz}\mathrm{log}\theta _{11}(z)+\mathrm{\Delta }^{}z,\mathrm{\Delta }^{}=\mathrm{const},$$
(2.6)
where $`\zeta (u)`$ is the Riemann zeta-function with parameters $`\eta _1`$ and $`\eta _3`$ such that for any $`u`$
$$\zeta (u)=_u^{\mathrm{}}\mathrm{}(u)𝑑u,\zeta (u+2\omega _1)=\zeta (u)+2\eta _1,\zeta (u+2\omega _3)=\zeta (u)+2\eta _3.$$
The constants $`2\eta _1,2\eta _3`$ are interpreted as the periods of the differential
$$\lambda d\lambda /2\sqrt{R_3(\lambda )}=\lambda (u)du.$$
Then, in view of (2.6),
$$\sqrt{\kappa }X(u+2\omega _1)=\sqrt{\kappa }X(u)+2\eta _1+2(\mathrm{\Delta }c)\omega _1,$$
$$\sqrt{\kappa }X(u+2\omega _3)=\sqrt{\kappa }X(u)+2\eta _3+2(\mathrm{\Delta }c)\omega _3.$$
Thus we have expressed the amplitude $`U=\lambda `$ and the phase $`X`$ in terms of the auxiliary complex variable $`u`$.
In what follows we study real solutions $`U(X)=\lambda (u(X))`$ which correspond to the case of all roots of the polynomial $`R_3(\lambda )`$ being real. We choose the half-period $`\omega _1`$ to be real and $`\omega _3`$ to be purely imaginary. The existence of branch points of $`U(X)`$ (see (2.2)) implies that the real solutions may have cusps. To demonstrate this we consider two different cases.
### Case 1: $`\kappa >0`$, $`a_1<a_2<c<\mathrm{}`$.
According to the theory of elliptic functions, parameterization (2.3) yields that
$$\lambda (\omega _1)=c,\lambda (\omega _3)=a_1,\lambda (\omega _1+\omega _3)=a_2,\lambda (0)=\mathrm{}.$$
(2.7)
Along the real axis $`\mathrm{}u`$ and the line $`\{u=\omega _3+u^{}|u^{}\}`$, $`\lambda (u)`$ is real valued. Since we are interested in nonsingular traveling wave solutions and $`\lambda (u)`$ has a pole at the origin, let us consider $`\lambda (\omega _3+u^{})`$ as a function of $`u^{}`$. It is periodic with period $`2\omega _3`$ and takes values in the interval $`[a_1,a_2]`$. The differentials in (2.1) and (2.5) do not have poles or zero’s on the line $`\{u=\omega _3+u^{}\}`$. Hence, $`X(\omega _3+u^{})`$ and the inverse $`u^{}(X)`$ are monotonic functions. As a result, due to (2.6), the composition function $`U(X)=\lambda (\omega _3+u^{}(X))`$ is a regular periodic function with period $`(2\eta _3+2(\mathrm{\Delta }c)\omega _3)/\sqrt{\kappa }`$, and its graph is merely a distortion of that of $`\lambda (\omega _3+u^{})`$ .
### Case 2: $`\kappa >0,a_1<c<a_2<\mathrm{}`$.
Here we have
$$\lambda (\omega _1)=a_2,\lambda (\omega _3)=a_1,\lambda (\omega _1+\omega _3)=c,\lambda (0)=\mathrm{}.$$
(2.8)
Again, $`\lambda (u)`$ is real valued along the real axis $`\mathrm{}u`$ and the line $`\{u=\omega _3+u^{}|u^{}\}`$. Since $`\lambda (u)`$ has a pole along the real axis, we again consider only the real function $`\lambda (\omega _1+u^{})`$, $`u^{}`$ which is $`2\omega _3`$-periodic and now takes values in the interval $`[c,a_2]`$. In this case, the differential (2.5) has a double zero at $`u=\omega _1+\omega _3`$ ($`u^{}=\omega _1`$), where the derivative $`d\lambda /du`$ has a simple zero. As a result, the derivative $`dU/dX=d\lambda /dudu/dX`$ blows up for the corresponding value of $`X`$, which implies that the graph of the function $`U(X)`$ has cusps with periodicity $`(2\eta _3+2(\mathrm{\Delta }c)\omega _3)/\sqrt{\kappa }`$ . This phenomenon was first detected in Alber et al .
### SW Equation.
Traveling wave solutions for the shallow water equation have a similar description. Substituting $`U(x,t)=\mu (xct)`$ into (1.1), we obtain an Abelian integral of the third kind
$$_{U_0}^U\sqrt{\frac{\mu c}{\kappa (\mu a_1)(\mu a_2)(\mu a_3)}}𝑑\mu =_{U_0}^U\frac{\mu c}{\sqrt{\kappa }\sqrt{R_4(\mu )}}𝑑\mu =xct=X,$$
(2.9)
defined on the even order elliptic curve
$$\stackrel{~}{}=\{W^2=R_4(\mu )\},R_4(\mu )=(\mu a_1)(\mu a_2)(\mu a_3)(\mu c),$$
where the roots of $`R_4`$ are supposed to be distinct. The differential in (2.9) has a pair of simple poles at the infinite points $`\mathrm{}_{},\mathrm{}_+`$ on $`\stackrel{~}{}`$. Then the complex inverse function $`U(X)`$ must have three independent periods on $``$: the periods of the differential along two nontrivial homology cycles on $`\stackrel{~}{}`$ and along a homology zero cycle around one of the infinite points. A meromorphic function with such a property does not exist.
Let $`A,B`$ be canonically conjugated cycles on $`\stackrel{~}{}`$ and $`\overline{\omega }`$ be the normalized holomorphic differential
$$\overline{\omega }=\frac{d\mu }{\mathrm{\Pi }\sqrt{R_4(\mu )}},$$
where the multiplier $`\mathrm{\Pi }`$ is chosen from the condition $`_A\overline{\omega }=2\pi i`$. Introduce a new variable $`z`$ parameterizing $`\stackrel{~}{}`$ as follows
$$_\mu ^c\frac{d\mu }{\mathrm{\Pi }\sqrt{R_4(\mu )}}=z,$$
(2.10)
where $`a_1,a_2`$ and $`c`$ denote the corresponding Weierstrass (branch) points on $`\stackrel{~}{}`$. Then we get the expression
$$\mu (z)=\rho \frac{d}{dz}\mathrm{log}\frac{\theta [\delta ](zq/2)}{\theta [\delta ](z+q/2)},$$
(2.11)
$$\rho =_A\mu \overline{\omega },q=2_c^\mathrm{}_+\frac{d\mu }{\mathrm{\Pi }\sqrt{R_4(\mu )}}.$$
Thus $`\mu (z)`$ is an elliptic function with periods $`2\pi i`$, $`𝐁=_B\overline{\omega }`$. The integral (2.9) takes the form
$$\sqrt{\kappa }X=_{z_0}^z(\mu (z)c)𝑑z=(\rho c)z\mathrm{log}\frac{\theta _{11}(zq/2)}{\theta _{11}(z+q/2)}+\mathrm{const}.$$
(2.12)
As a result, we have expressed the function $`U=\mu `$ and its argument $`X`$ in terms of $`z`$.
Since the differential in (2.9) has a double zero at the Weierstrass point $`\mu =c`$, similar to the case of the Dym equation, the inversion of (2.9) yields $`U(X)`$ with infinitely many branch points of the form (2.2). This results in real solutions having cusps.
Now we describe 2 different types of real traveling wave solutions of (1.1) assuming that $`\kappa >0`$ and all roots of $`R_4(\mu )`$ are real. In this case the period $`𝐁`$ is real as well.
### Case 1.
$`a_1<a_2<a_3<c`$. According to the parameterization (2.10), we have
$$q,\mathrm{\Pi }=_{a_1}^{a_2}\frac{d\mu }{\sqrt{R_4(\mu )}},𝐁=2_{a_3}^c\frac{d\mu }{\mathrm{\Pi }\sqrt{R_4(\mu )}},\mu (q/2)=U(q/2)=\mathrm{},$$
$$\mu (0)=c,\mu (𝐁/2)=a_1,\mu (\pi i+𝐁/2)=a_2,\mu (\pi i)=a_3.$$
The function $`\mu (z)`$ is real along the real axis $`\mathrm{}u`$ and the line $`\mathrm{}=\{z=\pi i+z^{}|z^{}\}`$, but it is finite only in the second case. The function $`\mu (\pi i+z^{})`$ changes periodically between $`a_2`$ and $`a_3`$ with period $`𝐁`$. Along the line $`\mathrm{}`$ the differential in (2.9) has no zeros, therefore $`X(\pi i+z^{})`$ and the inverse real function $`z^{}(X)`$ are strictly monotonic functions. According to (2.12), the composition function $`U(X)=\mu (\pi i+z^{}(X))`$ is a regular periodic function with real period
$$\frac{1}{\sqrt{\kappa }}\left[(\rho c)𝐁\mathrm{log}\frac{\theta _{11}(𝐁q/2)}{\theta _{11}(𝐁+q/2)}\right].$$
Such a function is shown in figure 1a.
### Case 2.
$`a_1<a_2<c<a_3`$. Now $`q,\mathrm{\Pi },𝐁`$ have the same expressions as above and
$$\mu (0)=c,\mu (𝐁/2)=a_2,\mu (\pi i+𝐁/2)=a_1,\mu (\pi i)=a_3.$$
The function $`\mu (z)`$ is real and finite only along the real axis, on which the differential has a double zero. As a result, the composition $`U(X)=\mu (z(X))`$ varies between $`a_2`$ and $`c`$, with the same period as above and has cusps for $`U=c`$. See figure 1b.
The cases of other positions of $`c`$ among the real roots $`(a_1,a_2,a_3)`$ lead to either one of the above two types of solutions or to unbounded solutions.
## 3 Periodic and solitary peakon solutions.
In this section we describe different periodic and solitary solutions to the Dym type and shallow water SW equations obtained as limits of the traveling wave solutions and associated with various singularizations of the elliptic curves described in the previous section. In particular, we encounter periodic and solitary peakon solutions, i.e. solutions with discontinuous derivatives.
In contrast to traveling wave solutions given above globally in a parametric form, peakon solutions can be given explicitly only in certain intervals.
### Periodic peakon solution for the Dym type equation.
Consider again the integral (2.1). Assuming $`\kappa =1`$, in the limit $`a_2c`$ the latter can be written in the equivalent differential form
$$\frac{d\lambda }{\sqrt{a_1\lambda }}=dX,\mathrm{or}\frac{d\lambda }{(\lambda c)\sqrt{a_1\lambda }}=dX^{},dX=(\lambda c)dX^{},$$
(3.1)
$`X^{}`$ being a new variable. Now putting $`\lambda =a_1\nu ^2`$, $`a_1c=\alpha ^2`$ and integrating the second differential in (3.1), we find that
$$\frac{1}{\alpha }\mathrm{log}\left(\frac{\nu \alpha }{\nu +\alpha }\right)=X^{}+C^{},C^{}=\mathrm{const},\mathrm{i}.\mathrm{e}.,\nu =\alpha \frac{1+e^{\alpha X^{}}}{1e^{\alpha X^{}}}.$$
Let us choose here $`C^{}=\pi i/\alpha `$. Then we have
$$\lambda (X^{})c=\alpha _1^2\alpha _1^2\left(\frac{e^{\alpha X^{}/2}e^{\alpha X^{}/2}}{e^{\alpha X^{}/2}+e^{\alpha X^{}/2}}\right)^2,\mathrm{or},\mathrm{equivalently},$$
$$\lambda (X^{})c=4_X^{}^2\mathrm{log}\tau (X^{}),\tau (X^{})=e^{\alpha X^{}/2}+e^{\alpha X^{}/2}.$$
(3.2)
This function is $`2\pi i`$ periodic in $`\alpha X`$ and takes real finite values along the real axis.
Next, using the expression (3.2), we integrate the relation $`dX=(\lambda (X^{})c)dX^{}`$ and get the connection between $`X`$ and $`X^{}`$:
$$X=4_X^{}\mathrm{log}\tau (X^{})+X_0=2\alpha \frac{e^{\alpha X^{}/2}e^{\alpha X^{}/2}}{e^{\alpha X^{}/2}+e^{\alpha X^{}/2}}+X_0,X_0=\mathrm{const}.$$
(3.3)
Notice that as $`X^{}\mathrm{}`$ or $`+\mathrm{}`$, $`X`$ has finite limits which differ by $`4\alpha `$. Thus, expressions (3.2) and (3.3) give the solution $`U(X)=\lambda (X^{}(X))`$ in a parametric form only in the interval $`[X_02\alpha ,X_0+2\alpha ]`$. As follows from (3.2) and (3.1), at the endpoints of the interval, $`U=c`$ and $`\mathrm{d}U/\mathrm{d}X=\pm \sqrt{a_1c}`$.
Now we define the global solution $`U(X)`$ for all values of $`X`$ by using periodic extension, i.e., by gluing an infinite number of pieces corresponding to $`X_0=4\alpha N`$, $`N`$ in (3.3). At the endpoint of each interval the derivative of the solution changes sign resulting in a peak, and we obtain a periodic peakon solution.
On the other hand, a direct integration of the first differential in (3.1) yields the following solution which holds between subsequent peaks
$$\lambda (X)=a_1\frac{1}{4}(XX_0)^2,X[X_02\alpha ,X_0+2\alpha ].$$
Thus, in contrast to $`\lambda (X^{})`$ and $`X(X^{})`$, the profile of $`U(X)`$ between peaks is not exponential, but a quadratic one.
### Periodic peakon solution for the SW equation.
In a similar way, consider a limit of the periodic solution of the SW equation by putting $`a_3=c`$ in (2.9). Then the curve $`\stackrel{~}{}`$ becomes singular having a double point at $`\mu =c`$. Let $`\stackrel{~}{}^{}`$ be the corresponding regularized curve. The integral (2.9) gives rise to the following differentials
$$\frac{d\mu }{\sqrt{(\mu a_1)(\mu a_2)}}=dX,\frac{d\mu }{(\mu c)\sqrt{(\mu a_1)(\mu a_2)}}=dX^{},$$
(3.4)
where, as before, $`X=xct`$, $`dX=(\mu c)dX^{}`$. Integrating and inverting (3.4) we obtain respectively
$$\mu (X)=\frac{a_1a_2}{4}(e^X+e^X)+\frac{a_1+a_2}{2},$$
(3.5)
$$\begin{array}{c}\mu (X^{})c=\frac{4}{(e^{Z/2}+e^{Z/2})^2/(a_1c)+(e^{Z/2}e^{Z/2})^2/(a_2c)},\hfill \\ Z=\sqrt{(ca_1)(ca_2)}X^{}.\hfill \end{array}$$
(3.6)
The last expression is $`2\pi i`$-periodic in $`Z`$ and provides a parameterization of the regularized curve $`\stackrel{~}{}^{}`$. Integrating (3.6) with respect to $`X^{}`$, we find that
$$\begin{array}{c}X=\mathrm{log}\frac{(a_1a_2)e^Z(a_1+a_2)+2c2\sqrt{(ca_1)(ca_2)}}{(a_1a_2)e^Z(a_1+a_2)+2c+2\sqrt{(ca_1)(ca_2)}}+\mathrm{const}.\hfill \end{array}$$
(3.7)
Now suppose that $`a_1,a_2,c`$ are real and $`c<a_1<a_2`$ or $`a_1<a_2<c`$. As follows from (3.6), (3.7), for real parameter $`X^{}`$, the variables $`\mu `$ and $`X`$ are real as well. As $`X^{}\pm \mathrm{}`$, the function $`X(X^{})`$ varies between different finite limits which we denote by $`X_1`$ and $`X_2`$, whereas, by (3.6), $`\mu (X^{})c`$ tends to zero. As a result, the composition function $`U(X)=\mu (X^{}(X))`$ is defined only in the interval $`[X_1,X_2]`$, where it takes values in $`[c,a_1]`$ or $`[a_2,c]`$. According to (3.4), at the endpoints of the interval, $`U=c`$, $`dU/dX=\pm \sqrt{(ca_1)(ca_2)}`$. Like for the Dym equation, we define the global solution $`U(X)`$ by periodic extension of $`\mu (X^{}(X))`$. This implies that $`U(X)`$ has periodic peaks on the infinite interval $`(\mathrm{},\mathrm{})`$.
The expression in (3.5) provides a piece of the solution between two subsequent peaks. Solving the equation $`\mu (X)=c`$ or using (3.7) directly, we find the period of the peakon solution to be
$$X_2X_1=\mathrm{log}\frac{2c(a_1+a_2)+2\sqrt{(ca_1)(ca_2)}}{2c(a_1+a_2)2\sqrt{(ca_1)(ca_2)}}.$$
(3.8)
See figure 1c. This completes the description of the periodic peakon solution.
One can show that the other possible case: $`a_1<c<a_2`$ does not provide a real bounded solution. For details about algebraic geometric approach to describing n-peakon solutions see Alber and Miller and Alber et al. \[2000a\].
### Solitons for the SW equation.
Consider another possible degeneration of the integral (2.9), assuming that $`a_2=a_3=a`$ and $`a,a_1,c`$ are distinct. Then we have
$$\begin{array}{c}\frac{(\mu c)d\mu }{(\mu a)\sqrt{(\mu a_1)(\mu c)}}=dX,\frac{d\mu }{(\mu a)\sqrt{(\mu a_1)(\mu c)}}=dX^{},\hfill \\ \\ X=xct+\mathrm{const},dX=(\mu c)dX^{}.\hfill \end{array}$$
(3.9)
The first differential has 2 pairs of simple poles on the rational curve
$`\{\nu ^2=(\mu a_1)(\mu c)\}`$ corresponding to $`\mu =a`$ and $`\mu =\mathrm{}`$ and a double zero for $`\mu =c`$. Therefore the inverse function $`U(X)`$ again has branching of the form (2.2).
Integrating the second differential in (3.9) and inverting, we obtain
$$\mu (X^{})a=\frac{4}{(e^{\stackrel{~}{Z}/2}+e^{\stackrel{~}{Z}/2})^2/(ca)+(e^{\stackrel{~}{Z}/2}e^{\stackrel{~}{Z}/2})^2/(a_1a)},$$
(3.10)
$$\stackrel{~}{Z}=\sqrt{(aa_1)(ac)}X^{}.$$
Then, integrating the relation between $`X`$ and $`X^{}`$ yields
$$\begin{array}{c}X=(ac)X^{}+\mathrm{log}\frac{(a_1c)e^{\stackrel{~}{Z}}(a_1+c)+2a2\sqrt{(aa_1)(ac)}}{(a_1c)e^{\stackrel{~}{Z}}(a_1+c)+2a+2\sqrt{(aa_1)(ac)}}+\mathrm{const}.\hfill \end{array}$$
(3.11)
As before, let us consider possible real solutions assuming that $`a,a_1,c`$ are real. According to (3.10), $`\mu (\stackrel{~}{Z})`$ is $`2\pi i`$-periodic and it takes real values along the lines Im$`\stackrel{~}{Z}=N\pi i`$, $`N`$.
### Case 1.
$`a<a_1<c`$. The function $`\mu (\stackrel{~}{Z})`$ is real and finite only along the lines Im$`\stackrel{~}{Z}=\pi i+2N\pi i`$, where it describes a smooth solitary wave tending to $`a`$ as $`\mathrm{}X`$ or $`\mathrm{}\stackrel{~}{Z}`$ tends to $`\pm \mathrm{}`$ and having the maximum $`a_1`$. Along these lines the derivative $`dX/dX^{}=\mu c`$ is always real, negative and separated from zero. Therefore, the composition function $`U(X)=\mu (X^{}(X))`$ gives a smooth solitary wave as well.
### Case 2.
$`a<c<a_1`$. Now $`\mu (\stackrel{~}{Z})`$ is real and finite only along the lines Im$`\stackrel{~}{Z}=2N\pi i`$, where it again describes a smooth solitary wave tending to $`a`$ as $`\mathrm{}X\pm \mathrm{}`$ and having the maximum $`c`$. On the other hand, along these lines the derivative $`dX/dX^{}=\mu c`$ is always real and negative except when $`\mu =c`$. At this point $`dX/dX^{}`$ has a double zero. As a result, for $`\mu =c`$ the derivative $`dU/dX`$ blows up and the composition function $`U(X)=\mu (X^{}(X))`$ describes a solitary cusp (cuspon).
### Solitary peakon for the SW equation.
As seen from (3.8), in the limit $`a_1=a_2=a`$ the period of the above peakon solution tends to infinity. This gives us a solitary peakon solution, which can also be regarded as the separatrix between the smooth soliton and cuspon solutions. Indeed, in this case the algebraic curve $`\stackrel{~}{}`$ has 2 double points with $`\mu =a`$ and $`\mu =c`$. Its reauthorization $`\widehat{}`$ consists of two disjoint copies of $`=\{\mu \}\mathrm{}`$. On them the differentials (3.4) take the following simple form
$$\frac{d\mu }{\mu a}=ldX,\frac{d\mu }{(\mu c)(\mu a)}=ldX^{},$$
(3.12)
where, as above, $`dX=(\mu c)dX^{}`$ and $`l=\pm 1`$, regarding to the copy of $``$. First, we suppose $`l=1`$. After integration in $`\mu `$ and inversion this yields
$$\mu (X)=a+e^{X+C_2},C_2=\mathrm{const},\mu (X^{})=\frac{ace^{(ac)X^{}}}{1e^{(ac)X^{}}}.$$
(3.13)
The function $`\mu (X^{})`$ has period $`2\pi i/(ac)`$ and its fundamental domain gives a parameterization of the cylinder $`\{\mu =a\}\{\mu =c\}`$.
Integrating the differential relation between $`X`$ and $`X^{}`$, we obtain
$$X(X^{})=_{X_0^{}}^X^{}\frac{(ac)e^{(ac)X^{}}}{1e^{(ac)X^{}}}𝑑X^{}=\mathrm{log}(e^{(ca)X^{}}1)+C_1,X_0^{},C_1=\mathrm{const}.$$
Suppose that $`a,c`$ are real and, for definiteness, put $`a<c`$. The function $`\mu (X^{})`$ is $`2\pi i/(ac)`$ periodic, it has a pole along the real axis, whereas along the line Im$`X^{}=\pi i/(ca)`$ it varies in the interval $`(a,c)`$. Let us put $`C_1=\pi i`$. Then $`X(X^{})`$ is real along this line. We notice that as $`\mathrm{}X^{}\mathrm{}`$, $`X`$ tends to $`\mathrm{}`$ and $`\mu `$ tends to $`a`$, whereas for $`\mathrm{}X^{}\mathrm{}`$, we have $`X0`$ and $`\mu c`$. Thus the composition function $`\mu (X^{}(X))`$ is defined on the interval $`[\mathrm{},0]`$ only, where it is also given by the first expression in (3.13) for $`C_2=\mathrm{log}(ca)`$.
Now assuming $`l=1`$, we obtain the same expressions with $`X`$ replaced by $`X`$. Let us choose the same values of the integration constants $`C_1,C_2`$. Then the composition function $`\mu (X^{}(X))`$ is defined on the interval $`[0,\mathrm{}]`$, where it has the form $`\mu (X)=a+(ca)e^X`$. As a result, the two copies of $``$ give rise to two branches of the real continuous solution
$$U(X)=a+(ca)e^{|X|},X,$$
which has a peak at the origin.
### Remark.
As seen from the above peakon solutions, the amplitude of peaks coincides with the velocity $`c`$, which is a specific property of soliton propagation.
## 4 Stationary quasi-periodic solutions.
Stationary solutions provide profiles of the quasi-periodic wave solutions. For the sake of clarity, in this paper we restrict ourselves to the simplest nontrivial case $`g=2`$. All the formulae and solutions below (except structure of real solutions) can be easily extended to the arbitrary $`g`$-dimensional case which is only notationally more complicated.
### Stationary quasi-periodic solutions for the SW equation.
According to the trace formula, in the genus $`2`$ case we have
$$U(x,t)=\mu _1+\mu _2\underset{j=1}{\overset{5}{}}a_j,$$
(4.1)
and equations (1.3) take the form
$$\begin{array}{c}\frac{\mu _1d\mu _1}{2\sqrt{R_6(\mu _1)}}+\frac{\mu _2d\mu _2}{2\sqrt{R_6(\mu _2)}}=dt,\hfill \\ \\ \frac{\mu _1^2d\mu _1}{2\sqrt{R_6(\mu _1)}}+\frac{\mu _2^2d\mu _2}{2\sqrt{R_6(\mu _2)}}=dx,\hfill \end{array}$$
(4.2)
where
$$R_6(\mu )=\kappa \mu (\mu a_1)\mathrm{}(\mu a_5),a_1,\mathrm{},a_5=\mathrm{const}.$$
Here we suppose that all the roots of $`R_6(\mu )`$ are distinct. The variables $`\mu _1,\mu _2`$ must be regarded as coordinates of points $`P_1=(\mu _1,w_1)`$, $`P_2=(\mu _2,w_2)`$ on the genus 2 hyperelliptic curve $`\mathrm{\Gamma }=\{w^2=R_6(\mu )\}`$. Equations (4.2) involve one holomorphic differential and one meromorphic differential of the third kind having a pair of simple poles at the infinite points $`\mathrm{}_{},\mathrm{}_+`$ on $`\mathrm{\Gamma }`$. Integrating (4.2), we obtain the mapping of the symmetric product $`\mathrm{\Gamma }^{(2)}`$ to $`^2=(t,x)`$
$$\begin{array}{c}_{P_0}^{P_1}\frac{\mu d\mu }{2\sqrt{R_6(\mu )}}+_{P_0}^{P_2}\frac{\mu d\mu }{2\sqrt{R_6(\mu )}}=t,\hfill \\ \\ _{P_0}^{P_1}\frac{\mu ^2d\mu }{2\sqrt{R_6(\mu )}}+_{P_0}^{P_2}\frac{\mu ^2d\mu }{2\sqrt{R_6(\mu )}}=x.\hfill \end{array}$$
(4.3)
where $`P_0`$ is a fixed basepoint of the mapping. Notice that for $`P_1`$ or $`P_2=\mathrm{}_{},\mathrm{}_+`$, we have $`x=\mathrm{}`$. Let us fix a canonical basis of cycles $`A_1,A_2,B_1,B_2`$ on $`\mathrm{\Gamma }`$ in a standard way (see, for example, Mumford ). The mapping has four independent (over the reals) 2-dimensional vectors of periods of the above differentials along the cycles. In addition, it has one extra period vector corresponding to a homology zero cycle around $`\mathrm{}_{}`$ or $`\mathrm{}_+`$. As a result, the mapping (4.3) has 5 period vectors in $`^2`$ hence its inversion is not well defined: there do not exist meromorphic functions on $`^2`$ with five periods. In particular, $`U(x,t)`$ is not a meromorphic or single valued complex function of $`t,x`$.
In order to describe properties of $`U(x,t)`$, we, first, fix time by putting $`t=t_0`$ ($`dt=0`$) and consider stationary solutions $`U(x,t_0)`$. Introduce a new coordinate $`x^{}`$ such that
$$dx=\mu _1\mu _2dx^{}.$$
(4.4)
Then equations (4.2) lead to the Abel–Jacobi mapping of $`\mathrm{\Gamma }^{(2)}`$ to the Jacobian variety Jac($`\mathrm{\Gamma }`$) of $`\mathrm{\Gamma }`$, which includes holomorphic differentials only:
$$\begin{array}{c}_{P_0}^{P_1}\frac{d\mu }{2\sqrt{R_6(\mu )}}+_{P_0}^{P_2}\frac{d\mu }{2\sqrt{R_6(\mu )}}=u_1,\hfill \\ \\ _{P_0}^{P_1}\frac{\mu d\mu }{2\sqrt{R_6(\mu )}}+_{P_0}^{P_2}\frac{\mu d\mu }{2\sqrt{R_6(\mu )}}=u_2,\hfill \\ \\ u_1=x^{}+\mathrm{const},u_2=\mathrm{const},\hfill \end{array}$$
(4.5)
where $`u_1,u_2`$ are coordinates on the universal covering $`^2`$ of Jac$`(\mathrm{\Gamma })`$.
Let $`\overline{\omega }_1,\overline{\omega }_2`$ be the dual basis of normalized holomorphic differentials on $`\mathrm{\Gamma }`$ with respect to the above choice of cycles and $`z_1,z_2`$ be the corresponding coordinates on the universal covering of Jac$`(\mathrm{\Gamma })`$:
$$\begin{array}{c}\overline{\omega }_1=\frac{d_{11}+d_{12}\mu }{2\sqrt{R_6(\mu )}}d\mu ,\overline{\omega }_2=\frac{d_{21}+d_{22}\mu }{2\sqrt{R_6(\mu )}}d\mu ,\hfill \\ \\ z_1=d_{11}u_1+d_{12}u_2,z_2=d_{21}u_1+d_{22}u_2.\hfill \end{array}$$
(4.6)
Here the normalizing constants $`d`$ are uniquely determined by the conditions $`_{A_i}\overline{\omega _j}=\delta _{ij}`$.
Recall that the standard theta-function related to a Riemann surface of genus $`g`$ and theta-functions with characteristics $`\alpha =(\alpha _1,\mathrm{},\alpha _g)`$, $`\beta =(\beta _1,\mathrm{},\beta _g)^g`$ have the form
$$\begin{array}{c}\theta (z|𝐁)=\underset{M^g}{}\mathrm{exp}\left(\frac{1}{2}(𝐁M,M)+(M,z)\right),\hfill \\ (M,z)=\underset{i=1}{\overset{g}{}}M_iz_i,(𝐁M,M)=\underset{i,j=1}{\overset{g}{}}𝐁_{ij}M_iM_j,\hfill \\ \\ \theta \left[\begin{array}{c}\alpha \\ \beta \end{array}\right](z|𝐁)=\mathrm{exp}\{(𝐁\alpha ,\alpha )/2+(z+2\pi i\beta ,\alpha )\}\theta (z+2\pi i\beta +𝐁\alpha |𝐁),\hfill \end{array}$$
(4.7)
$`𝐁`$ being the $`g\times g`$ period matrix of $`\mathrm{\Gamma }`$. In the sequel we shall omit it in the notation.
Now we choose the basepoint $`P_0`$ of the mapping (4.5) to be the last Weierstrass point $`(a_5,0)`$ on $`\mathrm{\Gamma }`$. Then, according to the trace formula for even order hyperelliptic curves (see e.g., Clebsch and Gordan , Dubrovin )
$$U=\mu _1+\mu _2\underset{j=1}{\overset{5}{}}a_j=\mathrm{const}_W\mathrm{log}\frac{\theta [\delta ](zq/2)}{\theta [\delta ](z+q/2)},$$
(4.8)
$$z=(z_1,z_2),q=(q_1,q_2)^T,q_i=_{\mathrm{}_{}}^\mathrm{}_+\overline{\omega }_i,$$
where, in view of the normalizing change (4.6), $`z_1=d_{11}x^{}+\mathrm{const}`$, $`z_2=d_{21}x^{}+\mathrm{const}`$, $`_W`$ is the derivative along a tangent vector $`W`$ of $`\mathrm{\Gamma }\mathrm{Jac}(\mathrm{\Gamma })`$ at $`\mathrm{}_+`$, namely, in the coordinates $`(u_1,u_2)`$, $`W=(0,1)^T`$, and in the coordinates $`(z_1,z_2)`$, $`W=(d_{12},d_{22})^T`$. Finally, $`\delta =(\delta ^{\prime \prime },\delta ^{})^T`$, $`\delta ^{\prime \prime },\delta ^{}\frac{1}{2}^g/^g`$ is the half-integer theta-characteristic corresponding to the vector of Riemann constants (see Mumford ). For the chosen standard canonical basis of cycles and the basepoint $`P_0=(a_5,0)`$
$$\delta ^{}=(1/2,\mathrm{},1/2)^T,\delta ^{\prime \prime }=(g/2,(g1)/2,\mathrm{},1,1/2)^T(\mathrm{mod}\mathrm{\hspace{0.33em}1}).$$
(4.9)
Thus, in our case
$$\delta =\left(\begin{array}{cc}0& 1/2\\ 1/2& 1/2\end{array}\right).$$
The function $`U(z_1,z_2)`$ is meromorphic on Jac$`(\mathrm{\Gamma })`$ and it has simple poles along 2 translates of the theta-divisor $`\mathrm{\Theta }=\{\theta (z)=0\}\mathrm{Jac}(\mathrm{\Gamma })`$:
$$\mathrm{\Theta }_{}=\{\theta [\delta ](zq/2)=0\},\mathrm{\Theta }_+=\{\theta [\delta ](z+q/2)=0\},$$
which are tangent to each other at the origin $`\{z=0\}`$. Thus, $`U(z_1(x^{}),z_2(x^{}))`$ is a quasi-periodic function of the complex variable $`x^{}`$. Notice that a quasi-periodic genus 2 solution of the nonlinear mKdV equation has the same form.
We also notice that the point $`E_0=(\mu =0,w=0)`$ is a Weierstrass (branch) point on $`\mathrm{\Gamma }`$. Then, following Clebsch and Gordan , we have the following expression for the symmetric polynomial
$$\mu _1\mu _2=\varrho \frac{\theta ^2[\delta +\eta _0](z)}{\theta [\delta ](z+q/2)\theta [\delta ](zq/2)},\varrho =\mathrm{const},$$
(4.10)
where $`\eta _0`$ is the half-integer theta-characteristic corresponding to the branch point $`E_0`$:
$$\eta _0=(\eta _0^{\prime \prime },\eta _0^{})^T,_{P_0}^{E_0}(\overline{\omega }_1,\overline{\omega }_2)^T=2\pi i\eta _0^{\prime \prime }+𝐁\eta _0^{}^2.$$
(4.11)
Thus, the product $`\mu _1\mu _2`$ is a meromorphic function on Jac$`(\mathrm{\Gamma })`$ having simple poles along $`\mathrm{\Theta }_{},\mathrm{\Theta }_+`$ and a double zero along another translate of the theta-divisor $`\mathrm{\Theta }`$, $`\mathrm{\Theta }_0=\{\theta [\delta +\eta _0](z)=0\}`$, passing through the origin and intersecting each of the translates $`\mathrm{\Theta }_{},\mathrm{\Theta }_+`$ at two points. The translate $`\mathrm{\Theta }_0`$ can also be interpreted as the image of the curve $`\mathrm{\Gamma }`$ itself under the Abel–Jacobi mapping (4.5):
$$\mathrm{\Theta }_0=\left\{_{P_0}^P(\overline{\omega }_1,\overline{\omega }_2)^T+_{P_0}^{E_0}(\overline{\omega }_1,\overline{\omega }_2)^T\right|P\mathrm{\Gamma }\}.$$
It follows from (4.4), (4.10) that generically the derivative of the function $`x(x^{})`$ is equal to $`\mu _1\mu _2`$ and that it has a double zero each time when the complex $`x^{}`$-flow intersects $`\mathrm{\Theta }_0`$, i.e., when $`\theta [\delta +\eta _0](z)`$ vanishes, except the points where the flow is tangent to $`\mathrm{\Theta }_0`$, i.e., when $`\theta [\delta +\eta _0](z)`$ has a higher vanishing order in $`x^{}`$ and $`\mu _1\mu _2`$ too. This takes place only at the origin of Jac$`(\mathrm{\Gamma })`$. Since at the origin the solution (4.8) blows up, we conclude that for bounded solutions the function $`\mu _1\mu _2`$ may have only a double zero and $`x(x^{})`$ a simple zero in $`x^{}`$.
On the other hand, in view of the second equation of (4.3), the original variable $`x`$ is a sum of Abelian integrals of third kind. Introduce the normalized differentials of third kind $`\mathrm{\Omega }_\mathrm{}_{}\mathrm{}_+`$ on $`\mathrm{\Gamma }`$ having poles at $`\mathrm{}_{},\mathrm{}_+`$ with residues $`\pm 1`$:
$$\mathrm{\Omega }_\mathrm{}_{}\mathrm{}_+=\frac{\mu ^2d\mu }{\sqrt{R_6(\mu )}}+h_1\overline{\omega }_1+h_2\overline{\omega }_2,$$
(4.12)
where $`h_1,h_2`$ are normalizing constants specified by the conditions for $`\mathrm{\Omega }_\mathrm{}_{}\mathrm{}_+`$ to have zero $`A`$-periods on $`\mathrm{\Gamma }`$.
According to Clebsch and Gordan ,
$$_{P_0}^{P_1}\mathrm{\Omega }_\mathrm{}_{}\mathrm{}_++_{P_0}^{P_2}\mathrm{\Omega }_\mathrm{}_{}\mathrm{}_+=\mathrm{log}\frac{\theta [\delta ](z+q/2)}{\theta [\delta ](zq/2)}+\mathrm{const}.$$
(4.13)
Then, in view of the second equation in (4.2) and (4.12), we get
$$\begin{array}{c}x(x^{})=\mathrm{log}\frac{\theta [\delta ](z+q/2)}{\theta [\delta ](zq/2)}h_1z_1h_2z_2+\mathrm{const},\\ \\ z_1=d_{11}x^{}+\mathrm{const},z_2=d_{21}x^{}+\mathrm{const}.\end{array}$$
(4.14)
As a result, we expressed the stationary quasi-periodic solution $`U`$ and the argument $`x`$ in terms of the auxiliary complex variable $`x^{}`$. The algebraic geometrical structure of the general solution $`U(x,t)`$ and the behaviour of real solutions will be considered in the next sections.
### Stationary quasi-periodic solutions for the Dym equation.
Now we pass to the Dym equation (1.2) and seek its solutions again in the form (4.1). In this case the variables $`\mu _1,\mu _2`$ again change according to equations of the form (4.2) with the only difference being that the order of the polynomial defining the corresponding hyperelliptic curve is odd:
$$\begin{array}{c}\frac{\mu _1d\mu _1}{2\sqrt{R_5(\mu _1)}}+\frac{\mu _2d\mu _2}{2\sqrt{R_5(\mu _2)}}=dt,\hfill \\ \\ \frac{\mu _1^2d\mu _1}{2\sqrt{R_5(\mu _1)}}+\frac{\mu _2^2d\mu _2}{2\sqrt{R_5(\mu _2)}}=dx,\hfill \\ \\ R_5(\mu )=\kappa \mu (\mu a_1)\mathrm{}(\mu a_4),\hfill \end{array}$$
(4.15)
hence the corresponding hyperelliptic curve $`\mathrm{\Gamma }=\{w^2=R_5(\mu )\}`$ has just one infinite point $`\mathrm{}`$. As a consequence, the equations (4.15) contain one holomorphic differential and one differential of the second kind.
As before, we first consider stationary solutions by putting $`t=t_0`$ $`(dt=0)`$ and assuming $`\kappa =1`$. Notice that under these conditions, (4.15) has the same structure as quadratures for the Jacobi problem on geodesics on a triaxial ellipsoid $`Q`$, where $`\mu _1,\mu _2`$ play the role of ellipsoidal coordinates on $`Q`$, parameters $`a_1,a_2,a_3`$ the squares of the semi-axes of $`Q`$, $`a_4`$ the constant of motion, and $`x`$ the length of a geodesic.
Under the change of parameter (4.4), we arrive at the Abel–Jacobi mapping
$$\begin{array}{c}_{P_0}^{P_1}\frac{d\mu }{2\sqrt{R_5(\mu )}}+_{P_0}^{P_2}\frac{d\mu }{2\sqrt{R_5(\mu )}}=u_1,\hfill \\ \\ _{P_0}^{P_1}\frac{\mu d\mu }{2\sqrt{R_5(\mu )}}+_{P_0}^{P_2}\frac{\mu d\mu }{2\sqrt{R_5(\mu )}}=u_2,\hfill \\ \\ u_1=x^{}+\mathrm{const},u_2=\mathrm{const},\hfill \end{array}$$
(4.16)
This change was first made by Weierstrass in order to find the theta-functional solution for the geodesic problem (see also Cewen ). Next we introduce normalized holomorphic differentials $`\overline{\omega }_1,\overline{\omega }_2`$ on $`\mathrm{\Gamma }`$ and coordinates $`z_1,z_2`$ on the universal covering of Jac$`(\mathrm{\Gamma })`$ according to (4.6) and, in addition, the normalized differential of the second kind having a double pole at $`\mathrm{}`$
$$\mathrm{\Omega }_{\mathrm{}}^{(1)}=\frac{\mu _i^2d\mu _i}{2\sqrt{R_5(\mu _i)}}+h_1^{}\overline{\omega }_1+h_2^{}\overline{\omega }_2.$$
(4.17)
Similarly to (4.12), the constants $`h_1^{},h_2^{}`$ are uniquely defined by the condition that $`\mathrm{\Omega }_{\mathrm{}}^{(1)}`$ have zero $`A`$-periods on $`\mathrm{\Gamma }`$.
Then, instead of the expressions (4.8), and (4.10), we have (see, e.g., Dubrovin , Dubrovin et al. )
$$\begin{array}{c}U(x^{})=\mu _1+\mu _2=\mathrm{const}_V^2\theta [\delta ](z),\hfill \\ \\ z_1=d_{11}x^{}+\mathrm{const},z_2=d_{21}x^{}+\mathrm{const},\hfill \end{array}$$
(4.18)
where $`_V`$ is the derivative along the tangent vector $`V`$ of $`\mathrm{\Gamma }`$Jac$`(\mathrm{\Gamma })`$ at $`\mathrm{}`$: $`V=(d_{12},d_{22})^T`$, and, respectively,
$$\mu _1\mu _2=\kappa \frac{\theta ^2[\delta +\eta _0](z)}{\theta ^2[\delta ](z)},\kappa =\mathrm{const},$$
(4.19)
where the characteristic $`\eta _0`$ is defined in (4.11). In addition, in contrast to (4.13), the sum of Abelian integrals of second kind has the form
$$_{P_0}^{P_1}\mathrm{\Omega }_{\mathrm{}}^{(1)}+_{P_0}^{P_2}\mathrm{\Omega }_{\mathrm{}}^{(1)}=\mathrm{const}_V\mathrm{log}\theta [\delta ](z).$$
(4.20)
Comparing this with (4.17), we find that the analog of the relation (4.14) between the parameters $`x`$ and $`x^{}`$ has the form
$$\begin{array}{c}x(x^{})=_V\mathrm{log}\theta [\delta ](z)h_1z_1h_2z_2+\mathrm{const},\hfill \\ \\ z_1=d_{11}x^{}+\mathrm{const},z_2=d_{21}x^{}+\mathrm{const}.\hfill \end{array}$$
(4.21)
This expression can be regarded as a 2-dimensional generalization of the Weierstrass zeta-function in (2.6).
Thus, we have expressed the stationary solution $`U`$ and the argument $`x`$ in terms of the auxiliary complex variable $`x^{}`$. Various types of real solutions defined by the above expressions will be considered in Section 6.
## 5 Time-dependent quasi-periodic solutions.
### The solutions for the SW equation.
In order to obtain general time-dependent solutions $`U(x,t)`$ of the SW equation given by (4.1), one has to invert the mapping (4.3). However, as already mentioned, the problem of inversion is unsolvable in terms of meromorphic functions.
To describe the structure of general solutions, let us first consider a divisor of 3 points $`P_i=(\mu _i,w_i)`$, $`i=1,2,3`$ on $`\mathrm{\Gamma }\{\mathrm{}_{},\mathrm{}_+\}`$ and the following extended equations
$$\underset{i=1}{\overset{3}{}}\frac{d\mu _i}{2\sqrt{R_6(\mu _i)}}=dy,\underset{i=1}{\overset{3}{}}\frac{\mu _id\mu _i}{2\sqrt{R_6(\mu _1)}}=dt,\underset{i=1}{\overset{3}{}}\frac{\mu _i^2d\mu _i}{2\sqrt{R_6(\mu _i)}}=dx,$$
(5.1)
including the extra variable $`y`$, two holomorphic differentials and one differential of the third kind on $`\mathrm{\Gamma }`$. The latter are linear combinations of the normalized differentials $`\overline{\omega }_1,\overline{\omega }_2,\mathrm{\Omega }_\pm \mathrm{}`$ defined in (4.6) and (4.12). According to Clebsch and Gordan , Fedorov , equations (5.1) describe a differential of a well defined mapping of the symmetric product $`(\mathrm{\Gamma }\{\mathrm{}_{},\mathrm{}_+\})^{(3)}`$ to generalized Jacobian variety Jac$`(\mathrm{\Gamma },\mathrm{}_\pm )`$, a noncompact algebraic group represented as the quotient of $`^3`$ by a lattice $`\mathrm{\Lambda }`$ generated by five vectors of periods of the differentials $`\overline{\omega }_1,\overline{\omega }_2,\mathrm{\Omega }_\pm \mathrm{}`$ on $`\mathrm{\Gamma }`$. Topologically, Jac$`(\mathrm{\Gamma },\mathrm{}_\pm )`$ is the product of the 2 dimensional variety Jac$`(\mathrm{\Gamma })`$ and the cylinder $`^{}=\{0\}`$. An analytical and algebraic-geometrical description of generalized Jacobians can be found in Clebsch and Gordan , Belokolos et all , Fedorov , Gavrilov .
Let $`z_1,z_2,Z`$ be coordinates on the universal covering of Jac$`(\mathrm{\Gamma },\mathrm{}_\pm )`$ such that
$$\underset{i=1}{\overset{3}{}}_{P_0}^{P_i}\overline{\omega }_1=z_1,\underset{i=1}{\overset{3}{}}_{P_0}^{P_i}\overline{\omega }_2=z_2,\underset{i=1}{\overset{3}{}}_{P_0}^{P_i}\mathrm{\Omega }_\pm \mathrm{}=Z,$$
(5.2)
where, as above, $`P_0=(a_5,0)`$. Then, according to (4.6), (4.12),
$$\begin{array}{c}z_1=d_{11}y+d_{12}t+\mathrm{const},z_2=d_{21}y+d_{22}t+\mathrm{const},\hfill \\ Z=x+h_1(d_{11}y+d_{12}t)+h_2(d_{21}y+d_{22}t)+\mathrm{const}.\hfill \end{array}$$
(5.3)
The problem of inversion of Abel–Jacobi mappings including differentials of the third and second kind is solved in terms of generalized theta-functions which are finite sums of products of customary theta-functions, rational functions, and exponentials (see Ercolani , Fedorov , Gagnon et al. ). To invert the mapping (5.2), we shall make use of the following theta-functions
$$\begin{array}{c}\stackrel{~}{\theta }(z,Z)=e^{Z/2}\theta (z+q/2)+e^{Z/2}\theta (zq/2),\hfill \\ \\ \stackrel{~}{\theta }[\eta ](z,Z)=e^{Z/2}\theta [\eta ](z+q/2)+e^{Z/2}\theta [\eta ](zq/2),\hfill \end{array}$$
(5.4)
$$z=(z_1,z_2),q=(q_1,q_2)^T,q_1=_{\mathrm{}_{}}^\mathrm{}_+\overline{\omega }_1,q_2=_{\mathrm{}_{}}^\mathrm{}_+\overline{\omega }_2,$$
where $`\theta (z),\theta [\eta ](z)`$ are customary theta-functions associated with the curve $`\mathrm{\Gamma }`$ with half-integer theta-characteristics $`\eta `$. Like $`\theta [\eta ](z)`$, generalized theta-functions have a quasi-periodic property: a shift of the argument $`(z,Z)`$ by any period vector of the generalized Jacobian results in multiplication of $`\stackrel{~}{\theta }[\eta ](z,Z)`$ by a constant factor.
Now consider the dissection $`\stackrel{~}{\mathrm{\Gamma }}`$ of $`\mathrm{\Gamma }`$ along the canonical cycles $`A_1,A_2,B_1,B_2`$, which is a one-connected domain having the form of an octagon. In addition, we cut $`\stackrel{~}{\mathrm{\Gamma }}`$ along the paths joining a point $`O`$ on the boundary $`\stackrel{~}{\mathrm{\Gamma }}`$ of $`\stackrel{~}{\mathrm{\Gamma }}`$ to the points $`\mathrm{}_{},\mathrm{}_+`$. On the obtained domain $`\stackrel{~}{\mathrm{\Gamma }}^{}`$ we introduce the single-valued function $`\stackrel{~}{F}(P)=\stackrel{~}{\theta }[\delta ](\stackrel{~}{𝒜}(P)(z,Z)^T)`$, where
$$\stackrel{~}{𝒜}(P)=(_{P_0}^P\overline{\omega }_1,_{P_0}^P\overline{\omega }_2,_{P_0}^P\mathrm{\Omega }_\mathrm{}_\pm )^T,$$
and the characteristic $`\delta `$ is defined in (4.9). Then the following analog of the Riemann theorem holds (see e.g., Fedorov , Gagnon et al. ).
###### Theorem 5.1
Let the coordinates $`z,Z`$ be such that the function $`\stackrel{~}{F}(P)`$ does not vanish identically on $`\stackrel{~}{\mathrm{\Gamma }}^{}`$. Then it has precisely 3 zeros $`P_1,P_2,P_3`$ giving a unique solution to the inversion of the generalized mapping (5.2).
Now let us consider the logarithmic differential $`\mu (P)d\mathrm{log}\stackrel{~}{F}(P)`$. By Theorem 5.1, the sum of the residues of its poles in the domain $`\stackrel{~}{\mathrm{\Gamma }}^{}`$ equals $`\mu (P_1)+\mu (P_2)+\mu (P_3)`$. Applying the residue theorem, after calculations, we get the following compact “trace formula”
$$\mu _1+\mu _2+\mu _3=\mathrm{const}\frac{e^Z\theta [\delta ](z+q)+e^Z\theta [\delta ](zq)}{\theta [\delta ](z)}$$
(5.5)
with the characteristic $`\delta `$ specified in (4.9).
The principal difference between the extended mappings (5.1) or (5.2) and the system (4.3) is that the latter contains only 2 points on $`\mathrm{\Gamma }\{\mathrm{}_{},\mathrm{}_+\}`$. On the other hand, (5.1) reduces to (4.2) by fixing $`P_3P_0`$ ($`\mu _3a_5`$, $`d\mu _30`$). Under this condition, (5.2) describes the embedding of the symmetric product $`(\mathrm{\Gamma }\{\mathrm{}_{},\mathrm{}_+\})^{(2)}`$ into Jac$`(\mathrm{\Gamma },\mathrm{}_\pm )`$. Its image is a 2-dimensional nonlinear analytic subvariety (stratum) $`W_2`$. Like the generalized Jacobian itself, it is a noncompact variety.
### Remark 5.1.
In case of customary Jacobian varieties, the corresponding nonlinear subvarieties and their stratification have been studied in Gunning and Vanhaecke . Such varieties or their open subsets often appear as (coverings of) complex invariant manifolds of finite-dimensional integrable systems (see Vanhaecke , Abenda and Fedorov ).
It follows from the above that on the stratum $`W_2`$ the variables $`z_1,z_2,Z`$ play the role of excessive (abundant) coordinates, hence they cannot be independent there. The analytic structure of $`W_2`$ is explicitly described by the following theorem (see e.g., Fedorov , Gagnon et al. ).
###### Theorem 5.2
The subvariety $`W_2\mathrm{Jac}(\mathrm{\Gamma },\mathrm{}_\pm )`$ coincides with the zero locus of the generalized theta-function:
$$W_2=\{e^{Z/2}\theta [\delta ](z+q/2)e^{Z/2}\theta [\delta ](zq/2)=0\}.$$
(5.6)
On the other hand, in view of relations (5.3), the coordinates $`z,Z`$ are linear functions of the variables $`x,t`$, and $`y`$. Thus, equation (5.6) can also be regarded as a constraint on them. It follows that on fixing $`P_3=P_0`$, $`y`$ becomes a transcendent function of $`x,t`$.
Now we notice that the sum $`\mu _1+\mu _2+a_5=\mu (P_1)+\mu (P_2)+\mu (P_0)`$ coincides with the restriction of the total sum $`\mu (P_1)+\mu (P_2)+\mu (P_3)`$, as a function on Jac$`(\mathrm{\Gamma },\mathrm{}_\pm )`$, onto $`W_2`$. Then, using expression (5.5), we conclude that the 2-phase solution of the SW equation has the form
$$U(x,t)=\mathrm{const}\frac{e^Z\theta [\delta ](z+q)+e^Z\theta [\delta ](zq)}{\theta [\delta ](z)}$$
(5.7)
$$z_1=d_{11}y+d_{12}t,z_2=d_{21}y+d_{22}t,Z=x+h_1(d_{11}y+d_{12}t)+h_2(d_{21}y+d_{22}t),$$
where the extra variable $`y`$ depends on $`x,t`$ according to (5.6). As a result, we arrive at the following algebro-geometric description of motion:
The $`x`$-flow ($`t`$-flow) defined by equations (4.2) evolves on the nonlinear variety $`W_2\mathrm{Jac}(\mathrm{\Gamma },\mathrm{}_\pm )`$ in such a way that $`y`$ is a nonlinear transcendent function of $`x`$ (respectively, of $`t`$). In this sense the flow is nonlinear.
We emphasize that the solution $`U(x,t)`$ is neither meromorphic in $`x`$, nor in $`t`$.
### Remark 5.2.
Let us consider the $`x`$-flow by putting $`t=`$const. It turns out that, up to an additive constant, the extra variable $`y`$ can now be identified with the auxiliary variable $`x^{}`$ introduced in (4.4) when we considered stationary solutions. Indeed, in view of (5.3), in this case the condition in (5.6) becomes
$$Z=x+h_1z_1+h_2z_2+\mathrm{const}=\mathrm{log}\frac{\theta [\delta ](zq/2)}{\theta [\delta ](z+q/2)}+\mathrm{const},$$
(5.8)
which is equivalent to the relation (4.14) between $`x`$ and $`x^{}`$. In view of (5.8) and the addition theorem for theta-functions, the solution (5.7) reduces to the stationary solution (4.8).
In contrast to $`x`$, the parameter $`t`$ enters both expressions for $`z`$ and $`Z`$ in (5.7). Therefore, in the case of the $`t`$-flow, $`t`$ cannot be explicitly expressed in terms of $`y`$ as in the case of the $`x`$-flow. This implies that solutions $`U(x_0,t)`$, $`x_0=`$const must have different properties in comparison with (4.8).
### Remark 5.3.
We notice that the subvariety $`W_2`$ of the generalized Jacobian becomes linear in rare cases when the curve $`\mathrm{\Gamma }`$ enjoys some nontrivial involutions, i.e., when it can be regarded as a covering of an elliptic curve. (Various examples of the involutions can be found in Belokolos et al .) In such cases $`U(x,t)`$ becomes a meromorphic function of its arguments.
### The solutions for the Dym equation.
Now we proceed to the problem of inversion of the reduction (4.15) of the Dym equation which is related to the odd order hyperelliptic curve $`\mathrm{\Gamma }=\{w^2=R_5(\mu )\}`$. As in the case of the reduction of the SW equation, in order to describe the function $`U(x,t)=\mu _1+\mu _2`$, we first consider an “excessive” divisor of 3 points $`P_i=(\mu _i,w_i)`$, $`i=1,2,3`$ on $`\mathrm{\Gamma }\{\mathrm{}\}`$ and the extended equations
$$\begin{array}{c}\underset{i=1}{\overset{3}{}}\frac{d\mu _i}{2\sqrt{R_5(\mu _i)}}=dy,\underset{i=1}{\overset{3}{}}\frac{\mu _id\mu _i}{2\sqrt{R_5(\mu _1)}}=dt,\underset{i=1}{\overset{3}{}}\frac{\mu _i^2d\mu _i}{2\sqrt{R_5(\mu _i)}}=dx,\hfill \\ \\ R_5(\mu )=\kappa \mu (\mu a_1)\mathrm{}(\mu a_4),\hfill \end{array}$$
(5.9)
including 2 holomorphic differentials and one differential of the second kind having a double pole at $`\mathrm{}\mathrm{\Gamma }`$. They are linear combinations of the normalized differentials $`\overline{\omega }_1,\overline{\omega }_2,\mathrm{\Omega }_{\mathrm{}}^{(1)}`$ defined in (4.6) and (4.17).
In contrast to (5.1), equations (5.9) describe a differential of a well defined mapping of the symmetric product $`(\mathrm{\Gamma }\{\mathrm{}\})^{(3)}`$ to the generalized Jacobian variety Jac$`(\mathrm{\Gamma },\mathrm{})`$, the quotient of $`^3`$ by the lattice generated by four period vectors of the differentials $`\overline{\omega }_1,\overline{\omega }_2,\mathrm{\Omega }_{\mathrm{}}^{(1)}`$ on $`\mathrm{\Gamma }`$. Topologically, this variety is a product of the 2 dimensional variety Jac$`(\mathrm{\Gamma })`$ and the complex plane $``$ (see Clebsch , Gavrilov ).
Let us introduce coordinates $`z_1,z_2,Z`$ by the mapping
$$\underset{i=1}{\overset{3}{}}_{E_0}^{P_i}\overline{\omega }_1=z_1,\underset{i=1}{\overset{3}{}}_{E_0}^{P_i}\overline{\omega }_2=z_2,\underset{i=1}{\overset{3}{}}_{E_0}^{P_i}\mathrm{\Omega }_{\mathrm{}}^{(1)}=Z$$
(5.10)
with the basepoint $`E_0=(0,0)`$ (we cannot choose the basepoint to be $`\mathrm{}`$ as in the previous section, since it is the pole of $`\mathrm{\Omega }_{\mathrm{}}^{(1)}`$). Next, comparing (4.6), (4.17) with (5.9), we find the following relations
$$\begin{array}{c}z_1=d_{11}y+d_{12}t+\mathrm{const},z_2=d_{21}y+d_{22}t+\mathrm{const}\hfill \\ Z=x+h_1^{}(d_{11}y+d_{12}t)+h_2^{}(d_{21}y+d_{22}t)+\mathrm{const}.\hfill \end{array}$$
(5.11)
Like (5.2), the mapping (5.10) is invertible in terms of meromorphic functions. The inversion problem is solved by means of the following rational degeneration of the customary theta-function
$$\widehat{\theta }(z,Z)=Z\theta [\delta ](z)+_V\theta [\delta ](z),$$
(5.12)
where $`_V`$ is defined in (4.18) (compare with the generalized theta-functions (5.4)). Like (5.4), the function (5.12) enjoys the quasi-periodic property.
Consider again the dissection $`\stackrel{~}{\mathrm{\Gamma }}`$ of $`\mathrm{\Gamma }`$ and cut it along a path joining a point $`O`$ on the boundary $`\mathrm{\Gamma }`$ to $`\mathrm{}`$. In the obtained domain we introduce the single-valued function
$$\widehat{F}(P)=\left(Z_{E_0}^P\mathrm{\Omega }_{\mathrm{}}^{(1)}\right)\theta [\delta ]\left(z_{E_0}^{\mathrm{}}\overline{\omega }_{E_0}^P\overline{\omega }\right)+_V\theta [\delta ]\left(z_{E_0}^{\mathrm{}}\overline{\omega }_{E_0}^P\overline{\omega }\right).$$
Using a modification of Theorem 5.1 and calculating the logarithmic differential $`\mu (P)d\mathrm{log}\widehat{F}(P)`$, we obtain
$$\begin{array}{c}\mu _1+\mu _2+\mu _3=\mathrm{const}(Z+_V\mathrm{log}\theta [\delta +\eta _0](z))^2_V^2\mathrm{log}\theta [\delta +\eta _0](z)\hfill \\ \\ =\mathrm{const}Z^2\frac{2Z_V\theta [\delta +\eta _0](z)_V^2\theta [\delta +\eta _0](z)}{\theta [\delta +\eta _0](z)},\hfill \end{array}$$
(5.13)
where $`\eta _0=(\eta _0^{\prime \prime },\eta _0^{})^T\frac{1}{2}^2/^2`$, such that $`2\pi i\eta _0^{\prime \prime }+𝐁\eta _0^{}=_{E_0}^{\mathrm{}}(\overline{\omega }_1,\overline{\omega }_2)^T`$. Now, similarly to the case of the SW equation, we fix $`P_3E_0`$ ($`\mu _3=0`$, $`d\mu _30`$) in the mapping (5.10). In this case its image becomes a 2-dimensional nonlinear noncompact analytic subvariety $`\widehat{W}_2\mathrm{Jac}(\mathrm{\Gamma },\mathrm{})`$. Comparing the third sum in (5.10) and expression (4.20), we find
$$\widehat{W}_2=\{Z+\mathrm{const}+_V\mathrm{log}\theta [\delta +\eta _0](z)=0\}.$$
(5.14)
Finally, taking into account the trace formula (5.13), we conclude that the solution of the Dym equation has the form
$$\begin{array}{c}U(x,t)=\mu _1+\mu _2=\mathrm{const}_V^2\mathrm{log}\theta [\delta +\eta _0](z),\hfill \\ \\ z_1=d_{11}y+d_{12}t+\mathrm{const},z_2=d_{21}y+d_{22}t+\mathrm{const},\hfill \end{array}$$
(5.15)
where the extra variable $`y`$ depends on $`x,t`$ in a transcendental way according to the constraint (5.14) and the expression for $`Z`$ in (5.11). The solution $`U(x,t)`$ is not meromorphic with respect to its arguments.
### Remark 5.4.
As in the case of SW equation, the stationary solutions for the Dym equation given in the previous section can be obtained from (5.15) by putting $`t=`$const. Then $`y`$ can be identified with the auxiliary variable $`x^{}`$ defined in (4.4) and the condition in (5.14) becomes equivalent to the relation (4.21) between $`x`$ and $`x^{}`$. As a result, (5.15) gives precisely the stationary solution (4.18).
## 6 Real bounded stationary 2-phase solutions.
In this section we impose reality conditions on the stationary complex solutions obtained in Section 4.
Let $`\sigma `$ be the antiholomorphic involution on a hyperelliptic curve $`\mathrm{\Gamma }=\{w^2=P(\mu )\}`$ of genus $`g`$. The part of $`\mathrm{\Gamma }`$ which is invariant with respect to $`\sigma `$ is called the real part $`\mathrm{\Gamma }()`$. On the plane $`^2=(\mathrm{}\mu ,\mathrm{}w)`$ it is either the empty set or a union of ovals. By the Abel–Jacobi mapping, the involution $`\sigma `$ lifts to Jac($`\mathrm{\Gamma }`$). By $`\mathrm{Jac}_{}(\mathrm{\Gamma })`$ we denote the real part of Jac($`\mathrm{\Gamma }`$) that is invariant under $`\sigma `$. One can show that the elementary symmetric functions of the variables $`\mu _1,\mathrm{},\mu _g`$ take real values on $`\mathrm{Jac}_{}(\mathrm{\Gamma })`$ and only there.
###### Theorem 6.1
(Comessatti ). Let $`s`$ be the number of connected components of $`\mathrm{\Gamma }()`$ and $`L`$ be the number of connected components of $`\mathrm{Jac}_{}(\mathrm{\Gamma })`$. If $`s0`$, then $`L=2^{s1}`$. If $`s=0`$, then $`L=1`$ provided the degree of $`R(\mu )`$ is even and $`L=2`$ in case the degree is odd.
### Shallow water equation.
Suppose all the roots of the polynomial $`R_6(\mu )`$ in (4.2) arising in the reduction of the SW equation are real, i.e., $`\mathrm{\Gamma }()`$ consists of 3 ovals about the segments $`[0,e_1]`$, $`[e_2,e_3]`$, and $`[e_4,e_5]`$. By Theorem 5.1, $`\mathrm{Jac}_{}(\mathrm{\Gamma })`$ has 4 connected components. They are characterized by the following behavior of $`\mu `$-variables, which reflects in different properties of real stationary solutions $`U(x,t_0)`$:
### Case 1.
The variables $`\mu _1,\mu _2`$ are real and $`\mu _1[a_2,a_3]`$, $`\mu _2[a_4,a_5]`$. The sum $`U=\mu _1+\mu _2`$ is thus a real quasi-periodic function of $`x^{}`$ having no poles and zeros. The product $`\mu _1\mu _2`$ has the same properties. Geometrically this means that the corresponding component of $`\mathrm{Jac}_{}(\mathrm{\Gamma })`$ does not intersect the translates $`\mathrm{\Theta }_{},\mathrm{\Theta }_+`$, and $`\mathrm{\Theta }_0`$. In view of (4.4), $`x(x^{})`$ and $`x^{}(x)`$ are strictly monotonic real functions. Therefore the composition $`U(x,t_0)=U(x^{}(x))`$ is a quasi-periodic regular function.
### Cases 2,3.
$`\mu _1[0,a_1]`$, whereas $`\mu _2[a_2,a_3]`$ or $`[a_4,a_5]`$. The function $`U(x^{})`$ has the same properties as above, whereas $`\mu _1\mu _2`$ does not blow up, but has zeros. As found in Section 4, the derivative $`dx/dx^{}`$ vanishes with a second order with respect to $`x^{}`$ as one of the $`\mu `$-variables vanishes. This happens when the real $`x^{}`$-flow on the considered components of $`\mathrm{Jac}_{}(\mathrm{\Gamma })`$ intersects $`\mathrm{\Theta }_0()=\mathrm{\Theta }_0\mathrm{Jac}_{}(\mathrm{\Gamma })`$. If follows that at this moment the derivative $`dU/dx=dU/dx^{}dx^{}/dx`$ blows up and the graph of the function $`U(x,t_0)`$ has an inflection point with vertical tendency line.
In addition, when $`(\mu _1,\mu _2)=(0,a_2)`$ or $`(0,a_4)`$, i.e., when the real $`x^{}`$-flow passes a half-period on $`\mathrm{Jac}_{}(\mathrm{\Gamma })`$, the function $`U(x^{})`$ has an extremum, which implies that the graph of $`U(x,t_0)`$ has a cusp. Due to quasi-periodicity of the flow, $`U(x,t_0)`$ has an infinite quasi-periodic sequence of cusps.
### Case 4.
Now the variables $`\mu _1,\mu _2`$ are complex conjugated. Using equations (4.2), we show that they cannot reach real axis. It follows that the product $`\mu _1\mu _2`$ is always nonzero and $`U(x,t_0)`$ is again a quasi-periodic regular function.
In a similar way one can show that when some of the roots of $`R_6(\mu )`$ are complex conjugate, the qualitative behavior of the real solution $`U(x,t_0)`$ coincides with one of the above four cases.
In a forthcoming paper we will consider different singular limits of the quasi-periodic solutions when the spectral curve becomes singular and its arithmetic genus drops to zero. The solutions are then expressed in terms of purely exponential tau-functions and, in the real bounded case, they describe an interaction of the two smooth solitons or cuspons, or a quasi-periodic train of peakons tending to a periodic one at infinity.
## 7 Peakon-soliton solutions and elliptic billiards.
In this section we continue studying degenerate solutions of the Dym equation. Consider another possible confluence of roots of the polynomial $`R_5(\mu )`$ in (4.15) by putting
$$a_1=0,a_2=a_3=b,a_4=a,\kappa =1.$$
(7.1)
As before, we first analyze stationary solutions ($`x`$-flow) by setting $`dt=0`$. Passing to the new variable $`x^{}`$ according to the change
$$dx=\mu _1\mu _2dx^{},$$
(7.2)
from (4.15) we get
$$\begin{array}{c}\frac{d\mu _1}{\mu _1\sqrt{a\mu _1}}+\frac{d\mu _2}{\mu _2\sqrt{a\mu _2}}=bdx^{},\hfill \\ \frac{d\mu _1}{(\mu _1b)\sqrt{a\mu _1}}+\frac{d\mu _2}{(\mu _2b)\sqrt{a\mu _2}}=0.\hfill \end{array}$$
(7.3)
Let us introduce the following normalized differentials of third kind on the Riemann surface $`=\{\xi ^2=a\mu \}`$ that have pairs of simple poles $`Q_1^{},Q_1^+`$ and $`Q_2^{},Q_2^+`$ with $`\mu =0`$ and $`\mu =b`$ respectively
$$\mathrm{\Omega }_1=\frac{\beta _1d\mu }{\mu \xi },\mathrm{\Omega }_2=\frac{\beta _2d\mu }{(\mu b)\xi },\beta _1=\sqrt{a},\beta _2=\sqrt{ab}.$$
Then equations (7.3) give rise to the generalized Abel–Jacobi equations
$$_{P_0}^{P_1}\mathrm{\Omega }_1+_{P_0}^{P_2}\mathrm{\Omega }_1=z_1,_{P_0}^{P_1}\mathrm{\Omega }_2+_{P_0}^{P_2}\mathrm{\Omega }_2=z_2,P_i=(\mu _i,\xi _i),$$
(7.4)
$$z_1=b\beta _1x^{}+z_{10},z_{10},z_2=\mathrm{const},$$
(7.5)
where we put $`P_0=(a,0)`$. These describe a well defined mapping of the symmetric product $`(\{Q_1^{},Q_1^+,Q_2^{},Q_2^+\})^{(2)}`$ to the generalized Jacobian Jac($`,Q_1^\pm ,Q_2^\pm `$).
As a result of inversion of (7.4), one finds the following expressions for symmetric polynomials of $`\mu _1`$ and $`\mu _2`$
$$\begin{array}{c}\mu _1+\mu _2=U(z_1,z_2)=_W^2\mathrm{log}\tau (z_1,z_2)+\beta _1^2\beta _2^2\hfill \\ =4(\beta _1^2\beta _2^2)\frac{\beta _1^2(e^{z_2}+e^{z_2})+\beta _2^2(e^{z_1}+e^{z_1})+2(\beta _1^2\beta _2^2)}{\tau ^2(z_1,z_2)}+\beta _1^2\beta _2^2,\hfill \\ \\ \mu _1\mu _2=\frac{1}{b\beta _1}_{z_1}_W\mathrm{log}\tau (z_1,z_2)\mathrm{or}\mu _1\mu _2=4ab\frac{(e^{z_2/2}+e^{z_2/2})^2}{\tau ^2(z_1,z_2)},\hfill \end{array}$$
(7.6)
where
$$_W=2\beta _1\frac{}{z_1}+2\beta _2\frac{}{z_2},z_1=b\beta _1x^{}+z_{10},z_{10},z_2=\mathrm{const},$$
and $`\tau (z_1,z_2)`$ is the 2-dimensional tau-function with $`\alpha _1,\alpha _2`$ replaced by $`\beta _1,\beta _2`$. The latter admits decomposition in the following sum of one-dimensional tau-functions
$$\begin{array}{c}\tau (z_1,z_2)=e^{z_1/2}\tau (z_2+q/2)e^{z_1/2}\tau (z_2q/2),\hfill \\ \tau (z_2)=e^{z_2/2}+e^{z_2/2},q=\mathrm{log}\left(\frac{\beta _1\beta _2}{\beta _1+\beta _2}\right)^2.\hfill \end{array}$$
(7.7)
Lastly, by using the second expression in (7.6) and the relation between $`z_1`$ and $`x^{}`$ in (7.5), we find
$$x=\mu _1\mu _2𝑑x^{}=_W\mathrm{log}\tau (z_1,z_2)+\mathrm{const}$$
or, in view of decomposition (7.7),
$$\begin{array}{c}x(x^{})=\beta _1\frac{e^{z_1/2}\tau (z_2+q/2)+e^{z_1/2}\tau (z_2q/2)}{e^{z_1/2}\tau (z_2+q/2)e^{z_1/2}\tau (z_2q/2)}\hfill \\ +\beta _2\frac{e^{z_1/2}_{z_2}\tau (z_2+q/2)e^{z_1/2}_{z_2}\tau (z_2q/2)}{e^{z_1/2}\tau (z_2+q/2)e^{z_1/2}\tau (z_2q/2)}+\mathrm{const},\hfill \\ \\ z_1=b\beta _1x^{}+z_{10},z_2=\mathrm{const}.\hfill \end{array}$$
(7.8)
### Remark 8.1.
As mentioned in Remark 4.1, equations (4.15) with $`dt=0`$ describing the quasiperiodic stationary solutions have the same structure as quadratures for the geodesic motion on an triaxial ellipsoid $``$ (more generally, a quadric) in $`^3`$. Parameter $`x`$ plays the role of length of a geodesic. Under the limit (7.1) one of the semiaxes of $``$ tends to zero whereas the geodesic motion passes to the asymptotic billiard motion inside the ellipse
$$\overline{}=\left\{X_1^2/a+X_2^2/b=1\right\}^2=(X_1,X_2).$$
Geodesics on $``$ transform to straight line segments passing through a focus of the ellipse between each subsequent elastic reflections (impacts) along $`\overline{}`$. As $`x\pm \mathrm{}`$, the billiard motion tends to oscillations along the bigger axis of the ellipse. Now the variables $`\mu _1,\mu _2`$ play the role of elliptic coordinates in $`^2`$ such that
$$X_1^2=\frac{(a\mu _1)(a\mu _2)}{ab},X_2^2=\frac{(b\mu _1)(b\mu _2)}{ba}.$$
Along $`\overline{}`$ one of the variables equals zero.
It follows that equations (7.3) can be regarded as describing the straight line motion of a point mass inside $`\overline{}`$. When the point meets the ellipse, one of the $`\mu `$-variables, say $`\mu _1`$, vanishes, and the corresponding point $`P_1=(\mu _1,\sqrt{R(\mu _1)})`$ on the Riemann surface $``$ coincides with one of the poles $`Q_1^{},Q_1^+`$ of the differential $`\mathrm{\Omega }_1`$. Then, as follows from the mapping (7.4), $`z_1`$ and $`x^{}`$ become infinite.
On the other hand, as $`x^{},z_1\pm \mathrm{}`$, the second expression in (7.6) vanishes, whereas the first one has finite limits giving the values of $`\mu _2`$ at the subsequent impact points. The variable $`z_2`$ plays the role of a constant phase defining position of the segment between the points.
According to (7.8), (7.6)
$$\begin{array}{c}x(\mathrm{},z_2)=\beta _1+\beta _2\frac{_{z_2}\tau (z_2q/2)}{\tau (z_2q/2)}+c_x,\hfill \\ x(\mathrm{},z_2)=\beta _1+\beta _2\frac{_{z_2}\tau (z_2+q/2)}{\tau (z_2+q/2)}+c_x,c_x=\mathrm{const}.\hfill \end{array}$$
(7.9)
and
$$\begin{array}{c}U(\mathrm{},z_2)=\frac{4\beta _2^2(\beta _1^2\beta _2^2)}{(\beta _1\beta _2)^2e^{z_2}+(\beta _1+\beta _2)^2e^{z_2}+2(\beta _1^2\beta _2^2)}+\beta _1^2\beta _2^2,\hfill \\ U(\mathrm{},z_2)=\frac{4\beta _2^2(\beta _1^2\beta _2^2)}{(\beta _1\beta _2)^2e^{z_2}+(\beta _1+\beta _2)^2e^{z_2}+2(\beta _1^2\beta _2^2)}+\beta _1^2\beta _2^2.\hfill \end{array}$$
(7.10)
Notice that $`x(\mathrm{},z_2)`$ and $`U(\mathrm{},z_2)`$ have the same value as $`x(\mathrm{},z_2+q)+2\beta _1`$ and $`U(\mathrm{},z_2+q)`$ respectively. All this results in the following algebro-geometrical description: As the point mass inside $`\overline{}`$ moves from one impact to the next one, the point $`P_1`$ on $``$ moves from the pole $`Q_1^{}`$ to $`Q_1^+`$. At the moment of impact, $`P_1`$ jumps from $`Q_1^+`$ back to $`Q_1^{}`$, whereas the phase $`z_2`$ in (7.6) increases by $`q`$. Then the story repeats.
Using this property, by induction, from (7.10) the elliptic coordinates of the whole sequence of impact points are found in form
$$\begin{array}{c}\mu _1=0,\mu _2=\frac{4\beta _2^2(\beta _1^2\beta _2^2)}{(\beta _1\beta _2)^2e^{z_{2N}}+(\beta _1+\beta _2)^2e^{z_{2N}}+2(\beta _1^2\beta _2^2)}+\beta _1^2\beta _2^2,\hfill \\ z_{2N}=z_{20}+Nq,\hfill \end{array}$$
(7.11)
$`N`$ being the number of impact and the constant $`z_{20}`$ is the same for all the segments of the billiard trajectory.
In addition, from (7.9) we find the length of the $`N`$-th segment of the billiard trajectory in form
$$x(\mathrm{},z_{2N})x(\mathrm{},z_{2N})=2\beta _1+2\beta _2\frac{e^{q/2}e^{q/2}}{\mathrm{exp}(z_{2N})\mathrm{exp}(z_{2N})+e^{q/2}e^{q/2}},$$
(7.12)
$`z_{20}`$ being the same as in (7.11).
According to the trace formula $`U(x,t_0)=\mu _1+\mu _2`$, expressions (7.6), (7.8) provide us stationary peakon solutions of Dym equation in a parametric form. Here the phase $`z_2`$ must be regarded as a certain function of $`t_0`$. Namely, the expressions describe one piece of the profile $`U(x,t_0)`$ corresponding to trajectory of the point mass between subsequent impacts. The other pieces are obtained by changing the phase $`z_2`$ in (7.6), (7.8) by $`q`$ and adding $`2\beta _1`$ to $`x`$. The pieces are glued at peak points, where the spatial derivative of $`U`$ changes sign and which correspond to impacts in the billiard problem. The profile $`U(x,t_0)`$ thus consists of an infinite sequence of peaks and knots between them. For this reason we call this solution the soliton-peakon solution. Notice that, in contrast to exponentials profiles of the functions $`U(x^{})`$, $`x(x^{})`$, any piece of $`U(x,t_0)`$ has quadratic profile, as will be explained below.
The heights $`U_N`$ of $`N`$-th peak is given by (7.11). The distance between subsequent peaks along $`x`$-axis is a quasiperiodic function of $`N`$ given by (7.12).
We emphasize that, in contrast to the peakon traveling wave solution considered in Section 3, now the $`x`$-distance between subsequent peaks is not constant. However, as seen from (7.12), for $`N\pm \mathrm{}`$ it tends to $`2\beta _1`$, the doubled bigger semi-axis of the ellipse, whereas the pieces tend to identical ones corresponding to periodic billiard motion along $`X_1`$-axis.
### Remark 8.2.
Expressions (7.6), (7.8) describe an asymptotic motion of an elliptic as well as a hyperbolic billiard. In the first case the initial phase $`z_{10}`$ is pure imaginary whereas in the second case it is real. According to the trace formula, the hyperbolic billiard corresponds to unbounded stationary solutions of HD equation, which is out of interest of this paper.
### Remark 8.3
. The above considerations can be extended to multi-dimensional case. Namely, following similar approach one can consider genus $`g`$ solution of HD equation described by equations (1.3), then its asymptotic stationary limit which is related to a generalized Abel–Jacobi mapping including $`g`$ meromorphic differentials of 3rd kind. Consider a billiard inside a $`g`$-dimensional ellipsoid. Then such a limit solution corresponds to asymptotic billiard trajectories that intersect $`g1`$ focal quadrics of the ellipsoid between any subsequent impacts. The resulting stationary solution $`U(x,t_0)`$ consists of an infinite series of peaks and between each subsequent peaks there are $`g1`$ knots.
In order to study time-dependent soliton-peakon solutions, we consider the system (4.15) under the limits (7.1) without changing the variable $`x`$. As a result, we arrive at
$$\begin{array}{c}\frac{d\mu _1}{2(\mu _1b)\sqrt{a\mu _1}}+\frac{d\mu _2}{2(\mu _2b)\sqrt{a\mu _2}}=dt,\hfill \\ \frac{d\mu _1}{2\sqrt{a\mu _1}}+\frac{d\mu _2}{2\sqrt{a\mu _2}}=dxbdt.\hfill \end{array}$$
(7.13)
The latter equations include one differential of third kind $`\mathrm{\Omega }_2=\frac{\beta _2d\mu }{(\mu b)\xi }`$ having simple poles $`Q_2^{}`$, $`Q_2^+`$ and one differential of second kind $`\mathrm{\Omega }_{\mathrm{}}^{(1)}`$ having a double pole at infinity on the Riemann surface $`=\{\xi ^2=a\mu \}`$. Consider the mapping
$$_{P_0}^{P_1}\mathrm{\Omega }_{\mathrm{}}^{(1)}+_{P_0}^{P_2}\mathrm{\Omega }_{\mathrm{}}^{(1)}=z_1,_{P_0}^{P_1}\mathrm{\Omega }_2+_{P_0}^{P_2}\mathrm{\Omega }_2=z_2,P_i=(\mu _i,\xi _i),$$
(7.14)
where, $`z_1=\beta _2t+Z_{10}`$, $`z_2=x+bt+z_{20}`$, $`z_{10},z_{20}=`$const and, as above, $`P_0=(a,0)`$. Integrating it explicitly, we obtain
$$\frac{(\xi _1\beta _2)(\xi _2\beta _2)}{(\xi _1+\beta _2)(\xi _2+\beta _2)}=e^{z_2},\xi _1+\xi _2=z_1.$$
(7.15)
Inverting these relations yields the following formal solution for the HD equation
$$\begin{array}{c}U(x,t)=\mu _1+\mu _2=(\xi _1+\xi _2)^22\xi _1\xi _2=z_1^22\beta _2(\beta _2z_1)\frac{e^{z_2/2}+e^{z_2/2}}{e^{z_2/2}e^{z_2/2}},\hfill \\ z_1=x+bt+z_{10},z_2=\beta _2t+z_{20},z_{10},z_{20}=\mathrm{const}.\hfill \end{array}$$
(7.16)
It is seen that $`U`$ depends on $`x`$ rationally (quadratically, as already mentioned above) and $`U`$ is unlimited as $`x\pm \mathrm{}`$.
However, this solution does not take into account the reflection phenomenon described above: when the variable $`\mu _1`$ vanishes, the corresponding point $`P_1`$ jumps from the pole $`Q_2^+`$ of the differential $`\mathrm{\Omega }_2`$ to $`Q_2^{}`$. According to mapping (7.14), this results in changing the phases $`z_1,z_2`$ in (7.16) by the constants
$$_{Q_2^{}}^{Q_2^+}\mathrm{\Omega }_{\mathrm{}}^{(1)}=2\beta _1,\mathrm{respectively}q=_{Q_2^{}}^{Q_2^+}\mathrm{\Omega }_2=2\mathrm{log}\frac{\beta _1\beta _2}{\beta _1+\beta _2},$$
the latter being already defined in (7.7). It follows that the actual solution $`U(x,t)`$ to HD equation consists of an infinite number of pieces described by (7.16) with
$$z_1=x+bt+2\beta _1N+z_{10},z_2=\beta _2t+Nq+z_{20},N$$
(7.17)
and glued along peak lines $`\{x=q_N(t)\}`$ in the plane $`(x,t)`$, where for a fixed time $`t`$, the function $`q_N(t)`$ gives $`x`$-coordinate of $`N`$-th peakon. Now if we assume $`z_{20}`$ to be imaginary and $`z_{10}`$ real, the series of pieces will provide a real bounded peakon solution.
The functions $`q_N(t)`$ can be found as follows. Along the peak lines one of the variables $`\mu `$, say $`\mu _1`$, vanishes implying $`d\mu _1=0`$, $`\xi _1\beta _2`$. Putting this into (7.15), we find
$$\frac{\xi _2\beta _2}{\xi _2+\beta _2}=e^{z_2q/2},\beta _2+\xi _2=z_1.$$
Substituting here (7.17), putting $`x=q_N(t)`$, and eliminating $`\xi _2`$, we obtain the sought expression
$$q_N(t)=btz_{10}2\beta _1N+\beta _1+\beta _2\frac{1+e^{z_2q/2}}{1e^{z_2q/2}},z_2=\beta _2t+Nq+z_{20}.$$
It follows that for $`|t|>>1`$ any peak moves with constant velocity $`b`$, and as $`t`$ evolves from $`\mathrm{}`$ to $`\mathrm{}`$ the peaks undergo the phase shift $`xx2\beta _2`$.
## Bibliography.
Abenda, S. and Fedorov.Yu , On the weak Kowalewski–Painlevé property for hyperelliptically separable systems, Acta Appl. Math. (to appear).
Ablowitz, M.J. and Segur, H , Solitons and the Inverse Scattering Transform, SIAM, Philadelphia.
Alber, M.S. and Alber, S.J. , Hamiltonian formalism for finite-zone solutions of integrable equations, C.R. Acad. Sc. Paris 301, 777-781.
Alber, M.S., R. Camassa, D.D. Holm and J.E. Marsden , The geometry of peaked solitons and billiard solutions of a class of integrable pde’s, Lett. Math. Phys. 32 137-151.
Alber, M.S., Camassa, R., Holm, D.D., and Marsden, J.E. , On the link between umbilic geodesics and soliton solutions of nonlinear PDE’s, Proc. Roy. Soc 450 677-692.
Alber, M.S., R. Camassa, Y. Fedorov, D.D. Holm and J.E. Marsden , On Billiard Solutions of Nonlinear PDE’s, Phys. Lett. A 264 171–178.
Alber, M.S., and C. Miller , On Peakon Solutions of the Shallow Water Equation, Appl.Math. Lett. (to appear).
Alber, M.S., Camassa, R., Fedorov, Yu., Holm, D.D., and Marsden J.E. , The geometry of new classes of weak billiard solutions of nonlinear PDE’s. (subm.)
Belokolos, E.D., A.I. Bobenko, V.Z. Enol’sii, A.R. Its, and V.B. Matveev Algebro-Geometric Approach to Nonlinear Integrable Equations. Springer-Verlag series in Nonlinear Dynamics.
Beals, R., D.H. Sattinger, J. Szmigielski , Multi-peakons and a theorem of Stietjes, Inverse Problems 15 L1–L4.
Beals, R., D.H. Sattinger, J. Szmigielski , Multipeakons and the Classical Moment, Advances in Mathematics (to appear)
Camassa, R. and Holm, D.D. , An integrable shallow water equation with peaked solitons, Phys. Rev. Lett., 71, 1661-1664.
Camassa, R., Holm, D.D., and Hyman, J.M. , A new integrable shallow water equation. Adv. Appl. Mech., 31, 1–33.
Cewen, C. , Stationary Harry-Dym’s equation and its relation with geodesics on ellipsoid, Acta Math. Sinica 6, 35–41.
Clebsch, A., Gordan, P. , Theorie der abelschen Funktionen. Teubner, Leipzig
Comessatti, A. and Sulle varietá abeliane reale I. Ann. Math. pure Appl., 2, (1924) 67–106.
Dmitrieva, L.A. \[1993a\] Finite-gap solutions of the Harry Dym equation, Phys.Lett.A 182 (2) 65–70.
Dmitrieva, L.A. \[1993b\], The higher -times approach to multisoliton solutions of the Harry Dym equation, J.Phys.A 26 6005–6020.
Dubrovin, B. , Theta functions and nonlinear equations. Russ. Math. Surveys, 36, 11–92.
Dubrovin, B.A., Novikov S.P., Krichiver, I.M. Integrable Systems. I. Itogi Nauki i Tekhniki. Sovr.Probl.Mat. Fund.Naprav. 4, VINITI, Moscow. English transl.: Encyclopaedia of Math.Sciences, Vol. 4, Springer-Verlag, Berlin 1989.
Ercolani, N. , Generalized theta functions and homoclinic varieties, Proc. Symp. Pure Appl. Math., 49, 87–100.
Fedorov, Yu. , Classical integrable systems and billiards related to generalized Jacobians, Acta Appl. Math., 55, 3, 151–201
Gagnon, L., Harnad, J., Hurtubise, J. and Winternitz, P. , Abelian integrals and the reduction method for an integrable Hamiltonian system, J.Math.Phys. 26, 1605–1612.
Gavrilov, L. , Generalized Jacobians of spectral curves and completely integrable systems, Math.Z. (to appear).
Gunning, R. , Lectures on Riemann Surfaces. Jacobi varieties. Princeton University Press.
Hunter, J.K., and Zheng, Y.X. , On a completely integrable nonlinear hyperbolic variational equation, Physica D 79, 361–386.
Li, Y.A. and Olver P.J. , Convergence of solitary-wave solutions in a perturbed bi-Hamiltonian dynamical system. Discrete and continuous dynamical systems, 4, 159–191.
Markushevich, A. I. , Theory of Functions of a Complex Variable, Chelsea Publishing Company: New York.
McKean, H.P. and A. Constantin , A Shallow Water Equation on the Circle, Comm.Pure Appl.Math. Vol LII 949–982.
Mumford, D. , Tata Lectures on Theta II, Birkhauser-Verlag.
Novikov, D.P. , Algebraic geometric solutions of the Harry Dym Equations, Siberian Math. J. 40 136–140.
Previato E. , Hyperelliptic quasi-periodic and soliton solutions of the nonlinear Schrödinger equation Duke Math. J. 52, I, 329–377.
Vanhaecke, P. , Integrable systems and symmetric products of algebraic curves, Math. Z. 40 143–172.
Weierstrass, K. , Über die geodätischen Linien auf dem dreiachsigen Ellipsoid, Mathematische Werke I, 257–266.
Wadati M., Konno, H., Ichikawa, Y.H. , J. Phys. Soc. Japan 47, 1698.
|
warning/0005/hep-ex0005008.html
|
ar5iv
|
text
|
# Prospects for Colliders and Collider Physics to the 1 PeV Energy Scale To appear in Proc. HEMC’99 Workshop – Studies on Colliders and Collider Physics at the Highest Energies: Muon Colliders at 10 TeV to 100 TeV; Montauk, NY, September 27-October 1, 1999, web page http://pubweb.bnl.gov/people/bking/heshop. This work was performed under the auspices of the U.S. Department of Energy under contract no. DE-AC02-98CH10886.
## I Introduction
No clear-cut consensus currently exists on the best long-term strategy for experimentation in high energy physics (HEP) over the next 50 years. This paper puts the case for continuing to aggressively raise the frontier energy reach of both hadron and lepton colliders. It is argued that a continuation at the historical rate of progress in the log-energy reach of colliders is plausible and would provide us with outstanding prospects for deepening our understanding of the elementary entities and organizing principles of our physical Universe.
In order to demonstrate the possible feasibility of such a push to higher collider energies, a proof-of-plausibility scenario is presented for future colliders that would continue at the historical pace in log-energy reach and would, by about the year 2040, attain a constituent energy reach of 1 PeV (i.e. 1000 TeV).
The proof-of-plausibility scenario is only one choice from a parameter space of plausible scenarios that might advance the energy reach of colliders at the historical pace and, even if viable, no claim is made that it is in any sense optimal. Instead, it is intended as a spur to constructive criticism and future research that will lead to its refinement and to alternative scenarios. Any such discussions will help us to assess the future prospects of HEP and to identify long-term R&D needs. In turn, this will enable the field to make more informed planning decisions towards our long-term future.
The paper begins with motivational background on the essential role of past and future colliders for our understanding of elementary particle physics. It then reviews the technical challenges of energy frontier colliders and presents, and then evaluates, the aforementioned proof-of-plausibility scenario.
In more detail, the paper is organized as follows. Section II provides a brief historical review of the impressive historical gains in physical understanding from past accelerator experiments and then turns to an outline of the heady physics goals for future colliding accelerators. Section III gives a wish-list, motivated by physics considerations, for the technical specifications and capabilities of future colliders. These include increased energy (mainly), specification of the physics requirements for luminosity, and the physics advantages in being able to study more than one type of projectile collisions. Section IV reviews the rate of historical progress in the energy reach of colliders as characterized by the famous Livingston plot. It then introduces the proof-of-plausibility scenario as an extension to the Livingston plot. The technical challenges and potential energy reaches future $`\mathrm{e}^+\mathrm{e}^{}`$, hadron and $`\mu ^+\mu ^{}`$ colliders are briefly assessed in sections V through VII, respectively, and justifications for the specific collider parameter choices of the proof-of-plausibility scenario are embedded in the more general discussions of these sections. The scenario as a whole is then assessed in section VIII, before concluding with a summary of the issues highlighted by the paper.
## II Colliders at the Energy Frontier are Indispensable
The continuing motivation for colliding accelerators (colliders) is to explore the most fundamental mysteries of the natural Universe: what are its fundamental entities and organizing principles ? What is the nature of space and time ? How did the Universe originate and evolve ?
Accelerators provide us with experimental insights on our Universe that could not plausibly be obtained in other ways and, to quote Harvard theorist Sidney R. Coleman Sci Am , “Experiment is the source of imagination. All the philosophers in the world thinking for thousands of years couldn’t come up with quantum mechanics”. The properties and spectrum of elementary particles have been no less hidden from theoretical understanding than was quantum mechanics.
This section reviews the central historical importance of accelerators in uncovering what is known today as the Standard Model of elementary particles. It then turns to our future aspirations for a yet deeper understanding of the elemental entities and physical principles of our Universe. In our most optimistic hopes, this might ultimately be described by a complete and logically self-consistent “theory of everything”.
### II.1 The Historical Importance of Accelerators
The past fifty years of experiments at accelerators have lead to remarkable progress in our understanding of the elementary processes and building blocks of our physical Universe. We discuss, in turn, the new insights gained on photons and on the building blocks of everyday matter, and then briefly summarize our current state of knowledge as encapsulated in the Standard Model of elementary particles.
#### A Context for Understanding Photons
Surprisingly, accelerators have greatly expanded our understanding of the multiple roles that photons play in the make-up and runnings of our Universe. It is manifestly obvious that these important insights could not have been attained without accelerator experiments – and this despite the fact that the photons themselves are all around us and can be studied in many ways.
In a famous quote Einstein , Albert Einstein once confessed that “All the fifty years of conscious brooding have brought me no closer to the answer to the question, “what are light quanta?””. While we certainly still can’t claim full understanding, discoveries at accelerators have at least moved us towards a context and framework for understanding the photon that even Einstein could not have suspected. Far from being an isolated entity, the photon has massive siblings – known as the $`Z^0`$ and $`W^\pm `$ – that have been produced and studied at colliders, with observed masses of $`\mathrm{M}_\mathrm{Z}=91.2\mathrm{GeV}/\mathrm{c}^2`$ and $`\mathrm{M}_\mathrm{W}=80.4\mathrm{GeV}/\mathrm{c}^2`$, respectively. Their close relationship to the photon has been well established from their observed properties, and the interactions and relative couplings of the photon, Z and W to other elementary particles are now precisely specified by a theory, known as the electroweak theory, that unites the electromagnetic and weak interactions. (Experiments at upcoming colliders may further expand the scope of this theoretical framework to include the mechanism for generating all mass, as will be presented in the following subsection II.2.)
In stark contrast to the photon itself, the large masses and immediate decays of its sibling Z and W bosons clearly preclude their direct study outside collider experiments. However, their effects are still seen in our everyday world since it is their interactions that turn out to be ultimately responsible for many radioactive decays – an everyday phenomenon that has been made a little less mysterious through the understanding provided by the electroweak theory.
#### Understanding Matter
Besides photons, our understanding of the building blocks of everyday matter has been revolutionized by accelerators. Electrons have passed progressively more stringent experimental tests of whether they are indeed point particles. On the other hand, protons and neutrons have been exposed as being composite entities rather than point particles. Experiments at accelerators have found them to be composed of hitherto unknown elementary particles: up-type, down-type and a few strange-type quarks bound together by gluons.
#### A Periodic Table of Elementary Particles
Accelerator experiments and, to a lesser extent, cosmic ray experiments have also shown us convincingly that the limitation of everyday matter to electrons, protons and neutrons is merely an “accident” of our low energy environment. Heavier forms of matter exist but cosmic ray interactions are the only naturally occurring process on Earth to supply the energy densities required to produce them. The few such particles that are produced sporadically by cosmic rays then decay almost instantaneously (sometimes in cascades of several stages) down to the familiar everyday particles – photons, electrons, protons and neutrons – plus the ghost-like neutrinos that are also all around us but are almost undetectable. Accelerator experiments are the only place where these particles can be systematically studied.
The list of additions to our everyday matter is impressive. Besides the aforementioned W and Z bosons, the electron has been found to have heavier siblings – the muon and the tau particle, and each of these has a neutral counterpart known as a neutrino that interacts so feebly that it is difficult to detect. Hundreds of quarks-plus-gluons bound states besides the proton and neutron have also been discovered, including bound states containing other, heavier quarks than the up and down quark – the so-called second generation strange and charm quarks and the third generation bottom (or beauty) and top quarks.
The new elementary particles fill out a veritable periodic table, which is shown in figure 1. Although smaller than the more familiar periodic table of the elements, the structural patterns are more complicated. The grouping of the particles in figure 1 reflects the particles’ properties and the interactions they participate in. These properties and interactions are all well described by the so-called Standard Model of elementary particles SM , which will be discussed further in the following subsection.
To summarize our past progress, accelerator experiments have already revolutionized our understanding of the elementary building blocks and interactions in the Universe around us. They have led to the standard model of elementary particles and have exposed the intrinsic naivete of any pre-accelerator picture of the Universe that had the proton, neutron, electron and photons as the sole elementary particles. However, they have also highlighted our continuing naivete. Further collider experiments will be essential to achieve a more satisfying level of understanding, as will now be explained.
### II.2 Heady Physics Goals for Future Colliders
#### The Standard Model is only a Stop-Gap Theory
Even though the crowning achievement of the past 50 years of HEP has been the construction of the Standard Model, our intellectual goals for its future are very much more wide-reaching than simply filling out the “periodic table” of figure 1 and further detailing its properties. For, despite its predictive power for existing HEP experiments, we know the Standard Model to be no more than a stop-gap theory with a limited domain of applicability. It is phenomenological rather than fully predictive, incorporating 19 free parameters that need to be determined by experiment. Even more damning, it becomes logically inconsistent when we try to extrapolate it to experimentally inaccessible energy scales.
Instead, the quest is for a deeper knowledge of where the Standard Model comes from and also for understanding its connection to the existence and bulk properties of the Universe such as its preponderance of matter over antimatter and its gravitationally-curved 3+1 dimensional space-time structure.
#### Strategies for Advancing beyond the Standard Model
This paper deals largely with the paramount importance of energy frontier colliders as discovery machines to further uncover the secrets of elementary particles and, hence, to learn more about the physical foundations of the Universe. This emphasis on energy reach is further justified in section III.1.
Apart from the approach emphasized here, it is worthwhile to briefly mention some other of the well-developed alternative or complementary strategies for examining these questions at the current and next generations of colliders. More detailed expositions can be found elsewhere in these proceedings Dawson ; Willis ; Parsa .
Very briefly, the origin of all mass in the Universe is hypothesized, in the Standard Model, to be intimately tied in with an as yet undiscovered particle known as the Higgs boson. This is a hypothesized extension to the electroweak theory that was discussed previously – that is to say, our context for understanding photons could further lead to a context for understanding the generation of all mass! As such, it provides a beautiful example of how experimental advances in elementary particle physics might build up a level of understanding of our Universe that would have be otherwise unattainable. The LHC has been optimized to search for this Higgs particle and proposed TeV-scale linear colliders have been optimized for follow-up precision studies of such a particle, if it exists. (Note, however, that both the LHC and $`\mathrm{e}^+\mathrm{e}^{}`$ colliders will anyway be at the energy frontier, so this is merely a shading of the approach emphasized in this paper rather than a true alternative strategy.)
As another thread Willis , colliders have the potential to reproduce and study the extreme energy densities that existed in the first moments after the Universe formed in the hypothesized “Big Bang”. Collisions of heavy ions – at the RHIC collider, for example – will provide the largest volumes under such conditions, even though these volumes are admittedly still miniscule.
Related to this, it has been hypothesized that the observed dominance of matter over antimatter in the observable Universe might have originated from the properties of exotic heavy particles and their anti-particles that could have been routinely produced in the earliest moments of the Universe and whose interactions could have involved large matter-antimatter asymmetries – this is known as CP violation in the obscuring lingo of the field. We may get an experimental handle on such possibilities from experiments at B factory colliders and elsewhere that study the effects of CP violation.
#### The “Theory of Everything”
Ultimately, the “holy grail” of high energy physics is to advance from the Standard Model to the hypothesized “Theory of Everything” that describes and explains the elementary entities, structure and organizing principles of our Universe and is predictive – at least in principle, even if not calculationally – for any experiment we could conceive of that involves elementary particles.
The nature of the Theory of Everything and any possible intermediate levels of understanding towards the Theory of Everything are the subjects for current speculation and future discoveries at accelerators, coupled to theoretical breakthroughs in interpreting these discoveries. Speculation on the elementary particle physics phenomena we might find at future accelerators is helpful in stimulating theoretical progress towards the Theory of Everything and also for the design of the future colliders and their experiments. Several example scenarios for what we could find at many-TeV muon colliders have been presented in these proceedings Berger ; Lane ; Rizzo , along with a very helpful classification scheme for the possibilities Lykken .
The Cambridge theorist and physics popularizer Steven Hawking gives the field a 50% chance Hawking of attaining the Theory of Everything within the next 20 years. This is the most optimistic assessment I am aware of and had many of the theorists at this workshop shaking their heads. At the opposite extreme, it is even logically possible that a unified understanding of the Universe is simply beyond human intellect or, as Steven Weinberg put it, like trying to explain Newtonian mechanics to your dog!
In any case, we know it will not be easy to reach a complete understanding of the physical foundations for our Universe, and collider experiments at the energy frontier will presumably be our main experimental tool in this heady quest.
## III A Wish-list for the Parameters of Future Colliders
This section begins by stressing the paramount importance of energy reach in determining the physics potential of future colliders. (It should be acknowledged that the viewpoint expressed here has certainly been influenced by other presentations at the HEMC’99 workshop and contributions to these proceedings, particularly that of Samios Samios and the summary presentation by Willis Willis .)
The auxiliary requirements for, and benefits of, adequate luminosity and utilizing a variety of colliding projectiles are then discussed in subsections III.2 and III.3. Beam polarization and clean event reconstruction are other relevant experimental capabilities whose discussion is left to elsewhere in these proceedings, in references Heusch and Kahn ; Rehak , respectively.
### III.1 The Paramount Importance of Energy Reach
#### Energy Reach versus Less Direct Experimental Strategies
Some specific experimental strategies for extending our knowledge beyond the Standard Model were discussed in the preceding section. However, the only way we can directly examine an energy scale is by cranking up the energy of our colliders to reach that energy scale. This will then allow us to observe any exotic particles or even more complicated entities Lykken that might exist at that energy scale, whether or not they had been previously forecast on theoretical grounds. Hence, a direct frontal assault on the collider energy frontier is intrinsically more powerful and more likely to result in major break-through discoveries than are alternative, more indirect experimental approaches.
#### Energy Frontier Colliders Can Also Do Lower Energy Physics
In weighing the balance for future frontier machines versus lower energy colliders it should be borne in mind that, besides their primary mission of discovery, frontier machines can also do well at studying lower energy processes – often even better than at dedicated lower-energy facilities.
As an example, the LHC, with $`E_{CoM}=14`$ TeV, will be one of the best places to do studies with B ($`M5GeV/c^2`$) and charm ($`M1.7\mathrm{GeV}/\mathrm{c}^2`$) particles. Even lepton colliders have the general property that lower mass particles are produced in higher order processes and in the decays of heavier particles.
A currently relevant example with lepton colliders is given by collider parameter sets for the 10 TeV and 100 TeV muon colliders that were studied at this workshop hemc99specs . The specified luminosities would correspond to the production of more than $`10^7`$ Standard Model Higgs particles if these existed at the 100 GeV mass scale. This would be orders of magnitude more events than at any of the lower energy electron or muon colliders that have been proposed with the principal goal of studying such a Higgs. (Admittedly, less precise event reconstruction may somewhat dilute the statistical advantage.) Therefore, at least some aspects of any Higgs particle, such as rare decay modes, might be better addressed at frontier colliders than at dedicated Higgs machines operating at the few hundred GeV energy scale.
Besides examples using future colliders, the case can also be made in a historical context, as now follows.
#### An Alternative History: The Standard Model could have been Reconstructed from Today’s Energy Frontier Experiments Alone
We now consider a historical “what if” question that highlights both the paramount importance of energy reach and the ability of energy frontier colliders to perform analyses concerning lower energy scales.
Consider the state of elementary particle physics a half century ago, in 1950. The positron (1933), muon (1937) and pion (1947) had been discovered in cosmic ray experiments, following up on the discovery of the neutron (1932) and the inferred existence of the neutrino from beta decay spectra (1932-3). The historical gedanken experiment is to imagine that, instead of the newly commissioned 184-in synchrocyclotron at Berkeley, which could produce pions, the HEP community of 1950 had been immediately gifted with today’s energy-frontier hadron and lepton colliders – the 1.8 TeV Tevatron proton-proton collider and the 90–200 GeV LEP electron-positron collider – along with the technology for their modern-day general purpose collider experiments.
We can then ask the following question: how much of today’s current understanding of elementary particles (i.e. the Standard Model) would have been promptly reconstructed from the data and what, if anything, would have been missed ?
It can be argued that the basic structure of the Standard Model would have been quickly recovered – either in its entirety or nearly so – since the Tevatron and LEP see evidence for all of the particles in table 1 (redundantly, in most cases) and provide measurements of their interactions and couplings.
In more detail, the copious production of W’s and Z’s would quickly arrive at the electroweak theory that was mentioned previously in section II.1. Knowledge of the strong interaction and of the point-like quarks and gluons it acts on would also come easily, from observations of the “jettiness” of hadronic events at both colliders and from other evidence, and these event signatures would also show immediately that the Tevatron’s proton projectiles were composed of these quarks and gluons.
Probably the last piece of the Standard Model structure to be experimentally established in this scenario would be the complex phase in the CKM quark mixing matrix that accounts for CP violation. The energy frontier collider experiments are poorly optimized for observing the small effects of CP violation in kaon decays where this phenomenon was experimentally discovered, and the alternative of experimental evidence for CP violation in B decays still has only marginal statistical significance at both LEP CPviolOPAL and the Tevatron CPviolCDF .
Even though CP violation is the part of the Standard Model least suited for study at the Tevatron and LEP, their data would certainly still provide the CKM matrix as a theoretical construct and would show it to be non-diagonal. From there, theoretical conjecture on a possible complex phase would be natural and this could well lead to a re-optimized detector (e.g. similar to the B-TeV detector that has been proposed for the Tevatron) that could follow up with more definitive measurements of CP violation to complete the picture of the Standard Model.
To summarize, the outcome of the above gedanken experiment reinforces the previous conclusions of this section by demonstrating that today’s energy frontier colliders can quickly provide access to all of the elementary particle physics structure that we are aware of from our 50 years of historical progress.
### III.2 Desirable Luminosities and their Scaling with Energy
The luminosity of future high energy colliders is the machine parameter that is second in importance only to energy reach.
A rule of thumb for hadron colliders that came into prominence in the 1980’s is that the physics gain from a factor of 10 in a hadron collider’s luminosity corresponds roughly to factor of 2 in energy reach for hadron colliders. The possibilities for such a trade-off are presumably more limited for the point-like projectiles of lepton colliders, where $`\mathrm{E}_{\mathrm{CoM}}`$ gives a more precise measure of the discovery energy reach.
Probably the best way to define the luminosity goals for energy frontier colliders is that the luminosity should be sufficient to gather good statistical samples for the study of any elementary particles existing at the energy scale, $`E\mathrm{E}_{\mathrm{CoM}}`$.
This definition raises a conundrum for discovery colliders at the energy frontier: the number of events is given by the product of the production cross section for the particle, $`\sigma `$, and the time integral of the luminosity, $``$:
$$\mathrm{no}.\mathrm{events}=\sigma \mathrm{dt};$$
(1)
yet how can we predict the cross sections for unknown particles ?
Fortunately, it is common knowledge that very approximate upper limits for production cross sections as a function of collider energy can be guessed at just from general considerations of relativistic quantum mechanics. We now give a version of the type of hand-waving argument that makes this connection. This argument works for the point-like projectiles of lepton colliders at any chosen $`\mathrm{E}_{\mathrm{CoM}}`$. The very approximate luminosity specifications that result could arguably be extended to hadron colliders by replacing $`\mathrm{E}_{\mathrm{CoM}}`$ with the equivalent energy reach, $`E_{reach}^{pp}\mathrm{E}_{\mathrm{CoM}}/6`$, for the collisions of the quark and gluon sub-components of protons. (See section IV for further discussion on the parameter $`E_{reach}^{pp}`$.)
The venerated Heisenberg uncertainty principle of non-relativistic quantum mechanics,
$$\mathrm{\Delta }p\mathrm{\Delta }x\frac{\mathrm{}}{2},$$
(2)
can, for elementary particles, be recast into the very approximate relativistic form
$$\mathrm{\Delta }E\mathrm{\Delta }x\mathrm{}c,$$
(3)
where $`\mathrm{\Delta }p`$ is the momentum spread of a particle’s wave-function, $`\mathrm{\Delta }x`$ is the position spread, $`\mathrm{}`$ is the reduced Plank’s constant, $`\mathrm{\Delta }E`$ can be considered as the energy scale of an interaction and $`\mathrm{\Delta }x`$ gives the corresponding spatial extent, and the conversion from equation 2 to equation 3 uses the approximate ultra-relativistic relation $`Epc`$ that neglects the incoming particle masses. The cross sectional area over which the interaction can occur will be of order the square of the spatial extent of the interaction, roughly $`(\mathrm{\Delta }x)^2`$, so the maximum cross section for a given center-of-mass energy will be roughly:
$$\sigma _{max}(\mathrm{\Delta }x)^2\left(\frac{\mathrm{}c}{E_{CoM}}\right)^2,$$
(4)
or, numerically,
$$\sigma _{max}[pbarn]\frac{400}{(E_{CoM}[TeV])^2},$$
(5)
where units are given in square brackets in this equation and throughout this paper, and 1 picobarn (pbarn) is $`10^{12}`$ barn or $`10^{36}\mathrm{cm}^2`$.
The crude estimate of equation 5 actually does surprisingly well at predicting the largest cross section at today’s 100-GeV-scale $`\mathrm{e}^+\mathrm{e}^{}`$ colliders: it predicts $`\sigma _{max}50`$ nbarn at the energy of the Z pole, 91.18 GeV, which agrees well with the actual cross section of 38 nbarn.
Apart from the Z resonance, however, most cross sections for point-like interactions have been observed to fall several orders of magnitude below the value of equation 5. This can be explained away, in the hand-waving spirit that the equation was derived, by saying that any coupling suppressions arising from the detailed physics process will generally reduce the probability of an interaction occurring for even the closest encounters.
As an acknowledgment that large coupling suppressions are the norm rather than the exception, the luminosities of lepton colliders are commonly bench-marked to a process other than resonant Z production, namely, to lepton-antilepton annihilations to fermions through photon exchange, e.g.,
$$\mathrm{e}^+\mathrm{e}^{}\stackrel{\gamma }{}\mu ^+\mu ^{}.$$
(6)
This has a cross section of:
$$\sigma _R[pbarn]=\frac{0.087}{(E_{CoM}[TeV])^2}.$$
(7)
(To be precise, this only gives accurate predictions for the cross section at energies well below the Z resonance, at 91 GeV. At energies above this, the cross section is substantially modified by interference with the corresponding process involving Z exchange. Instead, equation 7 is intended as the definition of a benchmark cross section that can be used at all energies, as we now explain.)
The inverse of the characteristic cross section of equation 7 defines a unit of integrated luminosity known as a “unit of R” such that a collider that collects one unit of R of integrated luminosity will produce, on average, one event that has the cross section of equation 7.
It was the guidance Peskin of SLAC theorist Michael Peskin that the luminosity for this workshop’s straw-man muon collider parameter sets should, if possible, allow an accumulated inverse luminosity of 10 000 units of R. To convert this to an average luminosity it can be noted that obtaining this integrated luminosity over $`5\times 10^7`$ seconds of running (five “Snowmass accelerator years”) requires an average luminosity of:
$$^{\mathrm{desired}}[\mathrm{cm}^2.\mathrm{s}^1]=2.3\times 10^{33}\times \left(\mathrm{E}_{\mathrm{CoM}}[\mathrm{TeV}]\right)^2.$$
(8)
This can be contrasted with the much more modest luminosity that would be needed to acquire 10 000 events produced at the approximate maximum cross section specified by equation 5:
$$^{\mathrm{borderline}}[\mathrm{cm}^2.\mathrm{s}^1]=5\times 10^{29}\times \left(\mathrm{E}_{\mathrm{CoM}}[\mathrm{TeV}]\right)^2.$$
(9)
The straw-man parameters for both the 10 TeV and “100 TeV ultra-cold beam” examples met or even exceeded Peskin’s request, each with 8700 units of R per detector in a single year. On the other hand, the “100 TeV evolutionary extrapolation” parameter set specified only 87 units of R per detector per year. This reflects the escalating luminosity demands with $`\mathrm{E}_{\mathrm{CoM}}`$ due to the $`1/\mathrm{E}_{\mathrm{CoM}}^{}{}_{}{}^{2}`$ cross section scaling of equations 5 and 7.
### III.3 The Complementarity of Different Projectile Types
The different experimental conditions and, particularly, the different interacting projectiles of hadron and lepton colliders will generally lead to different sensitivities for specific processes at the energy scale under consideration, so the two types of colliders are also complementary to a certain extent and there are advantages to operating both types of machines. This complementarity also applies to the two types of lepton colliders – $`\mathrm{e}^+\mathrm{e}^{}`$ and $`\mu ^+\mu ^{}`$ colliders – but to a lesser extent.
There are also many other possibilities for the colliding projectiles that will not be discussed further in this paper: gamma-gamma collisions, heavy ion colliders, like-sign lepton colliders Heusch ($`\mathrm{e}^{}\mathrm{e}^{}`$ and $`\mu ^{}\mu ^{}`$) and any one of the several options that collide dissimilar projectiles. These options all have some potential for complementary physics studies and should be looked at further. However, it should be noted that several of them are understood to be less suitable for exploring the energy frontier for various reasons, a few of which are discussed elsewhere in these proceedings Telnovprojectiles .
In the past, energy frontier hadron colliders have been regarded more as discovery machines while lepton colliders, following later but with cleaner experimental conditions, have been considered mainly as follow-up machines for precision studies. The following section reviews the history of collider facilities that led to this assignment, as well as introducing a speculated scenario for future colliders.
## IV The Livingston Plot for Progress in the Energy Reach of Colliders – Past, Present and Future
### IV.1 Presentation and Interpretation of the Plot
Figure 2 is the famous Livingston plot showing the historical exponential growth with time in the energy reach of both lepton and hadron colliders. The data for past and present lepton and hadron colliders has been taken from reference NLC and is discussed and parameterized in the following subsection.
The logarithmic energy scale in figure 2 is physically appropriate under the reasonable assumption that the underlying physical importance of the mass spectrum will lie in the ratios of particle masses as opposed to mass differences. As some confirmation of this assumption, the masses of the known elementary particles do indeed fall relatively evenly along a log-energy scale rather than being bunched at the low energy end. To rephrase this in a way that might sound depressing to accelerator builders, the past exponential progress in the energy reach of colliders can be considered to have corresponded to merely a steady (linear) rate of advance in their physics capabilities since the logarithm of the energy rather than the energy itself is the appropriate metric for assessing the discovery reach of colliders.
Some speculated future colliders beyond the LHC have been added in to figure 2. In sum, they are intended to comprise a straw-man proof-of-plausibility scenario to show that a sufficiently motivated and adequately funded HEP community may be able to continue constructing accelerators that lie on or near the lepton and hadron Livingston curves and that extend up to the PeV constituent energy scale (where 1 PeV = 1000 TeV). Discussion on this scenario occupies the final subsection in this section, subsection IV.3, as well as much of the remainder of this paper.
### IV.2 Parameterizations for the Historical Progress in the Energy Reach of Hadron and Lepton Colliders
The constituent energy reach for lepton colliders in figure 2 has been defined simply to be their center-of-mass energy,
$$E_{reach}^{lepton}E_{CoM}.$$
(10)
The fact that protons are composite particles rather than point-like elementary entities dilutes the constituent energy reach of hadron colliders relative to lepton colliders by a factor that depends on both the physics process and the collider luminosity. The choice of dilution factor for past hadron machines was copied from reference NLC and we have chosen the similar dilution factor of 6 for future hadron colliders:
$$E_{reach}^{pp}E_{CoM}/6.$$
(11)
For any given hadron collider, other estimates for the dilution factor may differ by a factor of two or more from this choice, in either direction. There is also a slight arbitrariness in some of the other data choices, so the reader is warned that the details in Livingston plots may vary from publication to publication.
The two dashed lines drawn through the data points give parameterizations for the constituent energy reach, $`E_{reach}`$, versus year of first physics, Y, for lepton and hadron colliders. They have equations:
$$\mathrm{log}_{10}(E_{reach}^{lepton}[TeV])=(Y2002)/13$$
(12)
and
$$\mathrm{log}_{10}(E_{reach}^{pp}[TeV])=(Y1994)/13,$$
(13)
respectively and substituting the energy dilution factor from equation 11 into equation 13 gives the required CoM energy reach for future hadron colliders:
$$\mathrm{log}_{10}(E_{CoM}^{pp}[TeV])=(Y1984)/13.$$
(14)
Equation 12 informs us that a decade of energy increase in lepton colliders has historically occurred every 13 years. Proton colliders have advanced at the same rate of progress as lepton colliders – an energy decade every 13 years – but have been about 8 years ahead of lepton colliders in attaining a given constituent energy reach.
As well as past and present colliders, figure 2 plots the planned 2005 completion date for the Large Hadron Collider (the LHC, currently under construction) and a region representing roughly the range of predicted or proposed turn-on dates and energies for contemplated electron-positron colliders. Beyond these, later and more speculative collider points at higher energies have also been added to figure 2. A straw-man “target” region for a Very Large Hadron Collider (VLHC) is shown with an energy range of 200–400 TeV corresponding to first physics dates of 2014-8. Three muon collider points/regions are also shown, extending up to the lepton Livingston curve and with CoM energies up to 1 PeV.
It can be noted that the cancelation of the SSC collider has hampered progress in the energy reach of hadron colliders, as is also stressed elsewhere in these proceedings Samios . The SSC would have been above the Livingston curve for hadrons if it had already been built at its design energy of $`E_{CoM}=40`$ TeV. Instead, it can be seen from figure 2 that the LHC has already fallen below the Livingston curve; to have stayed on the curve, equations 11 and 13 specify that it would have needed to have either turned on last year (1999) or, for its planned 2005 turn-on, had an $`E_{CoM}`$ of 42 TeV rather than 14 TeV.
### IV.3 Introducing the Proof-of-Plausibility Scenario for Future Colliders
Table 1 summarizes the collider type, energy reach and year of first operation for the colliders in the proof-of-plausibility scenario to extend the Livingston plot beyond the LHC.
The scenario is economical in requiring only 5 colliders to reach all the way up to the 1 PeV constituent energy scale – one each of $`\mathrm{e}^+\mathrm{e}^{}`$ and hadron colliders and three $`\mu ^+\mu ^{}`$ colliders. The correspondingly large leaps in energy continue the necessary trend that was first set by the SSC, with $`\mathrm{E}_{\mathrm{CoM}}=40`$ TeV, which improves by more than a factor of 20 from the existing energy frontier at $`\mathrm{E}_{\mathrm{CoM}}=1.8`$ TeV. (The SSC ended up being too expensive, but this should not be interpreted as a fundamental flaw in the concept.) The increasingly large energy jumps are dictated by the rising cost of each successive machine. They are also desirable to give each new machine a big enough window for a good chance to discover new physical phenomena or, at minimum, to rule out a big enough energy range for this to have theoretical significance.
In order to limit the number of colliders in table 1 to five, the energies of the colliders have all been pushed up as far as was considered practical towards, in each case, a natural and rather sharply defined technological bound. For the first collider in table 1 – a TeV-scale $`\mathrm{e}^+\mathrm{e}^{}`$ collider – the fundamental energy bound comes from the sharp rise with energy in the interactions between TeV-scale electrons and electromagnetic fields, as will be discussed in section V.
Only one extra hadron collider was included in the scenario since the energy scale of the LHC, with $`\mathrm{E}_{\mathrm{CoM}}=14`$ TeV, will already be enough energy to probably warrant jumping in a single step up to the few hundred TeV energy bound that results from synchrotron radiation. The limit and the 200-400 TeV hadron collider will be discussed in section VI.
The 3 muon colliders all butt up against different technological energy limits, as will be further addressed in subsection VII.2. The 4 TeV muon collider is about at the size limit for fitting on a laboratory site and, more importantly, is fighting against the energy-cubed rise in the off-site neutrino radiation hazard for populated locations. The middle muon collider pushes the synchrotron radiation energy limit for circular muon colliders, which is probably in the range of 30-100 TeV. Finally, 1 PeV is a limiting energy scale for any linear muon colliders that can’t call on exotic acceleration schemes to obtain accelerating gradients that greatly exceed today’s technological bounds.
While the scenario looks plausible at first sight, it is important to stress that this is only a first presentation, and the scenario’s feasibility or otherwise should be better established with further studies. Also, no claim is made that the scenario is in any way optimal and it is hoped that follow-up studies and technical progress will lead to refinements and also to alternative scenarios. This would give us more options and possibilities to progress with foresight towards a productive future for HEP.
## V Future Electron-Positron Colliders
### V.1 Energy Limitations on $`\mathrm{e}^+\mathrm{e}^{}`$ Colliders
The fundamental energy limitation for electron-positron colliders is the projectile itself. At only 1/1837 of the proton mass, electrons are relativistic enough at TeV energy scales and beyond to be overly sensitive to electromagnetic fields. They become difficult to bend, focus or collide because of their excessive tendency to throw off photons due to synchrotron radiation in the magnetic fields of bending and focusing magnets and then to beamstrahlung in the electromagnetic fields of the oncoming bunch at collision.
Even to advance to the TeV scale, these difficulties have already required the major technology shift from circular colliders to single pass linear colliders. Synchrotron radiation scales as the fourth power of the beam energy (or as the third power, for some parameters), so further technological innovations would clearly be needed to advance to still higher energies. To solve each of the three classes of problems, we would need all three of:
1. acceleration that doesn’t have a prohibitive linear cost coefficient with energy
2. focusing to collision that doesn’t induce excessive synchrotron radiation
3. some way of damping out the beamstrahlung at collision.
Research is underway to attempt solutions to each of these three problems, which have arguably been presented in order of increasing difficulty. Potential solutions include electron drive beams for acceleration, plasma focusing or focusing induced by auxiliary beams, and 4-beam schemes that also use additional beams to partially cancel out the electromagnetic fields at collision.
### V.2 This Scenario: A Single TeV-Scale Electron-Positron Collider
Planning for proposed TeV-scale $`\mathrm{e}^+\mathrm{e}^{}`$ colliders is far enough along that the shaded region for $`\mathrm{e}^+\mathrm{e}^{}`$ in figure 2 simply represents the very approximate range of energies and turn-on dates for proposed machines.
It can be seen that the shaded region extends well below the Livingston curve. This reflects the technological problems associated with energy that were discussed in the preceding subsection and, somewhat related to this, the widespread support for the alternative motive of studying identified physics processes – such as production of top quarks or Higgs particles in the 100-200 GeV mass range, if they exist – rather than a dedicated thrust towards the energy frontier.
The scenario assumes that exactly one TeV-scale $`\mathrm{e}^+\mathrm{e}^{}`$ collider will be constructed before the mantle of frontier energy lepton collider is passed on to muon colliders. Besides its own physics value, this $`\mathrm{e}^+\mathrm{e}^{}`$ collider would continue to establish the technology needed for the later muon colliders and, in particular, for the assumed 1 PeV muon collider that completes the scenario. The technological overlap between the two types of lepton colliders is discussed further in section VIII.1.
## VI Future Hadron Colliders for the Energy Frontier
This section discusses current research towards future hadron colliders and assesses the constraints on their ultimate potential energy reach. As a particular constraint on their energy reach, it is explained why linear hadron colliders will probably never be viable. Finally, the 200-400 TeV pp collider in the proof-of-plausibility scenario is discussed.
### VI.1 Current Research Towards Future Hadron Colliders
Beyond the LHC, research is underway for a follow-up hadron collider, usually referred to as the “Very Large Hadron Collider” (VLHC), whose goal would be to reach substantially higher energies than the LHC without being much more expensive. The U.S. research efforts towards a VLHC are conveniently summarized in the annual report of the U.S. Steering Committee for a VLHC VLHC .
Much of the VLHC research involves magnet design because the bending magnets for the collider storage ring are assumed to be a large, if not dominant, component of the cost for a VLHC. (For comparison, dipole magnet costs for the collider ring of the SSC were budgeted to comprise roughly 25% of the total cost.) The current emphasis on magnet R&D is largely divided between the design of low field (2 Tesla) superferric magnets and very high field (greater than 9 Tesla) magnets.
The low field superferric magnet designs might be very cheap to construct. As one of the challenges for this option, there is concern that the attainable luminosities for the low field superferric option may be limited due to beam instabilities caused by the combined effects of the small magnet apertures, the large collider circumference and the lack of synchrotron radiation damping. The potential lack of stability and tune-ability for such simple magnets is also a concern.
A major motivation given for very high field magnets, as opposed to intermediate field magnets, is their potential to cause a beneficial level of synchrotron radiation damping of the beam at an assumed beam energy of 50 TeV. The radiation damping is higher for high field magnets because the fractional energy loss per turn, $`\frac{\mathrm{\Delta }\mathrm{E}}{\mathrm{E}_{\mathrm{CoM}}}`$ due to synchrotron radiation scales with $`\mathrm{E}_{\mathrm{CoM}}`$ and with the average bending magnetic field, B, according to:
$$\frac{\mathrm{\Delta }\mathrm{E}}{\mathrm{E}_{\mathrm{CoM}}}\mathrm{E}_{\mathrm{CoM}}^{}{}_{}{}^{2}\times \mathrm{B}.$$
(15)
Very high field magnets cannot use the niobium-titanium superconductor that has been used in all collider magnets to date since this conductor has an impractically low critical current at magnetic fields above 9 Tesla. Other superconductor materials must be used, such as niobium-tin or the new high-$`T_C`$ superconductors, and these are – at least at present – considerably more expensive than niobium-titanium and have inferior mechanical properties.
Regardless of progress in superconducting materials, the mechanical stresses in magnets scale as the square of the magnetic field and will always conspire to raise the cost per unit length of high field magnets relative to those at lower fields. On the other hand, a more relevant yardstick than cost per unit length is the cost per Tesla-meter since a collider ring at a given energy will need a fixed number of Tesla-meters to bend the beams in a circle. This favors higher field magnets so long as the cost per meter increases less than proportionally to the field strength. Also, higher fields should reap further cost savings from the consequent reduction in tunnel length. Because of these trade-offs, the field strength for superconducting magnets that would give the optimal cost for a future hadron collider is not at all well established. (However, see the discussion in subsection VI.3 regarding a study by Willen Willen that uses today’s superconducting magnet technology.)
More generally, it appears that the cost and technology optimizations used for the VLHC could usefully be extended to include varying the energy of future hadron colliders away from the VLHC’s assumed $`\mathrm{E}_{\mathrm{CoM}}=100`$ TeV, as will be addressed in subsection VI.3.
### VI.2 The Ultimate Energy Reach for Hadron Colliders
#### Limits from Synchrotron Radiation
For the VLHC studies at $`\mathrm{E}_{\mathrm{CoM}}=100`$ TeV, the beneficial damping effects of synchrotron radiation must already be balanced against the problems it causes. Given the strong power-law rise in synchrotron radiation with energy, it can be surmised that synchrotron radiation should lead to a fairly sharp technical cut-off in the viability of hadron colliders by the few hundred TeV range. More quantitatively, the power radiated due to synchrotron radiation, $`\mathrm{P}_{\mathrm{synch}}`$, is given approximately by:
$$\mathrm{P}_{\mathrm{synch}}[\mathrm{MW}]0.6\times \mathrm{I}[\mathrm{A}]\times \mathrm{B}[\mathrm{T}]\times \left(\mathrm{E}_{\mathrm{CoM}}[100\mathrm{TeV}]\right)^3,$$
(16)
where I is the average current in each proton beam and B is the bending magnetic field which, for this equation, is simplistically assumed to be constant around a circular collider ring.
#### Constraints from the Experimental Environment
Probably the biggest technical challenges for the SSC and LHC came, not from the colliders themselves, but from the extreme operating conditions anticipated for the collider detectors. The experiments at future energy frontier hadron colliders will have even worse problems coping with luminosities that, as was shown in section III.2, should ideally rise as the square of $`\mathrm{E}_{\mathrm{CoM}}`$. The problems arise from the large cross section for soft background interactions:
$$\sigma _{TOT}^{pp}200\mathrm{millibarns},$$
(17)
rising only slightly with $`\mathrm{E}_{\mathrm{CoM}}`$. The average number of background events per bunch crossing, $`n_b`$, is given by,
$$n_b=\frac{\times \sigma _{TOT}^{pp}}{f},$$
(18)
with $`f`$ the bunch-crossing frequency, or, numerically,
$$n_b2\times \frac{[10^{34}\mathrm{cm}^2.\mathrm{s}^1]}{f[\mathrm{GHz}]}.$$
(19)
Desirable luminosities in the range $`10^{3536}\mathrm{cm}^2.\mathrm{s}^1`$ will require major advances in detector technology and event analysis to resolve the background event pile-up predicted from equation 19 and also to cope with event-induced radiation damage to central detectors. Particular attention will need to be paid to the radiation hardness of the central tracker and electromagnetic calorimeter, and to fast timing, triggering and read-out of the events.
Equation 19 shows that the bunch crossing frequency, $`f`$, should be made as large as is practical in order to minimize the event pile-up. Ideally, the time between crossings, $`1/f`$, should be comparable to the resolving time of the detector, which tends to be limited to on the order of nanoseconds. (The LHC is designed for one bunch crossing every 25 nanoseconds.) However, in practice this turns out to be an inefficient way to produce luminosity, requiring large stored beam currents that exacerbate the synchrotron radiation problem of equation 16 and bring on other technical headaches.
#### The Implausibility of Single Pass Hadron Colliders
An extreme example of the inefficiency of frequent bunch crossings would be provided by any attempted design for a single pass hadron collider that would use the linear accelerator technology developed for $`\mathrm{e}^+\mathrm{e}^{}`$ colliders. The magnitude of the problem appears to prohibit any serious speculation on using linear hadron colliders to extend the energy frontier, as we now show.
To obtain a crude scaling argument, we note that the luminosity as a function of the bunch crossing repetition rate, $`f`$, the number of particles per bunch, $`\mathrm{N}_\mathrm{b}`$, and the transverse beam dimensions, $`\sigma _x`$ and $`\sigma _y`$, is given roughly by:
$$\frac{f\mathrm{N}_\mathrm{b}^2}{\sigma _\mathrm{x}\sigma _\mathrm{y}}\frac{1}{f}\frac{(\mathrm{P}_{\mathrm{beam}})^2}{\sigma _\mathrm{x}\sigma _\mathrm{y}},$$
(20)
where we haven’t bothered to keep track of numerical factors depending on the precise definitions of $`\sigma _x`$ and $`\sigma _y`$ (several different conventions are in common use!) and have ignored effects of order unity such as the pinch enhancement of luminosity, and the second expression follows from the first because the average beam power, $`\mathrm{P}_{\mathrm{beam}}`$, is proportional to the average beam current, $`\mathrm{P}_{\mathrm{beam}}fN_b`$. Equation 20 shows that the luminosity at linear colliders falls off inversely with the repetition rate – at least to the extent that the pinch enhancement can be neglected and, as a more substantial caveat, provided that the beam power and transverse beam dimensions are fixed. Therefore, very low repetition rates are strongly favored and the nanosecond-scale repetition rates desired for hadron colliders are strongly disfavored.
To set the numerical scale, the first very speculative straw-man parameter set for a 1 PeV linear muon collider Zimmermann assumes a luminosity of $`=5.4\times 10^{35}\mathrm{cm}^2.\mathrm{s}^1`$ and a repetition rate of only $`f=3.2`$ Hz. (This corresponds to a very impressive per-collision integrated luminosity of 170 inverse nanobarns per collision!) If this parameter set was translated to a 1 PeV hadron collider rather than a muon collider then equation 19 predicts a manifestly unmanageable $`3\times 10^{10}`$ interactions per collision!
Even if one allows for a re-optimization of parameters from this rather extreme example, it is hard to imagine a single-pass hadron-hadron collider with both an interesting luminosity and viable experimental conditions. The conclusion of this subsection is then that the ultimate center of mass energy for hadron colliders will almost certainly be attained using circular hadron colliders and will probably be limited to a few hundred TeV.
### VI.3 A 200-400 TeV Hadron Collider for the Proof-of-Plausibility Scenario
#### Basic Specifications
The preceding subsection established that the collision energy often assumed in VLHC studies, $`\mathrm{E}_{\mathrm{CoM}}=100`$ TeV, is rather close to the ultimate energy scale possible for future hadron colliders, at – we will guess – perhaps 200 to 400 TeV. It is notable that the energy jump from the LHC ($`\mathrm{E}_{\mathrm{CoM}}=14`$ TeV) to this energy would be about the same factor of twenty-or-so as was planned to go from the Tevatron (1.8 TeV) to the SSC (40 TeV). Both rate-of-progress and economy therefore suggest that it makes eminent sense to try to reach this frontier energy in a single step. This motivates the choice for our proof-of-plausibility scenario of a single hadron collider after the LHC, at $`\mathrm{E}_{\mathrm{CoM}}=`$200-400 TeV. (The possibility has not been excluded for an upgrade from the lower end to the higher end of this energy range.)
For definiteness in the overall scenario, we can also assume a 400 km circumference for the 200-400 TeV hadron collider ring. As will be seen later, this fits in with the size scale for a new world HEP laboratory that also includes a muon collider.
#### Magnets and Synchtrotron Radiation Damping
A 400 km circumference corresponds to an average bending magnetic field around the collider ring of 5.3 (10.5) tesla for the assumed energy range of $`\mathrm{E}_{\mathrm{CoM}}=200`$ (400) TeV.
Because the 200-400 TeV energy range is higher than the $`\mathrm{E}_{\mathrm{CoM}}=100`$ TeV value usually considered for the VLHC, it is notable that it might give a better match between, on the one hand, the desirable level of synchrotron radiation damping sought by those designing high field VLHC magnets and, on the other hand, the lower magnetic field strengths that might correspond to a cost minimum. In fact, as a very intriguing possibility that is more general than this scenario, it is not ruled out that the global cost optimum for optimal synchrotron damping might be at lower magnetic fields but higher collision energies than are currently considered for the high field 100 TeV VLHC, i.e. the extra energy reach could conceivably come for free! The synchrotron radiation and cost aspects of the 200-400 TeV hadron collider will now be discussed in turn.
The levels of synchrotron radiation damping in this scenario are easily seen to range from just slightly above, to far beyond, those encountered in very high field magnet studies at a 100 TeV VLHC. From equation 15, the 5.3 T average field for the 200 TeV scenario gives the same damping as an average field 4 times larger at $`\mathrm{E}_{\mathrm{CoM}}=100`$ TeV, i.e. 21 tesla, which is slighty above the maximum considered for VLHC studies. In contrast, an unrealistic factor of 16 in magnetic field strength at 100 TeV would be needed to compensate for a four-fold increase in energy, to 400 TeV, so the level of synchrotron radiation damping in this scenario is obviously much larger. It is a subject for further studies to determine whether the level of synchrotron radiation at 400 TeV is desirable in, or even compatible with, any self-consistent set of hadron collider parameters that would presumably utilize lower beam currents, smaller spot sizes and a stronger final focus than any current VLHC parameters.
#### Cost Considerations
The basis for cautious optimism on the magnet costs for this scenario comes from a careful study, by superconducting magnet expert Erich Willen Willen , of the cost optimum in field strength that would be obtained by assuming today’s superconducting magnet technology.
Willen’s cost evaluation is for a $`\mathrm{E}_{\mathrm{CoM}}=200`$ TeV hadron collider. (As a cautionary note, he actually refers to his parameter sets as being for 100 TeV colliders, but close inspection reveals this to be the energy per beam rather than $`E_{CoM}`$.) The costing for the dipole magnets is a careful scaling of the costs for the dipole magnets used in the existing RHIC collider. The scaling takes into account some suggested design modifications to increase the magnetic field and reoptimize the magnet length, aperture and superconducting coil layout, all of which are compatible with currently available technology. The cost vs. magnetic field strength characteristic was found to have a rather broad minimum reaching down to 1436 1993 $ U.S. per tesla-meter at a field strength of 5.7 T. In table 7 of reference Willen , Willen presents an estimated cost of 6 $B for the dipole magnets in a 200 TeV collider using two rings of 5.7 T dipole magnets with 80% packing, for an average bending field of 4.6 T and a circumference of 460 km. Willen also priced the tunnelling costs for the collider ring at a little over 0.4 $B , costed at $900/m after studies at the 1996 Snowmass Workshop, i.e. a much smaller component of the total colider cost.
To use Willen’s study as a benchmark for the 200-400 TeV collider considered here, Willen’s cost minimum at a 4.6 T average field is rather close to the required 5.3 T field for the 200 TeV collider with a 400 km circumference, so his 6 $ B cost estimate for today’s magnet technology is directly applicable. Improving technology in high field superconducting magnets can then be expected to both lower the cost per tesla-meter and raise the field strength of the cost minimum, i.e., move the magnetic field strength some distance in the direction of the 10.5 T average bending field that has been assumed for the 400 TeV energy.
As a more quantitative statement of the progess that might be demanded in order for 200 TeV or 400 TeV hadron colliders to become economical, it would be helpful reduce the dipole magnet cost to about 3 $B. This would require a factor of about 2 or 4 reduction, respectively, in magnet costs-per-Tesla-meter from Willen’s estimate of $1436 /Tesla-meter. It is not unrealistic to hope that such savings could come from economies of scale and from a decade of technological advances in magnet components, design and manufacture that builds on the current magnet R&D program for the VLHC.
## VII Muon Colliders
### VII.1 Circular and then Linear Muon Colliders to 1 PeV and Beyond ?
#### Switching from Electrons to Muons
All of the problems with TeV-scale $`\mathrm{e}^+\mathrm{e}^{}`$ colliders that were discussed in section V are associated with the relative smallness of the electron mass. The proposed technology of muon colliders aims to solve, or at least greatly reduce, these problems by instead colliding muons, which are leptons that are 207 times heavier than electrons.
Replacing the mass-related problems of $`\mathrm{e}^+\mathrm{e}^{}`$ colliders, the main problems at muon colliders arise because muons are unstable particles, with an average lifetime of approximately 2.2 microseconds in their rest frame. The preparation, acceleration and collision of the muon beams must all be done quickly and the supply of muons must be replenished often. The products of the muon decays also cause problems: the decay electrons deposit energy all along the path of the muon beams and create backgrounds in the detectors and, more surprisingly, the neutrinos can cause a radiation hazard in the surroundings of the collider ring nurad ; hemc99nurad .
The technology and status of R&D on muon colliders has been covered in detail in reference status and the specific issues involving many-TeV muon colliders were examined at this workshop and form the topic of many of the papers in these proceedings. A focal point for the studies at this workshop was provided by three self-consistent parameter sets for muon collider rings, one set at 10 TeV and two sets at 100 TeV. Reference hemc99specs of these proceedings discusses the parameter sets and their evaluation at the workshop. A very significant development at the workshop was the presentation Zimmermann of parameter sets for linear muon colliders at energies ranging from 3 TeV all the way up to 1 PeV. The general assessments on the energy reach for both circular and linear muon colliders will now be briefly reviewed.
#### Synchrotron Radiation Limits for Circular Muon Colliders
The potential energy reach for circular muon colliders appears to hit a fairly hard limit at about $`\mathrm{E}_{\mathrm{CoM}}=100`$ TeV, where the radiated power from synchrotron radiation has risen to become approximately equal to the beam power. Additional constraints arise from beam heating due to the quantum fluctuations in synchrotron radiation, as was pointed out in the workshop and is discussed elsewhere in these proceedings Telnovsynch . Like the beam power, the quantum fluctuations also rise as a relatively high power of the beam energy (both rising as energy cubed, for some benchmark parameters – c.f. equation 16 for hadron colliders) – and this further pins the ultimate potential for circular muon colliders down to the 100 TeV energy scale. See reference hemc99specs for further discussion on the limits imposed by synchrotron radiation.
#### Single Pass Muon Colliders
Following the historical path of $`\mathrm{e}^+\mathrm{e}^{}`$ colliders, muon collider energies above 100 TeV can be contemplated by switching to the technology of linear colliders, as was shown by Zimmermann Zimmermann . All of Zimmermann’s parameter sets – at 3, 10, 100 and 1000 TeV – require specified “exotic” technologies both for preparing the muon beams and for acceleration, although these are not implausible at first reading by this author, who is admittedly not particularly knowledgable about linear colliders; see reference hemc99specs for further discussion. (Zimmermann’s provocative parameter sets clearly need further review by people more expert than this author.)
It is a remarkable feature of the progression in parameters that the final parameter set, at $`\mathrm{E}_{\mathrm{CoM}}=1`$ PeV, appears to be not so much more “exotic” or less technologically plausible than the initial parameters at 3 TeV. This justifies the inclusion of a PeV-scale linear muon collider as the last collider in the proof-of-plausibility scenario, where it can benefit from about three decades of R&D to develop and refine the necessary technologies. Conversely, circular muon colliders have been favored over linear colliders up to the 100 TeV scale, where their technologies seem better established.
Continuation of linear muon colliders to even beyond the PeV scale has not yet been rigorously excluded, although the technological challenges would obviously be formidable. For example, the final focus design is far removed from anything we can seriously contemplate today. Also, the 2 linacs for even a 1 PeV collider are each already 500 km long for an assumed accelerating gradient of 1 GV/m, giving a maximum depth of 5 km below a spherical Earth. (See reference hemc99nurad for an illustration of the geometry.)
A 10 PeV collider would need either 10 times the tunnel length or 10 times the accelerating gradient of the 1 PeV example, or some compromise between these parameters. However, the maximum depth below the Earth’s surface goes as the square of tunnel length, so trying to lengthen the tunnel quickly gets one into hot lava! The barriers against increasing the gradient to 10 GV/m have more loopholes. Gradients of 1 GV/m are already pushing the material limits for any known surfaces in the accelerating structures, even at the highest frequencies people consider, but the solution to this might plausibly come from exotic acceleration schemes using plasmas driven by lasers. Various experiments to test such acceleration schemes are already underway.
### VII.2 Muon Colliders for the Proof-of-Plausibility Scenario
This subsection provides discussion that is specific to the 3 muon colliders of table 1, at $`\mathrm{E}_{\mathrm{CoM}}=4`$ TeV, 30-100 TeV and 1 PeV. In order to spell out the entire progression, however, we begin by discussing a neutrino factory – a simpler accelerator that is not actually a collider but which would likely serve as an important staging point towards the construction of the 4 TeV muon collider that is the first $`\mu ^+\mu ^{}`$ collider entry in table 1.
#### Leading-in with a Neutrino Factory Muon Storage Ring
Because no muon collider has yet been built, a convincing demonstration of muon collider technology would be prudent before investing in an energy frontier collider. A neutrino factory would afford this demonstration while providing much useful and complementary physics to the colliders.
A neutrino factory is a simpler, non-colliding muon storage ring that would be optimized for the collimated decay of muons into one or more intense neutrino beams aimed at neutrino physics experiments. As an aside on the choice of beam energy for the neutrino factory, current studies are concentrating mainly on neutrino factories with beam energies of 20 to 50 GeV that are optimized for a particular range of neutrino oscillation studies. However, neutrino factories at higher beam energies of perhaps 100–200 GeV are more optimal for the complementary high-rate experiments that study neutrino interactions and also for some other neutrino oscillation scenarios. This option for neutrino factories at higher energies might also be further considered since it would better test the acceleration technology needed for high energy muon colliders and should also be easier to upgrade to such a collider.
In the straw-man scenario presented here, the design of the neutrino factory demonstration machine would be made compatible with an upgrade to a site-filling 4 TeV muon collider, to occur immediately on completion if not beforehand. The neutrino factory provides an ideal intermediate step in the construction of a 4 TeV collider because it would be a valuable HEP facility in its own right if, for some reason, the continued upgrade to a collider was found not to be technologically feasible. It can therefore be built before a final decision has been made on the collider technology, and its construction can better inform that decision.
As another nice feature of this staging scenario, the experimental investments and the incremental physics from neutrinos is not lost with the upgrade to a collider. Quite the contrary, it turns out nuphysref that much of the high-rate neutrino physics can anyway be performed even better in parasitic running at a multi-TeV collider than in a dedicated lower-energy neutrino factory. The schedule for the proof-of-plausibility scenario would require a decision by 2007-8 on whether or not to proceed to a site-filling muon collider. By this time, the decision could presumably be guided by preliminary results from 14 TeV CoM proton-proton collisions at the LHC.
#### A 4 TeV Muon Collider
Four TeV was the muon collider energy chosen for the design study Snowmass presented at the Snowmass’96 workshop. However, it has since become more widely appreciated that neutrino radiation from the collider ring is a concern at these energies nurad , so the muon beam current would likely be limited to more than an order of magnitude below the Snowmass’96 parameters. The reduced current would anyway save on the cost and power of the proton driver, which might have been considerable for the Snowmass’96 specifications.
Reference epac98 contains a self-consistent parameter set for a low-current 4 TeV muon collider that is appropriate for the proof-of-plausibility scenario given here. The parameter set has a luminosity of $`=6\times 10^{33}\mathrm{cm}^2.\mathrm{s}^1`$ and an average neutrino radiation dose in the plane of the collider that is below one-thousandth of the 1 mSv/year U.S. federal off-site limit.
#### The Highest Energy Circular Muon Collider, at 30-100 TeV
The next step up in energy is assumed to carry all the way to the highest feasible energy for a circular machine, which we assume to be in the range $`\mathrm{E}_{\mathrm{CoM}}=30`$-100 TeV.
In order to consolidate resources and minimize the overall cost of the scenario, it is sensible that the 30-100 TeV $`\mu ^+\mu ^{}`$ collider should be constructed in the same laboratory whose 400 km boundary is defined by the 100-200 TeV hadron collider of the preceding section. A boundary of such a size will anyway be dictated by the requirements of neutrino radiation, as is discussed in detail elsewhere in these proceedings hemc99nurad . Briefly, the radiation disk that is emitted in the plane of the collider ring would rise to approximately 300 meters above the Earth’s surface for such a boundary and in the approximation of a spherical Earth. This is assured to be far enough above any structures that no practical off-site radiation hazard would remain, as is discussed further in hemc99nurad .
The high end of the energy range, $`\mathrm{E}_{\mathrm{CoM}}=100`$ TeV, corresponds to two of the straw-man parameter sets for this workshop – see the parameter table in reference hemc99specs – while parameters for the low end, 30 TeV, can be estimated by interpolating between the 10 TeV and 100 TeV straw-man parameter sets in reference hemc99specs . As we learned in the workshop Telnovsynch , beam heating effects from synchrotron radiation push the parameters in the direction of a lower average bending field than the 10.5 T assumption of the workshop’s parameter sets. A sensible value might, e.g., be half of this, i.e. a 5.3 Tesla average for a 200 km circumference at 100 TeV, as was indicated in reference hemc99specs . Such effects should be much less important if the maximum collider energy turns out to be limited towards the lower end of the energy range under consideration. A 30 TeV collider would presumably still use as high an average bending field as is practicable in order to maximize its luminosity, e.g., perhaps as high as a 10.5 T average. This would correspond to the much smaller circumference of only 30 km.
As for the hadron collider, bending magnets are likely to be the major cost component of this $`\mu ^+\mu ^{}`$ collider. The collider ring magnets should be much cheaper for a 30-100 TeV muon collider than for the 200-400 TeV hadron collider as they would require an order of magnitude less tesla-meters of total bending: the beam energy is several times lower and there is a further factor-of-two saving because only one ring of magnets is needed instead of two (counter-rotating beams of opposite charges can share the same magnet ring). Indeed, a presentation at this workshop by Mike Harrison Harrison suggested a total magnet cost of only about 400 million dollars for the collider ring magnets in the 10 TeV parameter set. This is very encouraging to the extent that it can be scaled up to higher energy muon colliders (although see Harrison’s caveats regarding such a scaling). Instead, the magnets for the muon acceleration are likely to be a much larger cost component than the collider ring magnets, with perhaps several times the total tesla-meters of bending as well as additional costs associated with the need to transport large momentum spreads. (See reference hemc99specs of these proceedings for a more detailed discussion.) The design of the magnets for both the acceleration and collider rings will also need to cope with the energy deposited from decay electrons and sychrotron radiation, and this should also feed down into more expensive magnets than at hadron colliders. The upshot of the discussion in this paragraph is that the cost of the 30-100 TeV muon collider might plausibly be similar to that of the 200-400 TeV hadron collider in the scenario. There would probably also be a substantial overlap in the magnet technologies that drive the costs of the two colliders.
It is noted that, as a follow-up to the workshop’s parameter sets hemc99specs , new $`\mu ^+\mu ^{}`$ collider parameter sets are currently being generated epac2000 at the $`\mathrm{E}_{\mathrm{CoM}}=30`$ TeV and 100 TeV lower and upper limits of the range specified here.
#### The 1 PeV Linear Muon Collider
The final collider in table 1 is a 1 PeV linear $`\mu ^+\mu ^{}`$ collider. This item in the proof-of-plausibility scenario simply defers to an expert in linear colliders by assuming the 1 PeV parameters that were presented elsewhere in this workshop by Zimmermann Zimmermann and have already been discussed previously in this paper. (If a change to Zimmermann’s scenario were to be guessed at, it might be in the direction of more frequent bunches with smaller emittances but fewer muons per bunch. Such a refinement appears to move towards the anticipated potential capabilities of emittance reduction using the proposed method of optical stochastic cooling Zholents .)
Zimmermann does not specify a linac length or accelerating gradient, so this scenario makes the additional assumption of an accelerating gradient of 1 GV/m. This corresponds to a 500 km total length in each of the 2 linac tunnels. This gradient is rather ambitious, as was pointed out in the preceding subsection. The new laboratory site should have provision for a 1000 km long linear tunnel centered in the laboratory site, as is shown in figure 3.
In order for the beam acceleration to have some hope of an acceptable cost scaling with energy, Zimmermann assumes an electron-drive beam technology for the acceleration, such as is now under development for the CLIC linear $`\mathrm{e}^+\mathrm{e}^{}`$ collider. To be affordable at 1 PeV, this technology will need to reduce the cost-per-unit-length of the linac by more than an order of magnitude over the more conventional klystron-driven technology, to below $ 10 000/meter, corresponding to a total linac cost below $ 10B.
The luminosity of Zimmermann’s 1 PeV parameter set is $`5.4\times 10^{35}\mathrm{cm}^2.\mathrm{s}^1`$. This corresponds to only about half a unit of R in integrated luminosity per $`10^7`$ second accelerator year (see section III.2) and so falls squarely on top of the “borderline” luminosity assignment of equation 9:
$$^{\mathrm{borderline}}[\mathrm{E}_{\mathrm{CoM}}=1\mathrm{PeV}]=5\times 10^{35}\mathrm{cm}^2.\mathrm{s}^1.$$
(21)
The 1 PeV linear $`\mu ^+\mu ^{}`$ collider is the farthest extrapolation in time, energy and technology of all the colliders in the proof-of-plausibility scenario, and its overall feasibility and choice of parameters also have the largest uncertainties. Even so, reference Zimmermann was really only a first look at linear $`\mu ^+\mu ^{}`$ colliders and there is presumably much that can be done to further assess and develop this possibility with studies that wouldn’t be too time-consuming. The interested reader is encouraged to take a further look!
## VIII Assessment of Challenges and Prospects for the Proof-of-Plausibility Scenario
The individual challenges for each of the $`\mathrm{e}^+\mathrm{e}^{}`$, pp, and $`\mu ^+\mu ^{}`$ collider technologies in the proof-of-plausibility scenario have been briefly addressed in the preceding sections. The three subsections in this section now discuss the global features: the common technologies, overall costs and the potential physics rewards.
### VIII.1 Technological and Logistical Requirements for the Overall Scenario
#### Common Technologies
It has been noted in the preceding sections that pp and circular $`\mu ^+\mu ^{}`$ colliders share the technology of high field bending magnets as their cost drivers. In fact, it is true more generally that the technologies for progress in the three accelerator types are very much intertwined, as is itemized in table 2.
The considerable overlap serves to remind us of the common future of the field and to provide further impetus for cooperative technical studies between experts in each of the 3 accelerator types, and the technological health and vigor of each one of these accelerator types will trickle down to affect the others.
All of these technologies will require considerable R&D if a viable rate of progress towards the energy frontier is to be maintained. This implies a concerted and, probably, expanded commitment of resources to accelerator technologies, as was already pointed out several years ago by the Nobel laureate experimental physicist Samuel C. C. Ting Sci Am : “We need revolutionary ideas in accelerator design more than we need theory. Most universities do not have an accelerator course. Without such a course, and an infusion of new ideas, the field will die.”
#### The Desirability of a New World HEP Laboratory
As well as the technological overlap, the proof-of-plausibility scenario assumed that the 200-400 TeV hadron collider and the final circular and linear $`\mu ^+\mu ^{}`$ colliders would be housed in the same new world HEP laboratory. An “artist’s conception” example illustration of the layout for this scenario is shown in figure 3 and its large size scale has been emphasized in figure 4 by placing it somewhat arbitrarily on a map of Australia.
The choice of an appropriate large laboratory site would obviously be a major project involving a considerable amount of research followed by detailed political negotiations. Site selection would involve the optimization of many factors and the satisfaction of several requirements besides isolation Colin . For example, the site should be in a politically and economically stable country and should have ready access to an industrial base and resources such as power, transport and cooling water. Besides Australia, other candidate regions for a site would obviously include the U.S.A., Canada and several parts of Northern Europe or Asia. The example of figure 4, showing the site in an unpopulated desert region of Australia, would presumably use closed cooling loops to conserve water. It recalls the discussion of around 1980 on the “desertron”, which was the original popular name for a mooted energy frontier hadron collider that then evolved into the SSC project.
The consolidation of resources into a single new HEP laboratory should be more generally beneficial in all scenarios that envisage more than one collider extending the high energy frontier, not to mention that even a single such collider will anyway be too big to fit on existing laboratory sites. This would also help financially by avoiding duplication of laboratory operating costs – an aspect that will be covered in more detail in the following subsection.
### VIII.2 Could the Scenario be Affordable ?
For a plausible funding scenario, note that the straw-man scenario includes 5 energy-frontier colliders beyond the LHC (1 $`\mathrm{e}^+\mathrm{e}^{}`$, 1 pp and 3 muon colliders) over a time period of about 40 years. The following suggested cost goals appear to be plausible and correspond to an assumed level of worldwide construction funding on these machines of 1 B$/year over these 40 years:
1. 10 $B for the combined cost of the $`\mathrm{e}^+\mathrm{e}^{}`$ collider, the neutrino factory and the 4 TeV muon collider
2. 8 $B for the 200-400 TeV hadron collider
3. 10 $B for the final circular muon collider, at 30-100 TeV
4. 12 $B for the 1 PeV linear muon collider.
The 10 $B cost combined cost for the first machines in the scenario can be estimated with slightly more confidence than the others and appears difficult but plausible. The cost drivers contributing to the rest of these guessed figures have been at least touched on in the preceding sections but the potential reasonableness or otherwise of such cost goals is clearly a subject requiring much more careful and detailed study. Ideally, a global cost assessment of funding scenarios would benefit from explicit funding algorithms that have been benchmarked to recent large accelerator projects such as the SSC, LEP, HERA, RHIC and the LHC, and then peer-reviewed and continually refined by the HEP community as our technical understanding improves.
Besides the costs of the accelerators themselves, additional costs would include accelerator R&D and the commissioning and operating costs of the new world HEP laboratory. The laboratory would presumably cost several billion dollars to set up, although this might reasonably be partially or fully covered by a one-off contribution from the host country. Operating costs and upkeep for such a large laboratory might amount to several hundred million dollars per year, considering that the world’s largest current laboratory, CERN, has an annual budget in the range of half a billion dollars. Electricity for the colliders would be a significant part of the laboratory’s operating cost; to set the scale, operating for a $`10^7`$ second accelerator year with a total wall-plug power of 1 Gigawatt would cost 140 million dollars at an assumed rate of 5 cents per kilowatt hour.
All-in-all, the potential viability of the proof-of-plausibility scenario assumes something around 2 $B per year as the world-wide HEP budget devoted to the high energy frontier. This can be compared to the total funding for HEP in the U.S. alone Drell , which has bounced around at an average slightly below 1 $B per year since the early 1960’s but has risen as high as 1.2 $B in 1970 and 1.4 $B in 1992, at the height of SSC funding. (These figures are in FY 1995 U.S. dollars.) This indicates that the scenario could not easily be funded by one country acting alone but would instead require the combined commitments of all of the U.S., continental Europe, Britain, Japan and other, smaller contributors.
To summarize the cost discussion, the obvious answer to the question in the section title is “not easily”. The cost constraints will be extremely challenging in the straw-man collider scenario presented here, or in any other that holds to the historical precedent for progress along the Livingston plot. The costs of each machine will need to be agressively minimized and, even so, the field will need to make a united and convincing case for concerted world-wide funding at a level that is at least comparable to the historical norm for each of the contributing countries.
### VIII.3 Benchmarking the Experimental Potential of Future Colliders
The preceding discussion in this section has addressed the “pain” involved in keeping to the Livingston curves for accelerator progress. We now turn to the “gain” that would make all the effort worthwhile. If the future colliders covered in this proof-of-plausibility scenario turned out to be indeed feasible and were constructed, it would extend our energy reach for elementary particles from the current 300 GeV flagship in energy reach (the 1.8 TeV proton-antiproton Tevatron) all the way up to the 1 PeV energy scale.
The true significance that such an experimental advance would have is unknowable until it happens since it depends on what we find and how successful we are in tying it in to our theoretical understanding of the Universe. All we can do for now is extrapolate from past experience in particle physics. Such an advance of $`3\frac{1}{2}`$ decades in energy reach can be judged against these benchmarks:
* the past and present colliders on the Livingston plot span from about 1 GeV to 300 GeV. These $`2\frac{1}{2}`$ energy decades of collider experiments have been sufficient to revolutionize our knowledge of the elementary constituents of our universe. The final span of 6 energy decades would be an increase on this by a factor of 2.4, and it would be pessimistic to not predict that this would again revolutionize our level of understanding.
* those elementary particles in figure 1 that are not massless have a mass spectrum extending from the electron, with $`m_e5.11\times 10^4`$ GeV, to the top quark, with $`m_t1.75\times 10^2`$ GeV. (We exclude for the moment the very preliminary and indirect experimental evidence for non-zero neutrino masses that, if confirmed, would be much lower.) Therefore, the additional $`3\frac{1}{2}`$ energy decades could potentially broaden the known spectrum of elementary particle masses from the current $`5\frac{1}{2}`$ decades in mass to 9 decades, a substantial increase of more than 60%.
* $`3\frac{1}{2}`$ decades in energy reach would explore more than 20% of the log-energy gap from our current experimental reach all the way up to the Planck mass scale, at $`10^{19}`$ GeV. The Planck scale is defined by where quantum gravitational effects necessarily become important enough that even the framework for our current theories no longer makes sense and some theory that would surely be a close approximation to the Theory of Everything would be required to describe the physical processes. By biting off a significant fraction of this log-energy span we would presumably be giving ourselves a good shot at such an elevated level of understanding of our Universe.
## IX Conclusions
This paper has reviewed the past and future importance of accelerators for understanding our cosmos and its elementary constituents. The prospects for future $`\mathrm{e}^+\mathrm{e}^{}`$, pp and $`\mu ^+\mu ^{}`$ colliders were also reviewed and a proof-of-plausibility scenario was presented, incorporating the 5 plausible future colliders in table 1, that is able to hold to the historical rate of progress in the log-energy reach of hadron and lepton colliders and to reach the 1 PeV constituent mass scale by the early 2040’s.
While the challenges to a further half century of concerted progress on energy frontier colliders are great, the potential rewards are grander. Experimental discoveries at colliders would be expected to feed further advances in theoretical areas such as, e.g., string theory and, to reciprocate, any such theoretical advances would then motivate and inspire continued advances in colliders and collider experiments. The side-by-side progress of experiment and theory have the common and lofty goal of uncovering the long sought after “Theory of Everything” – the sum total of the elementary entities and organizing principles that underly the structure and processes of our physical Universe. Such an immortal pillar of knowledge and understanding, if attained, could justifiably be regarded as the greatest scientific achievement in all of human history, no less!
The high stakes and long-term nature of this scientific endeavor underly the wisdom of devoting some small fraction of our energies towards better understanding the possible options and required technologies for the years and decades to come. Proof-of-plausibility scenarios such as the one presented in this paper can guide this planning. However, they encompass diverse and often speculative areas of expertise and so will necessarily start out poorly informed and in need of refining in the furnace of scientific peer review, to then be augmented by alternative scenarios and either developed further or else disgarded as unrealistic.
Such a spirit of friendly and cooperative problem-solving, constructive criticism and model-building was a hallmark of this workshop. HEMC’99 was restricted to exploring the technologies and collider physics of many-TeV muon colliders but planning is underway for a follow-up study and workshop, in the Summer and Fall of 2001, that will explore the long-term prospects for all types of energy frontier colliders and the physics processes they might illuminate.
## X Acknowledgements
|
warning/0005/astro-ph0005032.html
|
ar5iv
|
text
|
# Orbital dynamics of Cygnus X-3
## 1. INTRODUCTION
First discovered in the pioneering X-ray surveys of the 1960s, Cygnus X-3 (Cyg X-3) remains among the least understood of the long-standing X-ray sources known in the sky. In the early 1970s, it was discovered to show a flux modulation on a period of 4.8 hours, first in X-rays (Parsignault et al. 1972; Sanford & Hawkins 1972) and then also at near-infrared wavelengths (Becklin et al. 1973), which was assumed to be the orbital period of the binary system. Modulation of the light curve at this period is asymmetric and extremely similar in both X-ray and infrared bands (Mason, Cordova & White 1986), with the exception of irregular, rapid flaring superposed on the orbital modulation in the infrared (Mason et al. 1986; Fender et al. 1996). The period of this orbit was further shown to be rapidly increasing, with a time-scale of just 850,000 yrs (Kitamoto et al. 1995).
Van den Heuvel & De Loore (1973) made the prediction that the system was comprised of a compact object and a helium star, thus representing a remarkably rare, late evolutionary stage of a massive X-ray binary. Nearly 20 years later, persuasive evidence to support this prediction was found in near-infrared spectroscopic data presented by van Kerkwijk et al. (1992; 1996). They discovered strong, broad emission lines of neutral and ionized helium, combined with a marked lack of hydrogen features. These properties are reminiscent of massive, highly evolved stars known as Wolf-Rayet stars (Smith 1968). As such, Cyg X-3 represents one of the most unique binary systems in our Galaxy. Should the helium companion star become a supernova, as is generally predicted for Wolf-Rayet stars (however, cf. Conti 1996), it could evolve into a double compact object binary system, such as the Hulse-Taylor pulsar, PSR 1913+16 (Hulse & Taylor 1975; Burrows & Woosley 1986). In addition, Cyg X-3 is one of the most luminous sources of radio jets in our Galaxy, displaying repeated ejections at (probably) relativistic velocities (e.g. Schalinski et al. 1998).
While much data have been gathered on the system through X-ray, near-infrared and radio observations, the most fundamental properties of the Cyg X-3 system have been difficult to constrain. This is a consequence of line emission found to be dominated by a non-axisymmetric wind component rather than one or both of the stellar objects. Furthermore, the wind features are generally weak, very broad and at times heavily blended, making line identification difficult or impossible. Consequently stellar masses have not been measured from binary motion. There exist plausible arguments for both a high (Schmutz, Geballe & Schild 1996) and low (Mitra 1996, 1998) mass companion star, which has inevitable consequences for the state of the compact object, be it a neutron star or a black hole. As the only known Wolf-Rayet + compact object binary and (probably) the shortest-period high-mass X-ray binary, accurate constraints on the mass and dimensions of the Cyg X-3 system are crucial for our understanding of many aspects of massive binary evolution.
In a previous paper (Fender, Hanson, & Pooley 1999, hereafter FHP99), we found a dramatic spectral change during an outburst of the system. Though the lines suffer from serious blending, it would appear that during outburst the spectrum of the system evolves from showing high-excitation single-peaked emission lines to one dominated by low-excitation double-peaked profiles. In this current paper we present a first analysis of these data in order to map the velocity field of the line-emitting regions.
## 2. OBSERVATIONS
We have recently completed a monitoring program of the Cyg X-3 system spanning 18 months. Our observations are based on broadband radio observations at 15 GHz combined with the Rossi XTE All-Sky Monitor program (FHP99). During this monitoring program we obtained the best near-infrared spectroscopy of the system, intermittently throughout the 18 months. These data, comprising over 100 individual spectra taken of Cyg X-3, were obtained over four different observing campaigns between May 1996 and October 1997. The data are available through the CDS archive website (see FHP99).
The near-infrared spectroscopic data used in this analysis are presented and fully described in FHP99. Observations were made using the Steward Observatory’s infrared spectrometer, FSpec (Williams, et al. 1993) on the Multiple Mirror Telescope prior to the recent upgrade to a single mirror. The spectra have an effective resolution of R $``$ 1200 at 2.12 $`\mu `$m, and a spectral coverage from 2.02 $`\mu `$m to 2.22 $`\mu `$m. Our analysis in this paper is based on epochs A, C and D, as defined by FHP99. The Cyg X-3 system was in quiescent-, outburst- and post-outburst mode, respectively, during these three epochs. Figure 1 displays spectra taken from each epoch in order to show the main spectral features present during these three epochs. In quiescence the K-band is dominated by broad, weak He ii and N v emission; in outburst strong twin-peaked He i emission dominates; for a detailed discussion see FHP99.
## 3. ANALYSIS
Because of the night-to-night variations of the emission line profiles observed during the outburst phase (see FHP99), it was important to obtain full orbital coverage on a single night for this analysis. We have time-resolved spectra providing full orbital sampling during two nights, 1996 June 2–3, when the system was in quiescence and 1997 June 19–20, when the system was in outburst. Despite the heavy blending, we were able to identify a few of the major features within the spectra using the laboratory transition lists of Morris et al. (1996).
### 3.1. June 1996 Quiescent Spectra
Trailed spectra from 1996 June are presented in the upper three panels of Fig 2. In each case, the continuum was removed by subtracting a 3-spline fit to nearby line-free wavelength regions and the data were recast into constant velocity bins of 121 km s<sup>-1</sup>. Spectra were further mean-averaged into 10 orbital phase bins using the ephemeris of Kitamoto et al. (1995). Doppler tomograms were constructed from the trailed spectra using the Maximum Entropy Method (Marsh and Horne 1988), an approach preferred here over other mapping algorithms because it allows the mapping of blended lines such as the He ii 2.1653,2.1891 $`\mu `$m doublet. It involves the transformation of velocity–phase information provided by the Doppler-broadened emission line profiles ($`V`$,$`\mathrm{\Phi }`$) to a velocity–velocity field ($`V_x`$,$`V_y`$) using:
$$f(V,\mathrm{\Phi })=I(V_x,V_y)\times $$
$$g(V\gamma +V_x\mathrm{cos}\mathrm{\Phi }V_y\mathrm{sin}\mathrm{\Phi })\text{d}V_x\text{d}V_y\text{.}$$
(1)
The mapping results from a $`\chi ^2`$-iteration between model and a fit and entropy minimization (Skilling and Bryan 1984). $`f`$ is the emission line intensity at velocity $`V`$ and orbital phase $`\mathrm{\Phi }`$, $`I`$ the emission distribution of the resulting map in velocity coordinates ($`V_x`$,$`V_y`$), $`\gamma `$ the systemic velocity of the binary center of mass and $`g(V)`$ the local line profile. From the combination of Doppler-broadened line profiles and rotational profile variations over the orbit, the maps are reconstructions of line emission in the velocity field around the binary, where $`V_y`$ is the velocity in the direction parallel to a line joining the two stellar centers and $`V_x`$ is the velocity orthogonal to $`V_y`$ in the plane of the orbit.
Any motion out of the orbital plane will not be recognized by the mapping and consequently Doppler tomograms are of most use in systems where line emission is confined to gas moving in a plane, such as distributed over an accretion disk or one of the stellar companions. If emission originates in a spherical wind the tomograms become difficult, if not impossible, to interpret with any confidence. Our motivation to apply tomographic mapping to the lines in Cyg X-3 is two-fold. First, in quiescence, the line emitting wind region is thought to be within the cone shadowed from the X-ray source by the companion star (van Kerkwijk et al. 1996). Motion in the cone is somewhat constrained to the orbital plane, although this assumption breaks down if the mass ratio $`q=M_{\text{WR}}/M_\text{X}`$ is too large. Secondly, FHP99 reveal that line emission profiles are double-peaked in outburst, suggesting that a disk or flattened, rotating wind may dominate profiles during these epochs, in which case Doppler tomography is well suited to these spectra.
The tomograms derived from the quiescent data of Fig 2 are shown in Fig 3. The reduced-$`\chi ^2`$ of these fits are listed in Table 1. For reference, the binary center of mass occurs at ($`V_x`$,$`V_y`$) = (0,0) km s<sup>-1</sup>, The companion star center of mass corresponds to ($`V_x`$,$`V_y`$) = (0,$`K_{\text{WR}}`$), where $`K_{\text{WR}}`$ is the radial velocity of the Wolf-Rayet star, $`K_{\text{WR}}`$ $`0`$. The compact object center of mass occurs at ($`V_x`$,$`V_y`$) = (0,$`K_\text{X}`$) where $`K_\text{X}`$ is the radial velocity semi-amplitude of the accreting star. The ratio $`K_\text{X}/K_{\text{WR}}=q`$, where $`q`$ is the ratio of stellar masses. Rotating structure around the compact object or WR star will manifest itself as an annulus on the map centered on $`K_\text{X}`$ or $`K_{\text{WR}}`$, respectively. The lower panel of Fig 2 shows the model emission profiles produced from the tomographic fit in Fig 3.
The trailed spectra and Doppler tomograms indicate that the line emission in quiescence arises in a restricted region of velocity space. This is consistent with the van Kerkwijk model in which the line emission originates in a region of the stellar wind shadowed from ionizing X-rays by the He star. Note that the cool wind component and binary motion vectors are orthogonal: this results in the wind phasing and WR star phasing being separated 90 degrees and the tomographic location of the emission occurring in a region offset approximately $`90^{}`$ from the tomographic location of the Wolf-Rayet star. Because of the moderate $`\chi ^2`$ of these fits the small scale structure is not significant.
Two-dimensional Gaussian fits to the tomograms provide a first-order estimate of both velocity, $`K=\sqrt{V_x^2+V_y^2}`$, and phasing, $`\mathrm{\Phi }=\mathrm{tan}{}_{}{}^{1}(V_x/V_y)`$, of the emitting gas and these are listed in Table 2. Uncertainties are not propagated through the mapping algorithm and therefore the errors are taken to be the size of a pixel element on the maps. Similar results are obtained by fitting one-dimensional, time-dependent Gaussian profiles to the raw data and this provides confidence in our tomographic fits. These further Gaussian fits are also presented in Table 2 and the resulting velocities plotted against orbital phase in Figure 4.
By taking a variance-weighted mean of the one-dimensional fits, the line forming region(s) have an average wind velocity of $`K`$ = 480 $`\pm `$ 50 km s<sup>-1</sup>. This is consistent with the value measured by Schmutz et al. (1996). We do not detect any significant difference in $`K`$ between the N v and the three He ii lines in this measure. From the fits, an average full-width half-maximum (FWHM) value was determined for each of the four transitions, and is given in Table 2; if the lines are well represented by Gaussian profiles, these values imply a terminal wind velocity of at least 1500 km s<sup>-1</sup> in the system, similar to that found by van Kerkwijk et al. (1996).
We have also more accurately determined the maximum of the blue-shifted wind features of He ii and N v against the X-ray modulation. From a weighted-mean of the four measurements, the He ii and N v features are at their maximum blue-shift when the X-ray intensity is at a minimum (e.g. $`\mathrm{\Phi }`$=0), with an error of 0.04 in orbital phase.
Note that there is no well-defined solution for the systemic velocity of the binary center of mass, $`\gamma `$ (Eqn. 1), from the 1-dimensional fits in Table 2. This could be the result of turbulence within the wind (e.g., Lépine & Moffat 1999). Furthermore, the laboratory wavelength for the N v transition is not accurately known. $`\gamma `$ can only be constrained to $`|\gamma |200`$ km s<sup>-1</sup>, precluding us from making a distance estimate. This upper limit on the systemic velocity seems rather difficult to reconcile with the $`+`$800 km s<sup>-1</sup> velocity field recently reported by Chandra observations of the system (Paerels et al. 2000).
### 3.2. June 1997 Outburst Spectra
The upper panel of Figure 5 shows the spectra taken during the June 1997 outburst. All lines seen have been attributed to He i, though exact line identifications for the $`\lambda =2.162.18`$ $`\mu `$m region are difficult to make. There is likely some contamination from the 2.1891 $`\mu `$m He ii feature normally seen during quiescence. Doppler tomograms, derived from the outburst spectra in Figure 5 are shown in Figure 6. Reduced-$`\chi ^2`$ of each fit are provided in Table 1. Finally, computed spectra, derived from the tomograms shown in Figure 6, are given in the lower panel of Figure 5.
The line profile behavior of the four He i lines appears at least marginally similar to that seen during quiescence: strong blue shifted emission at X-ray minimum and strong red shifted emission at X-ray maximum. However, the 2.0587 $`\mu `$m He i line in the outburst spectrum shows much stronger red-shifted emission at $`\mathrm{\Phi }`$ = 0.5 than the blue shifted emission at $`\mathrm{\Phi }`$ = 0.0 (Fig 5; see also Fig 4 of FHP99). Furthermore, in the region of the the 2.0587 $`\mu `$m He i line, a narrow, permanently blue-shifted absorption feature is seen, threaded through the background emission (Fig. 7).
The 2.1623 $`\mu `$m He i line appears almost double-peaked though, like the 2.0587 $`\mu `$m transition, the peak velocities are a bit asymmetric, showing strongest emission while at maximum red shift. Double-peaked emission may be evidence for rotating gas either in an accretion disk or a wind. However the stronger red peak also suggests these transitions may be contaminated by a P Cygni profile.
All three tomograms describe a ring structure similar to the tomographic signatures of an accretion disk (Fig. 6), consistent with rotational velocities on the order of 1000 km s<sup>-1</sup>. Unfortunately the mapping algorithm assumes that all line components have a velocity modulating symmetrically about $`\gamma `$. P Cygni profiles break this assumption and it is possible that the entire ring structure in each map is the result of P Cygni absorption. P Cygni components will not be recreated in the computed spectra and this explains the large discrepancies between the real data and fits in Fig. 5. It is important that we consider only the real data in the upper panels when interpreting line behavior in this system. Note though that the fits are poor for the 2.0587 $`\mu `$m transition (Table 1) but improve for the other maps suggesting that any P Cygni contributions are reduced in these transitions. Consequently the case for a disk or rotating wind cannot be proven or ruled out in any of these transitions.
Fortunately, the sinusoidal absorption feature found in the 2.0587 $`\mu `$m spectral region is far less ambiguous, being clearly discerned over the entire temporal sequence of Figure 5 and 7. This continuous blue absorption feature was fit with a Gaussian to determine its velocity modulation over the orbital cycle. Without knowing the shape of the underlying emission profile, the fits to the absorption feature are dubious. However since the absorption feature is narrow, being just a few resolution elements across, any bias to the result should be small. The feature was found to oscillate with a semi-amplitude of $`K=109\pm 13`$ km s<sup>-1</sup>. Most importantly, the phase of the velocity shifts, shown in Figure 8, are not like those seen in the N v and He ii features during quiescence (Figure 2) or like the He i emission features during outburst, whose red and blue maxima occur at $`\mathrm{\Phi }`$ = 0.5 and 0.0 (Figure 4). Instead, the absorption shows red and blue maximum at phases, $`\mathrm{\Phi }`$ = 0.18 and 0.68 (with an error of $``$ 0.02), relative to the orbital ephemeris of Kitamoto et al. (1995).
If $`\mathrm{\Phi }=0`$ corresponds to superior conjunction of the compact object in Cyg X-3, red and blue maxima of features associated with the mass donor He star should occur at $`\mathrm{\Phi }=0.25`$ and $`\mathrm{\Phi }=0.75`$, respectively. Thus the observed phasing of the absorption feature is approximately consistent with it tracing the motion of the He star. The impact of this result is explored further in Section 4.1. Fixing the phases of the maximum blue and red-shift to correspond with the binary phases $`\mathrm{\Phi }=0.25`$ and $`\mathrm{\Phi }=0.75`$ and re-fitting the radial velocity curves produces only a slightly poorer fit, $`K=100\pm 14`$ km s<sup>-1</sup>, but with a consistent semi-amplitude.
## 4. DISCUSSION
### 4.1. The He i 2.0587 $`\mu `$m transition
Some very important differences in the formation of the He i transitions have been described. This has lead to very different line fluxes and quite possibly different line profiles seen in the numerous lines detected in the Cyg X-3 outburst spectrum. First, 2.0587 $`\mu `$m emission is produced by direct radiative recombination into the 2$`{}_{}{}^{1}P`$ state, or through recombinations at higher levels which then cascades down into the 2$`{}_{}{}^{1}P`$ state. This is not a problem in very dense extended winds, as the radiation field pumps electrons via the He i resonance lines into upper levels, leading to an enhancement of recombinations to the 2 $`{}_{}{}^{1}P`$ level. However, once in this state, the atom has only a $``$ 1:1000 chance of decaying via the 2.0587 $`\mu `$m (2 $`{}_{}{}^{1}P`$ $``$ 2 $`{}_{}{}^{1}S`$) transition compared to the 584 Å (2 $`{}_{}{}^{1}P`$ $``$ 1 $`{}_{}{}^{1}S`$) resonant transition. Thus strong 2.0587 $`\mu `$m emission further requires a very high optical depth in the 584 Å line. There is a second equally important point about the He i 2.0587 $`\mu `$m line. Its lower level, 2 $`{}_{}{}^{1}S`$, is meta-stable and becomes increasingly overpopulated (relative to LTE). Frequently, a 2.0587 $`\mu `$m P Cygni profile is seen in hot, dense WR winds because in the outermost regions the absorption process becomes quickly favored due to the overpopulated lower level (Williams & Eenens 1989). Thus, the expected behavior of the 2.0587 $`\mu `$m line in an extended wind is for it to self-absorb readily, i.e., create a P Cygni feature. It is of interest to note that several months after the outburst in June 1997, the dominant feature in the spectrum of Cyg X-3 is the P Cygni profile occurring at 2.0587 $`\mu `$m (Fig 1, see also FHP99).
The remaining transitions, most specifically the $`4s3p`$ triplet at 2.1126 $`\mu `$m (the singlet $`4s3p`$ line at 2.1138 $`\mu `$m has about one third the flux of the combined emission from the triplet transition), are not directly coupled to a resonance transition. Their emission level is set only by the recombination rate of He<sup>+</sup>. Furthermore, they are not predisposed to increased absorption in dense winds like the 2.0587 $`\mu `$m transition, since their lower states readily cascade to lower levels. These differences can give rise to rather different behaviors in the He i lines seen in our outburst spectra, and may possibly explain why the weak sinusoidal absorption feature seen threaded through outburst spectrum shown in Figure 5, is detected only in the 2.0587 $`\mu `$m He i line.
### 4.2. Orbital dynamics from the He i 2.0587 $`\mu `$m absorption line
Fortunately, an exact identification of the absorption feature seen in Figure 5 is not necessary for our purposes. While we feel confident the feature is due to He i 2.0587 $`\mu `$m given the arguments given above, even if the line is unknown, or due to an unresolved blend of features, we can still use its velocity signature given in $`K`$ to measure the orbital dynamics of the Cyg X-3 system. As already noted, the phasing of the absorption feature seems consistent with a source which is distributed approximately isotropically around the star and hence should reflect the binary motion of the He star in the system. However, before continuing with that interpretation, we note that there have been alternative suggestions for the phasing of the system. For example Schmutz et al. (1996) propose that the He ii and N v emission lines commonly observed in quiescence (Figure 2) track the motion of the He star (in which case X-ray minimum light would be occurring one quarter cycle before superior conjunction of the compact accretor). So it is clear that there is not a consensus about the interpretation of the phasing in the Cyg X-3 system. As a result we have investigated the implications of an origin for the absorption line to originate in material centered on the accretor and on the donor.
Dynamical solutions for the first case, that the absorption line in the material is centered on the accretor, are presented in Fig 9a. The solid curves show the mass solution assuming the compact object is a neutron star of M<sub>x</sub> = 1.4 M. For this solution only, two solid curves are given to demonstrate the error range on the velocity measurement (109 $`\pm `$ 13 km s<sup>-1</sup>). This orientation, where the absorption is centered on the accretor, is illustrated in Fig 9b. We see from Fig 9a that for all accretor masses, unless the He star is much less massive than implied from luminosity arguments (i.e. $`2`$M), the inclination of the system must be much lower ($`50^{}`$) than is generally accepted based on the X-ray and infrared modulation. Furthermore, any solution that allows for a reasonable inclination, $`50^{}`$, simultaneously gives an absurdly small, nearly non-physical, value for the Roche lobe radius of the donor. As a result we do not consider an origin for the absorption line in the accretion disc or any other material centered on the accretor to be very likely.
Rather, our analysis implies the most likely configuration is the second case, that the material creating the absorption feature is centered on the mass donor. Solutions for this interpretation are presented in Fig 10a and a schematic illustrating the orientation with phase is shown in Fig 10b. Again, two solid curves are given to illustrate the range of solutions consistent with the mass accretor being a neutron star, M<sub>x</sub> = 1.4 M, given our error on the velocity measurement. In this orientation, X-ray minimum occurs at superior conjunction of the accretor. It is immediately apparent from Fig 10a that realistic high-inclination solutions are now available. Using this interpretation we directly derive the mass function of the system, which is 0.027 M. The reason this does not constrain at all the nature of the accretor is the large, probably dominant, mass of the He donor star. For a neutron star accretor of 1.4 M the mass of the He star is most likely in the range 5 M $``$ M$`_{\text{WR}}`$ $``$ 11 M, assuming the system has a moderately high inclination, $`i>60^{}`$.
However, if the accretor is a black hole, there is much less confidence about the mass of the He star. Charles (1998) lists derived masses for black holes in low-mass X-ray binaries and finds them to be in the range 5–12 M. Careful inspection of Fig 10a reveals that this allows a wide range of possible solutions, most of which are compatible with very large masses indeed (i.e. several tens of M) for the He star. This is because the Roche radius given in Figure 10a increases for greater He star mass and easily stays within the expected radius limit for such stars (Langer 1989). Moreover, the Roche radius of the He star shown in Figure 10a can even be less than its R, as shown in Langer’s Figure 5. This is because the tenuous outer atmospheric radius of the star, identified as R, contributes an insignificant fraction of the mass of the star, and can extend well beyond the dynamical Roche radius. Nonetheless, we see from Figure 10a that solutions for a M$`_{\text{BH}}`$ $``$ 10 M would be inconsistent with our assumption of even a moderate inclination for the binary system, for any realistic companion star mass, M$`_{\text{WR}}`$ $``$ 70 M. For this reason, we put an upper limit on the mass of the compact object of M$`_{\text{BH}}`$ $``$ 10 M. The well known high mass X-ray binary system Cyg X-1 has a similar mass estimate for its black hole (Charles 1998).
We note that in the binary orientation proposed by Schmutz et al. (1996), which lead them to argue for a most likely value of M$`_{\text{BH}}`$ $``$ 17 M for the compact object, the absorption feature would be required to arise from a region centered not on the He star or accretor, but a quarter of a cycle out of phase with either orientation.
### 4.3. The nature of the stellar companion
The assumptions used to determine stellar mass in Figure 10a also provide limits for the radius of the stellar companion of Cyg X-3. We are able to test whether this result is consistent with the characteristics, most specifically the luminosity, predicted from atmosphere models and observations of typical WR stars.
The distance to Cyg X-3 is not directly known. However, Dickey (1983) has suggested a lower limit on the distance, 11.6 $`\times `$ $`\pi `$/10 kpc, where $`\pi `$ is the distance in kpc to the galactic center, based on the detection of high velocity neutral hydrogen absorption features and applying a flat rotation curve for the Galaxy. With an updated value for the distance to the Galactic center of $``$ 8 kpc (Reid 1993), this gives a lower limit to Cyg X-3 of $``$ 9 kpc. Original estimates of the line of sight extinction to Cyg X-3, A$`{}_{V}{}^{}=15`$, were made by Becklin et al. (1973). More recent measurements suggest a slightly higher value of A$`{}_{V}{}^{}=19`$ (Molnar et al. 1988), A$`{}_{J}{}^{}=5.5`$ (van Kerkwijk et al. 1996) which corresponds approximately to A$`{}_{V}{}^{}=19`$, and A$`{}_{J}{}^{}=6.0`$ (Fender et al. 1996) corresponding to approximately A$`{}_{V}{}^{}=20.5`$ (Rieke & Lebofsky 1985). Determining the line of sight extinction is dependent on measuring the flux distribution over a large enough wavelength range to note the differential change, as done by both van Kerkwijk et al. and Fender et al. But equally important, one must simultaneously make assumptions about the functional form of that extinction and the functional form of the underlying continuum of the source. For this reason, without an a priori quantitative understanding of the stellar companion, the line of sight extinction cannot be as well constrained as one might hope. Mindful of these limitations, we have used the apparent magnitudes of Cyg X-3 given by Wagner et al. (1989) and Fender et al. (1996), $`m_I=20.0`$, $`m_H=13.11`$, $`m_K=11.76`$ and applied an extinction of approximately $`A_V=20`$ and a distance of 9 kpc to obtain absolute magnitudes of $`M_I=4.8`$, $`M_H=5.2`$, $`M_K=5.1`$ for the stellar companion. It is precisely these values, the very luminous lower limits for the absolute magnitude of the stellar companion, that most strongly challenge the notion of Cyg X-3 being a low mass X-ray binary system (Mitra 1998)
We have argued that while in quiescence, the emission lines seen in Cyg X-3, presumably dominated by the companion star, look most similar to the lines seen in early WN-type (WNE) WR stars (FHP99). Based on direct measurements of WNE stars given in Smith et al. (1994), a mass range from 6 to 12 M, corresponds to a typical, $`4>M_V>4.8`$. Naturally, there is a large scatter in the observed relation. Without a direct determination of $`M_V`$, with the uncertainties in the bolometric correction and with the difficulty in extrapolating the extinction characteristics towards Cyg X-3, the luminosity for the stellar component of Cyg X-3 is not highly inconsistent with the luminosity of other WNE stars.
### 4.4. The nature of the accretor
The crucial question of whether the accretor is a neutron star or a black hole cannot be decided with confidence. The only direct observational evidence for the presence of a neutron star in Cyg X-3 was in the report of 12.6 ms $`\gamma `$-ray pulsations, $`15`$ years ago (Chadwick et al. 1985). However, this result has never been widely accepted. As already stated, Schmutz et al. (1996) argue that the observational evidence supports a black hole accretor, which prompted Ergma & Yungelson (1998) to investigate the evolution of such a system. They concluded that Cyg X-3 may contain a black hole accreting at super-Eddington rates. Mitra (1998) however argued strongly against the interpretation of Schmutz et al. (1996), preferring a low-mass system (which is extremely hard to reconcile with the very high luminosity of the system). For now we will have to consider a fairly large range of accretor masses up to M$`_{\text{BH}}`$ $``$ 10 M (see discussion in §4.2).
## 5. CONCLUSIONS
The X-ray binary, Cyg X-3, is the only known system containing a (presumably) massive He star with a compact object. This makes Cyg X-3 a uniquely important link in our understanding and testing of massive X-ray binary evolution. In this paper we pursued an emission line analysis of the quiescent and outburst spectra presented in FHP99. We reported that while the double-peaked emission seen in the He i lines during outburst are consistent with a disk-wind geometry as proposed in FHP99, our tomographic analysis in particular reveals unresolvable ambiguities in the mechanism responsible for the formation of double-peaked emission. For this reason, additional constraining spectral information or methods are needed to resolve the nature of the double-peaked emission and its relation to the wind or disk geometry of the system during outburst.
Of greater significance was the detection of a weak absorption feature threading through the blue wing of the 2.0587 $`\mu `$m He i line. This feature moves $``$ 1/4 out of phase with all other spectral modulations seen in Cyg X-3, consistent with He i absorption originating in an isotropic, asymptotic wind from the companion star. Consequently it can be used to derive the first radial velocity curve for the Cyg X-3 system. Employing reasonable geometric assumptions, we derive a mass range of 5 M $``$ M$`_{\text{WR}}`$ $``$ 11 M for the He star, if the accretor is a neutron star. Additionally, should the accretor be a black hole, we determine M$`_{\text{BH}}`$ $``$ 10 M, based on an upper mass limit, M $``$ 70 M, for the He star companion.
Identification of the critical absorption feature used in obtaining a mass function for the Cyg X-3 system was made possible only through phased-resolved near-infrared observations. A further key was obtaining observations during outburst. Clearly, periods of high X-ray variability would make the most opportune times to catch this rare infrared state once more. The spectra presented in FHP99 represent the best quality a 4-meter class telescope will likely ever deliver on this object, and yet these spectra barely detect the very weak, sometimes very broad, and often heavily blended features. Its not certain that moving to an 8-meter class telescope would resolve all the problems with the current data set, as many of the lines are already resolved with the current spectra. However, the narrow absorption feature detected in the outburst spectrum may be just the first of many similarly weak and narrow features which might better be revealed with higher-quality spectra. Such features offer promise in resolving the ambiguities plaguing our analysis.
FHP99 showed there to be a marked decrease in X-ray flux during the four days prior to the near-infrared outburst, the minimum occurring just after outburst. This may have resulted from drops in the ionization state of the wind following a temporary decrease in the X-ray flux. FHP99 showed that the outburst occurring in June 1997 and previous outbursts suggested in earlier spectra (van Kerkwijk et al. 1996) were short lived ($``$ 24-h). Careful day-to-day monitoring of Cyg X-3 in the X-ray and radio, particularly during epochs of high X-ray and radio activity, would potentially precede suitable windows for near-infrared spectroscopic observations of P Cygni lines.
We are grateful for comments made by our referee which lead us to greatly improve the presentation of this work. Observations reported in this paper were obtained with the Multiple Mirror Telescope, operated by the Smithsonian Astrophysical Observatory and the University of Arizona. We are grateful to G. and M. Rieke for their critical support in this program. M. M. H. received support for this work provided by The University of Cincinnati through a University Research Council Grant and a Faculty Summer Fellowship.
|
warning/0005/cond-mat0005383.html
|
ar5iv
|
text
|
# Optimizing the classical heat engine.
## Abstract
A pair of systems at different temperatures is a classic environment for a heat engine, which produces work during the relaxation to a common equilibrium. It is generally believed that a direct interaction between the two systems will always decrease the amount of the obtainable work, due to inevitable dissipation. Here a situation is reported where, in some time window, work can be gained due the direct coupling, while dissipation is relevant only for much larger times. Thus, the amount of extracted work increases, at the cost of a change of the final state.
Introduction.— The classical problem of thermodynamics is the determination of the maximal amount of work that can be extracted from a non-equilibrium system, during the relaxation to the equilibrium state . In recent years the interest in this time-honored problem was renewed (see for review). In many cases the limits proposed by the founders of the classic thermodynamics have appeared as too idealistic, and attention was focused on the study of dissipative effects, which restrict the abilities of realistic heat engines . One of the most popular examples of that kind is a direct interaction between the thermal baths that drive the standard heat engine . Evidently, this is a way to dissipate energy. The common opinion, expressed in textbooks is that dissipation is the main, if not the only effect of any direct interaction.
The purpose of the present paper is to show that a direct interaction between the baths may have energy transfer, rather than energy dissipation, as its main physical effect. The reason is that the relevant timescale of the transfer arising from the direct coupling can be widely separated from the dissipative timescale. This leads to an optimization of the classical heat engine, in a certain time window. As a result, the final equilibrium state of the heat baths is changed. The mechanism arises from the recently proposed steady adiabatic state (see also in the related context).
First, we shall reproduce the classical discussion of the maximal amount of work that can be extracted from a non-equilibrium system. The underlying ideas are, of course, well-known, and we shall need them for a careful interpretation of our main result.
The classical analysis.— Consider two subsystems which have different temperatures $`T_1`$ and $`T_2`$. Depending on the concrete relaxation process, the whole thermally isolated system will produce different amounts of work. This model is well-known and described in textbooks on thermodynamics . We shall assume that the total volume of the system is unchanged by the relaxation (although it can vary during the process), since we wish to take into account only the work that can be done due to the non-equilibrium initial state, and not due to the general expansion. Denoting by $`U_i`$, $`S_i`$ the initial energy and entropy of the system, we get the following expression for the work $`W`$ performed by the system
$$W=U_iU(S),$$
(1)
where $`U(S)`$ is energy of the final equilibrium state as a function of its entropy. Because the temperature is positive, $`U`$ is a monotoneous function of $`S`$ ($`U/S|_V=T>0`$ in equilibrium). Therefore, $`W`$ is maximal when $`S`$ is as small as possible. Since the whole system is thermally isolated, the second law demands $`SS_i`$. The maximal amount of work is attained for $`S=S_i`$, i.e., for a reversible process towards equilibrium. It is believed that for obtaining the maximal amount of work any direct interaction between subsystems should be removed, because it would induce irreversible relaxation. Thus, a third body (“engine”) should operate between our subsystems to perform the work. At the end of the relaxation process the working body must return to its initial state. The expression of the total amount $`W`$ of the extracted work can be given as
$$W=U_1(T_1)+U_2(T_2)U_1(T_0)U_2(T_0),$$
(2)
where the final temperature $`T_0`$ is determined by the reversibility condition
$$S_1(T_1)+S_2(T_2)=S_1(T_0)+S_2(T_0),$$
(3)
and $`U_k`$, $`S_k`$ ($`k=1,2`$) are energy and entropy of the corresponding subsystems. The reversible relaxation process consists of an infinite amount of elementary cycles of the third body. The famous Carnot process is the most representative example for such a cycle. The local efficiency $`\eta `$ of the optimal cycle, defined as the maximal extracted work during the cycle divided by the input energy, depends only on the conditions of reversibility and conservation of energy: $`\eta =1(T_1/T_2)`$, if $`T_1<T_2`$ . This efficiency is universal and system-independent, emphasizing the power of thermodynamics. In contrast, the maximal amount of the extracted work (2) is not universal, and can vary from one system to another.
The steady adiabatic state.— Steady adiabatic systems have two distinctive properties : 1) different subsystems are not in the mutual equilibrium, but possess different temperatures, 2) the local characteristic relaxation times of these subsystems are very well separated. Let us consider a statistical system which has two subsystems with coordinates $`x_1`$ and $`x_2`$. ($`x_1`$ and $`x_2`$ can thus also code also a set of variables; we will not denote that explicitly.)
The corresponding relaxation times are denoted by $`\mathrm{\Gamma }_1`$ and $`\mathrm{\Gamma }_2`$. The condition of well separated timescales is ensured by $`\gamma =\mathrm{\Gamma }_1/\mathrm{\Gamma }_21`$. The stationary distribution of this system can be found from the following heuristic arguments (for a more rigorous presentation see ). Due to the large difference between relaxation times the $`x_1`$-subsystem comes to an equilibrium while the $`x_2`$-variable is almost unchanged. The corresponding Gibbs stationary distribution reads
$$P(x_1|x_2)=\frac{1}{Z(x_2)}\mathrm{exp}[\beta _1H(x_1,x_2)],$$
(4)
where
$$H(x_1,x_2)=H_1(x_1)+H_2(x_2)+gH_{int}(x_1,x_2)$$
(5)
is the total system’s Hamiltonian, $`T_1=1/\beta _1`$ is the temperature of the $`x_1`$-subsystem, and $`Z(x_2)`$ is the partition sum at fixed $`x_2`$. For obtaining the coarse-grained distribution $`P(x_2)`$ notice that after integrating out of the vast variable $`x_1`$, the effective Hamiltonian for the slow variable $`x_2`$ is just $`T_1\mathrm{ln}Z(x_2)`$, the free energy of the $`x_1`$-subsystem at fixed $`x_2`$. The stationary distribution of $`x_2`$ then reads
$$P(x_2)=\frac{Z^{T_1/T_2}(x_2)}{𝒵},𝒵=dx_2Z^{T_1/T_2}(x_2).$$
(6)
The complete stationary distribution can now be written as $`P(x_1,x_2)=P(x_2)P(x_1|x_2)`$. The mean energy and entropy of the system are given by the following general definitions
$`U={\displaystyle dx_1dx_2H(x_1,x_2)P(x_1,x_2)},`$ (7)
$`S={\displaystyle dx_1dx_2P(x_1,x_2)\mathrm{ln}P(x_1,x_2)},`$ (8)
$`S`$ can be decomposed as the sum of the entropies of the slow and fast subsystems, $`S=S_1+S_2`$, with
$`S_1={\displaystyle dx_2P(x_2)[dx_1P(x_1|x_2)\mathrm{ln}P(x_1|x_2)]},`$ (9)
$`S_2={\displaystyle dx_2P(x_2)\mathrm{ln}P(x_2)}.`$ (10)
$`S_1`$ is the entropy of the fast variable $`x_1`$ at $`x_2`$, averaged $`x_2`$, while $`S_2`$ is the entropy of this slow variable itself. As was shown in the considered system admits a thermodynamical description. In particular, defining the free energy as $`F=T_2\mathrm{ln}𝒵`$, we get:
$$F=UT_1S_1T_2S_2,$$
(11)
where the entropies can also be obtained by the standard relations:
$$S_1=\frac{F}{T_1}|_{T_2},S_2=\frac{F}{T_2}|_{T_1}.$$
(12)
The energy and entropies are constant in the steady state, but a direct coupling induces a steady entropy production at rate $`\dot{S}`$ and an energy dissipation at rate $`\dot{\mathrm{\Pi }}`$. These quantities were analyzed in Ref. , and the obtained formulas read $`\dot{\mathrm{\Pi }}=T_2\dot{S}+𝒪(\gamma ^2)`$, with
$`\dot{S}=\gamma g^2{\displaystyle \frac{(T_1T_2)^2}{\mathrm{\Gamma }_1T_1^2T_2}}{\displaystyle }\mathrm{d}x_2\mathrm{d}x_1P(x_1,x_2)\times `$ (14)
$`\left[{\displaystyle \frac{H_{int}(x_1,x_2)}{x_2}}{\displaystyle dyP(y|x_2)\frac{H_{int}(y,x_2)}{x_2}}\right]^2+𝒪(\gamma ^2)`$
Although these results were obtained for the strictly steady state, they can be applied also in the time-dependent case, if the characteristic time of this quasi-stationary process is much larger than the largest relaxation time, $`\mathrm{\Gamma }_2`$. For instance, to obtain the time-dependent distribution function for the case of slowly (adiabatically) changing temperatures or other parameters, one just inserts these time-dependent values directly in Eqs. (4, 6). In this context, the change of free energy (11) can be shown to be the adiabatic work $`𝒲_{ad}`$ done on the system, when varying a parameter $`\alpha `$ (for example, the width of the potential, or a coupling constant) from the initial value $`\alpha _i`$ to final value $`\alpha _f`$ at constant temperatures:
$`𝒲_{ad}`$ $`={\displaystyle _{\alpha _i}^{\alpha _f}}d\alpha {\displaystyle dx_1dx_2P(x_1,x_2,\alpha )\frac{H(x_1,x_2,\alpha )}{\alpha }}`$ (17)
$`=F(T_1,T_2;\alpha _f)F(T_1,T_2;\alpha _i)`$
This is fully analogous to the property of the usual (single-temperature) free energy.
The dissipative effects (given by Eq. (14)) are small for small $`\gamma `$, $`g`$. Let us neglect them for the moment; later we shall show that this is allowed in a certain timewindow.
The maximal amount of work extracted from the steady adiabatic state.— Certainly, we can apply the above-mentioned general analysis, concerning the maximal work, to our adiabatic system. In fact, this analysis does not use any concrete property of the initial non-equilibrium state, but we should take into account that our subsystems interact directly, and not only through the third body. As compared to the case without direct coupling, the system will now relax to a different equilibrium state, and this is the reason why one can get more work done by it. The total amount of the gained work can be again written as
$$W(g)=U(T_1,T_2)U(T_g,T_g),$$
(18)
and the temperature $`T_g`$ of the final equilibrium state is defined from the condition of reversibility
$$S(T_1,T_2)=S(T_g,T_g).$$
(19)
This condition involves the total entropy of the interacting subsystems, and now assumes that there are no additional sources of dissipation besides (14).
We shall investigate Eqs. (18, 19) to first order in the small parameter $`g`$. Hereafter quantities of the order $`g^0`$ and $`g^1`$ will be indicated by the index $`0`$ and $`1`$, respectively. It is evident from (5) that to order $`g^1`$ it holds
$$F=F^{(0)}+gV_0$$
(20)
where
$$V_0(T_1,T_2)=dx_1dx_2P(x_1,x_2)H_{int}(x_1,x_2).$$
(21)
Using Eqs. (7-12), one gets
$`S_1=S_1^{(0)}+gS_1^{(1)}=S_1^{(0)}g_{T_1}V_0,`$ (22)
$`S_2=S_2^{(0)}+gS_2^{(1)}=S_2^{(0)}g_{T_2}V_0,`$ (23)
$`S=S^{(0)}+gS^{(1)}=S^{(0)}g\{_{T_1}+_{T_2}\}V_0,`$ (24)
$`U=U^{(0)}+gU^{(1)}=U^{(0)}+g\{1T_1_{T_1}T_2_{T_2}\}V_0`$ (25)
Let us now obtain from Eq. (19) an expression for the final temperature $`T_g`$ to order $`g`$, given the value of $`T_0`$, the final temperature for $`g=0`$,
$$T_g=T_0\left(1+g\frac{S^{(1)}(T_1,T_2)S^{(1)}(T_0,T_0)}{C_1+C_2}\right).$$
(26)
Here $`C_k=T_kS_k^{(0)}/T_k|_V=U_k^{(0)}/T_k|_V`$, with $`k=1,2`$, are the heat capacities of the subsystems when they are uncoupled. Starting from Eq. (18) and using Eqs. (22)-(25), we finally derive the excess work at order $`g`$
$`W^{(1)}=\underset{g0}{lim}{\displaystyle \frac{W(g)W(0)}{g}}`$ (27)
$`=V_0(T_1,T_2)V_0(T_0,T_0)+{\displaystyle \underset{k=1}{\overset{2}{}}}(T_0T_k)_{T_k}V_0(T_1,T_2).`$ (28)
This is the first main result of our work. It remains to be shown in a specific example that this quantity can be positive. Let us first point out that further simplifications occur when $`T_1`$ is close to $`T_2`$. To first order in the parameter $`T_1T_2`$ eq. (3) gives us
$$T_0=\frac{C_1+C_2}{C_1T_1+C_2T_2}T_1T_2,$$
(29)
while Eq. (27) can be approximated by
$$W^{(1)}=\underset{k=1}{\overset{2}{}}(T_0T_k)(_{T_k}V_0\{_{T_k}V_0\}|_{T_1=T_2=T_0}).$$
(30)
To illustrate the general results let us present a concrete model, where the direct interaction increases the total amount of work: $`W(g)>W(0)`$. One of the most popular models of the thermal bath is a set of harmonic oscillators , which is frequently used to derive kinetic equations or to gain fundamental insight. Following this well-established tradition, we shall model our first (second) thermal bath by $`N_1`$$`(N_2`$) oscillators at temperature $`T_1`$ ($`T_2`$), and weakly-anharmonic interaction:
$$H=\frac{1}{2}\underset{i=1}{\overset{N_1}{}}x_{1,i}^2+\frac{1}{2}\underset{i=1}{\overset{N_2}{}}x_{2,i}^2+g\underset{i=1}{\overset{N}{}}x_{1,i}^2x_{2,i}^2,$$
(31)
where $`g>0`$, and $`N_1,N_2N`$. It is straightforward to show that to order $`g`$ one has the partial partition sum
$$Z(x_2)=T_1^{N_1/2}\mathrm{exp}\left(\frac{1}{2}\beta _1\underset{i=1}{\overset{N_2}{}}x_{2,i}^2g\underset{i=1}{\overset{N}{}}x_{2,i}^2\right)$$
(32)
and the full partition sum
$$𝒵=T_1^{\frac{1}{2}N_1T_1/T_2}T_2^{\frac{1}{2}N_2}(1+2gT_1)^{\frac{1}{2}N}$$
(33)
The latter result yields the free energy $`F=T_2\mathrm{ln}𝒵`$
$$F=\frac{1}{2}N_1T_1\mathrm{ln}T_1\frac{1}{2}N_2T_2\mathrm{ln}T_2+gNT_1T_2$$
(34)
in agreement with the fact that $`V_0=NT_1T_2`$. According to previous rules we derive
$`S_1`$ $`=`$ $`{\displaystyle \frac{1}{2}}N_1(\mathrm{ln}T_1+1)gNT_2`$ (35)
$`S_2`$ $`=`$ $`{\displaystyle \frac{1}{2}}N_2(\mathrm{ln}T_2+1)gNT_1`$ (36)
The internal energy follows as $`U=F+T_1S_1+T_2S_2`$,
$$U=\frac{1}{2}N_1T_1+\frac{1}{2}N_2T_2gNT_1T_2$$
(37)
Notice that the sign of the order $`g`$ correction is negative, due to entropic effects. To determine $`T_0`$ from Eq. (3) we need entropies $`S_1^{(0)}`$, $`S_2^{(0)}`$, that can be read of from Eqs. (35), (36) at $`g=0`$. One obtains $`T_0=T_1^{\nu _1}T_2^{\nu _2}`$ and then from Eq. (26)
$`T_g=T_1^{\nu _1}T_2^{\nu _2}[1+2g\nu (2T_1^{\nu _1}T_2^{\nu _2}T_1T_2)],`$ (38)
where $`\nu =N/(N_1+N_2)`$, $`\nu _k=N_k/(N_1+N_2)`$, $`k=1,2`$. According to Eq. (1) and using $`U_k^{(0)}=N_kT_k/2`$, one gets for $`g=0`$ the maximal amount of work
$$W(0)=\frac{1}{2}N_1(T_1T_1^{\nu _1}T_2^{\nu _2})+\frac{1}{2}N_2(T_2T_1^{\nu _1}T_2^{\nu _2})$$
(39)
Taking into account that $`V_0(T_1,T_2)=NT_1T_2`$, one gets from Eq. (27) a non-negative shift in the maximal of amount of work
$`gW^{(1)}`$ $`=`$ $`gN(T_1^{\nu _1}T_2^{\nu _2}T_1)(T_2T_1^{\nu _1}T_2^{\nu _2})0.`$ (40)
The equality is realized in the trivial case $`T_1=T_2`$. Only for $`g<0`$ this mechanism would reduce the work.
Eq. (27), and especially our model-dependent result (40), show that the full amount of the extracted work can increase due the direct coupling. Let us now return to the dissipative effects. For the considered model the energy dissipated per unit of time $`\dot{\mathrm{\Pi }}=T_2\dot{S}`$ can be derived from Eq. (14). It reads
$$\dot{\mathrm{\Pi }}=\frac{8g^2N}{\mathrm{\Gamma }_1}\gamma T_2(T_1T_2)^2+𝒪(\gamma ^2)$$
(41)
Our aim is now to obtain the characteristic time $`𝒯`$, after which the energy dissipated due to the direct coupling is comparable with the energy $`gW^{(1)}`$ gained according to Eq. (40). For the dissipated energy an upper estimate can be given as $`\dot{\mathrm{\Pi }}𝒯`$, and we get from $`\dot{\mathrm{\Pi }}𝒯=gW^{(1)}`$,
$$𝒯=\frac{\mathrm{\Gamma }_2}{8g}\frac{(T_1^{\nu _1}T_2^{\nu _2}T_1)(T_2T_1^{\nu _1}T_2^{\nu _2})}{(T_1T_2)^2}.$$
(42)
To be able to neglect the dissipated energy, the duration of our process $`t`$ must be much smaller than $`𝒯`$. On the other hand, since we are getting the work in the relaxation process, its duration $`t`$ must be much higher than the largest relaxation time $`\mathrm{\Gamma }_2`$. Thus, for times
$$\mathrm{\Gamma }_2t𝒯$$
(43)
it is possible to perform more work due to the presence of the direct coupling. The necessary condition $`\mathrm{\Gamma }_2𝒯`$ is realized mainly when $`g`$ is small. If one was not able to complete the relaxation in the time-window (43), then for $`t𝒯`$ the gained work will be equal to that without any direct coupling. The same analysis can be applied for the general case.
So far we have compared the efficiencies of two systems with the same initial temperatures $`T_1`$, $`T_2`$, and different values of $`g`$. One can also compare cases of identical initial energies, $`(T_1,T_2,g=0)`$ and $`(\overline{T_1},\overline{T_2},g>0)`$, where the temperatures $`\overline{T_1},\overline{T_2}`$ are defined by $`U(T_1,T_2,g=0)=U(\overline{T_1},\overline{T_2},g>0)`$. The analysis is very similar to that given above, and indicates that our main result remains valid also in this case. There are examples of $`(\overline{T_1},\overline{T_2})`$, for which the direct coupling enhances the work.
In the context of our main result it is useful to investigate which amount of work $`𝒲(0g)`$ should be spent by external sources to switch on the small coupling $`g`$, starting from the state with $`g=0`$. We shall consider the two extremes, very slow and very fast switching, which happen to give the same answer for small $`g`$. In the first case one uses Eqs. (17, 20) to obtain ($`T_1`$, $`T_2`$ are constant)
$$𝒲_{ad}(0g)=gV_0$$
(44)
For the very fast switching the initial state does not change, and the main change comes from the Hamiltonian (5): $`𝒲_{fast}(0g)=H(g)H(0)_0`$, which for small $`g`$ leads to the same result as in Eq. (44). Using Eqs (27, 40, 44) one readily notices that there are temperatures, for which $`𝒲_{ad}(0g)<gW^{(1)}`$, implying that the cost for the switching is less than the gain due to coupling: $`W(0)+𝒲_{ad}(0g)<W(g)`$.
Conclusion.— Untill now it was believed that the presence of a direct interaction between the baths of a heat engine reduces its efficiency . The purpose of the present paper is to demonstrate that it can enhance the efficiency. Having changed the initial and final states, a direct coupling introduces, of course, both a change in work and dissipation. We show that the characteristic times of these two quantities can be well-separated. For times in the window (43) the work can be enhanced, though the dissipation is not yet relevant. This additional amount of work, which can be obtained from Eqs. (27, 30, 40), is provided by the modified final state of the baths.
Finally, we will briefly discuss related studies. Refs. consider the local thermodynamic efficiency of brownian motors and related models. The statement of this problem differs from the one considered by us, but it is interesting to mention that the role of a direct interaction between baths was studied also in this context .
|
warning/0005/quant-ph0005096.html
|
ar5iv
|
text
|
# Polarization dependence of four-wave mixing in a degenerate two-level system.
## I Introduction.
The generation of new optical fields is the most spectacular consequence of the intrinsic non-linearity of the light-matter interaction. Four-wave mixing (FWM), i.e. the generation of a fourth field as the result of the interaction of a material sample with three electromagnetic fields is the simplest non-linear optics generation process allowed in every material.
This paper is concerned with nearly degenerate four wave mixing (NDFWM) occurring when two of the three incident fields (designated as pump fields) have the same frequency $`\omega _1`$ and the third field (probe field) has an independent frequency $`\omega _2\omega _1+\delta `$ which is tunable around $`\omega _1`$. Of particular interest is the case where all fields are nearly resonant with the same optical atomic transition. In such case the spectroscopic study of the NDFWM signal as a function of the probe to pump frequency offset can provide useful information about the atomic dynamics. NDFWM was used by Rothberg and Bloembergen for studying collisional dynamics in sodium vapor. Boyd et al have studied theoretically the atomic polarization and the propagation of the probe and NDFWM waves in the case of a pure two-level system (PTLS) driven by a arbitrarily intense pump. Their work illustrate the critical influence of the atomic damping mechanism in the FWM process. Steel et al explored NDFWM in both open and closed PTLS. Berman and coworkers clearly established the suitability of NDFWM for the determination of all relevant relaxation parameters. Andersen and coworkers analyzed the NDFWM process in PTLS in terms of the dressed state picture. Their work stresses the importance of the initial atom-field state on the FWM yield.
In spite of the fact that the experiments on NDFWM generally involve degenerate energy levels, most experiments to date were carried under conditions where the level degeneracy does not play an essential role. In such cases, simplified theoretical models using two-level schemes were successfully used. Nevertheless, the role of the level degeneracy for NDFWM was clearly appreciated by Berman and coworkers who stressed the importance of the non-conservation in the evolution of the multipolar expansion terms of the atomic density matrix, for the observation of sub-natural resonances in the NDFWM signal spectrum. Their work include a detailed calculation, valid for degenerate atomic levels, of the nonlinear atomic polarization to the lowest order in the incident fields. Recently, a model allowing the numerical investigation of the complete response of a degenerate two-level system (DTLS) to first order in the probe field for an arbitrarily intense pump and arbitrary pump and probe polarizations was presented.
Recent experimental observations illustrate the need for a deeper understanding of the role of the level degeneracy in the non linear response of DTLS. An important example is provided by the spectroscopic response of alkaline atoms in a magneto-optical trap (MOT). In this system the atoms are submitted to the excitation of the trapping beams which are quasi resonant with a closed degenerate transition. The absorption and FWM spectra of magneto-optically trapped atoms present characteristic features resulting from the level degeneracy. Recently, the use of DTLS for efficient and highly selective NDFWM generation in an atomic vapor was reported.
The aim of this paper is to address the role of the level degeneracy in NDFWM on a closed transition between two degenerate atomic levels. The dependence of NDFWM generation on the intensity of the pumping field and the polarizations of the pump and probe will be analyzed. In the next section, NDFWM spectra calculated for an homogeneous ensemble of DTLS under different conditions are presented. The following section is devoted to the experimental observation of NDFWM in cold cesium atoms. Discussion of the experimental results in regard to the theoretical predictions follows.
## II Theoretical predictions.
The NDFWM generation in an homogeneous ensemble of DTLS is analyzed using the semiclassical procedure already presented in . In Fig. 1 the essential elements of this model are reminded. The two-level atoms are driven by a pump field of frequency $`\omega _1`$ of arbitrary intensity and polarization. The reduced pump field Rabi frequency is $`\mathrm{\Omega }_1`$. The driven atoms are probed by a weak field of frequency $`\omega _2`$ $`\omega _1+\delta `$ and arbitrary polarization. Both fields are nearly resonant with the atomic transition (transition frequency $`\mathrm{}\omega _A`$) between the ground level $`g`$ of total angular momentum $`F_g`$ and the excited level $`e`$ of total angular momentum $`F_e`$. Having in mind the experimental observation presented below, we will restrict our attention to closed atomic transitions with $`F_e=F_g+1`$. Level $`e`$ decays back into $`g`$ through spontaneous emission at rate $`\mathrm{\Gamma }`$. In addition to this relaxation mechanism a state independent decay rate $`\gamma `$ ($`\gamma \mathrm{\Gamma }`$) is assumed which represents the departure of the atoms from the interaction zone. This departure is compensated at steady state by the arrival (pumping term $`\gamma \rho _0`$) of “fresh” atoms isotropically distributed in the ground level. ($`\rho _0`$ is the system density matrix in the absence of any applied field). Since no other relaxation mechanism exist for the ground state, $`\gamma `$ effectively plays the role of a ground state relaxation rate. The calculation is exact to all orders in the pump field and to first order in the probe field. The spectra represent the variation of the squared modulus of the atomic polarization induced at the frequency $`\omega _32\omega _1\omega _2\omega _1\delta `$ as a function of the probe to pump frequency offset $`\delta `$. Since the calculation corresponds to the response of an homogeneous ensemble of atoms at rest, no propagation effects such as phase matching, spatial emission pattern of the generated wave or beam absorption (or amplification) are accounted for.
Before presenting the spectra calculated for DTLS, let us remind the main features of the NDFWM spectra for PTLS. Figure 2 represent the spectra of the NDFWM power as a function of $`\delta `$ for $`\mathrm{\Delta }=0`$ and $`\mathrm{\Delta }=2\mathrm{\Gamma }`$ ($`\mathrm{\Delta }\omega _A\omega _1`$) for different values of the pump field Rabi frequency $`\mathrm{\Omega }_0`$. We restrict ourselves to closed transition where spontaneous emission is the only relaxation mechanism. For all values of $`\mathrm{\Omega }_0`$, the spectra present two symmetric sidebands at $`\delta =\pm \mathrm{\Omega }`$ where $`\mathrm{\Omega }\sqrt{\mathrm{\Omega }_0^2+\mathrm{\Delta }^2}`$ is the generalized Rabi frequency. The width of these two peaks is determined by $`\mathrm{\Gamma }`$. When $`\mathrm{\Omega }_0\mathrm{\Delta },\mathrm{\Gamma }`$ a third peak develops at $`\delta =0`$ whose width is also determined by $`\mathrm{\Gamma }`$. Only when $`\mathrm{\Omega }\mathrm{\Gamma }`$ the central peak overcomes the two sidebands. As discussed in , the relative amplitude of peak at $`\delta =0`$ is determined by the coherence relaxation rate. In the present case this rate equals $`\mathrm{\Gamma }/2`$ since only radiative decay is assumed. When the transition is open and when the escaping rates out of the two-level systems are different for the upper and lower level, an additional peak appears at $`\delta =0`$ with a width determined by the lower level relaxation rate. Finally let us mention that in PTLS the maximum NDFWM power is obtained for $`\mathrm{\Omega }_0\mathrm{\Gamma }`$.
We examine now the NDFWM power spectra for a DTLS with $`F_g=1`$ and $`F_e=2`$ (no magnetic field present). Three polarization cases were considered: circular and equal pump and probe polarizations, parallel linear pump and probe polarizations and linear and perpendicular pump and probe polarizations. The corresponding level schemes are shown in Fig. 3 where an appropriate choice of the quantization axis is made in each case.
Figure 4 shows the spectra calculated for pump and probe fields with the same circular polarization ($`\sigma ^+`$). As expected, due to the optical pumping of the population towards the Zeeman sublevel in $`g`$ and $`e`$ with the highest magnetic quantum number (see Fig. 3 a), the system behaves as a PTLS in the limit of large $`\mathrm{\Omega }_1`$. However, striking differences appear for $`\mathrm{\Omega }_1\mathrm{\Gamma }`$ in which case the Zeeman optical pumping is only partial. Notice the presence at $`\delta =0`$ of narrow resonances: a peak for $`\mathrm{\Omega }_1=\mathrm{\Gamma }`$, $`\mathrm{\Delta }=2\mathrm{\Gamma }`$ and a dip for $`\mathrm{\Omega }_1=\mathrm{\Gamma }`$, $`\mathrm{\Delta }=0`$ and $`\mathrm{\Omega }_1=2\mathrm{\Gamma }`$, $`\mathrm{\Delta }=2\mathrm{\Gamma }`$. The width of this narrow resonance is determined by $`\gamma `$. In this configuration, the pump and probe fields interact with pairs of Zeeman sublevels of $`e`$ and $`g`$ with $`m_e=m_g+1`$ ($`m_i`$ is a magnetic quantum number). Coupling between different pairs occurs only via spontaneous emission. NDFWM signal arises from the coherent contribution of different Zeeman sublevels pairs excited by the field. Due to spontaneous emission, a given pair is not a close system in which case, a narrow structure (width $`\gamma `$) is expected at $`\delta =0`$. As observed in Fig. 4 the sign of this narrow feature may vary depending on the pump field intensity and detuning. It is the result of the quantum interference of the contributions from the different Zeeman sublevels pairs to the non linear atomic polarization. A similar situation occurs in the case of parallel linear polarizations discussed next.
Figure 5 refers to the spectra calculated for pump and probe fields with parallel linear polarizations. Unlike the previous case, in the limit of large $`\mathrm{\Omega }_1`$, the system does not behave as a PTLS. Instead, taking as the quantization axis the direction of the common pump and probe polarization, it should be seen as three ($`2F_g+1`$) two-level systems linked through spontaneous emission (Fig. 3 b). Each two-level system consist of a two Zeeman sublevels with $`m_g=m_e`$. Since the strength of the atom field coupling dependents only on $`\left|m_g\right|`$, two different values of the pump Rabi frequency occur in the present example. This explains the splitting observed in the two sidebands for large $`\mathrm{\Omega }_1`$. At low pump intensities a narrow resonance (width determined by $`\gamma `$) appears at $`\delta =0`$ for $`\mathrm{\Delta }=2\mathrm{\Gamma }`$. Also a narrow dip is present at $`\delta =0`$ for $`\mathrm{\Delta }=0`$ and $`\mathrm{\Omega }_1<\mathrm{\Gamma }`$ (not shown in the figure). The appearance of narrow features for zero pump to probe detuning with a width determined by the effective ground state relaxation rate was predicted for open two-level systems. In the present case this feature should be seen as the consequence of the population transfer, due to spontaneous emission, between pairs of levels connected by the fields. A detailed look into the evolution of this narrow resonance with the pump field power (for $`\mathrm{\Delta }=2\mathrm{\Gamma }`$) is presented in Fig. 6. The resonance at $`\delta =0`$ is negative (dip) for $`\mathrm{\Omega }_1>0.4\mathrm{\Gamma }`$. This result is to be compared with the contribution to the nonlinear atomic polarization arising from the different pairs of Zeeman sublevels with $`m_e=m_g`$. These contributions can be extracted from the calculation after identification of the density matrix coefficients corresponding to a given value of $`\left|m_i\right|`$ ($`i=e,g`$). Figs. 7 and 8 represent the square modulus of the nonlinear atomic polarization contributions arising from the Zeeman sublevels pairs with $`\left|m_g\right|=0`$ and $`\left|m_g\right|=1`$ respectively. Only in the case of $`\left|m_g\right|=0`$ and for the largest value of $`\mathrm{\Omega }_1`$ the central resonance is opposite to the lateral sidebands. In all other cases this resonance has the same sign than the lateral sidebands. This is always the case for $`\left|m_g\right|=1`$. Consequently, the dip observed at $`\delta =0`$ in the NDFWM spectra in Fig. 6 is the result of (destructive) quantum interference between the two contributions coming from $`\left|m_g\right|=0,1`$. Indeed, the two contributions have opposite phase over all the considered range of $`\mathrm{\Omega }_1`$.
The NDFWM spectra obtained for perpendicular linear pump and probe polarizations (Fig. 9) are rather different to the ones presented above. For all the considered pump field intensity range the spectra are dominated by the features occurring around $`\delta =0`$. In fact for $`\mathrm{\Omega }_12\mathrm{\Gamma }`$ the spectral sidebands are barely visible and the spectra is mainly composed by a large central resonance with a width of the order of $`\gamma `$ . This configuration produces for $`\mathrm{\Omega }_1\mathrm{\Gamma }`$ the largest NDFWM yield. For increasing $`\mathrm{\Omega }_1`$ the central peak splits into two components. Also in this case the spectra are asymmetric for $`\mathrm{\Delta }0`$. The main features of these spectra can be qualitatively understood using the dressed states picture of the degenerate atomic system in the presence of the pump field. A similar analysis was carried in (Appendix) to examine the probe absorption spectra of driven DTLS in this configuration. The dressed-state energy level scheme for a $`F_g=1F_e=2`$ transition driven by a $`\pi `$ polarized pump is presented in Fig. 10 following the conventions adopted in . NDFWM resonances are expected to occur when the probe field is resonant with a transitions between dressed levels that are coupled to the pump field photons. Four different probe frequencies satisfy this condition. The corresponding values of $`\delta `$ (see Table II in ) are indicated with solid arrows in the two lower spectra of Fig. 9. They correspond to the main features in the NDFWM spectra. From a simple dressed-state analysis, no resonance is expected to occur for the probe frequencies corresponding to transitions ending in the $`|1\mathrm{"},N`$ dressed-atom levels since these states are not coupled to the pump photons due to the $`m_e=m_g`$ selection rule valid for a $`\pi `$ polarized pump. Nevertheless small features appear in the calculated spectra at these positions (dashed arrows in Fig. 9). They are due to non-secular terms usually neglected in the dressed atom approach.
## III Experimental observation of NDFWM in cold cesium.
The observation of the NDFWM spectra was performed in a sample of cold cesium atoms produced in a magneto-optical trap (MOT). Light from a Ti:sapphire laser, nearly resonant with the cesium cycling transition $`6S_{1/2}(F=4)6P_{3/2}(F=5)`$, was employed both for the trapping of the atoms and for the investigation of the NDFWM. The experimental setup is shown schematically in Fig. 11. The frequency of the Ti:sapphire laser is red-detuned by approximately two natural linewidth $`(\mathrm{\Gamma }/2\pi =5.3MHz)`$. A repumping diode laser, not shown in the figure, recycles the population lost to the $`6S_{1/2}(F=3)`$ ground state. We use a backwards FWM configuration, where the two counter-propagating pump fields, the forward (F) and backward (B), have the same frequency and the same linear polarization. The probe beam (P), is linearly polarized and makes a small angle ($`\theta =4^o`$) with the pumping beams. The probe beam has its frequency scanned around the frequency of the pump beams with the help of two acousto-optic modulators as shown in Fig. 11. The relative polarization between the pump and the probe fields is controlled by a half-wave plate. The generated (nearly) phase conjugated signal (PC), which propagates in opposite direction with respect to beam P, is reflected out of a $`50/50`$ beam splitter and detected by a fast photodiode. The trapping beams are switched on and off by a mechanical chopper with a transmission duty cycle of $`95\%`$. The number of cold atoms was estimated by measuring the absorption of the probe beam and is of order of $`10^7`$. The NDFWM spectra were recorded within a $`1ms`$ time interval during which the trapping beams were blocked and the quadrupole magnetic field was turned off. Each of the F and B pump beams have an intensity of $`7mW/cm^2`$. The probe beam intensity is approximately equal to $`0.7mW/cm^2`$. No significant modification of the spectra was observed at lower probe beam intensities. The maximum NDFWM power generated was of the order of $`1\mu W`$. Typical spectra, recorded as a function of the pump to probe frequency offset $`\delta `$, are shown in Figs. 12a) and 13a) for the probe beam polarization respectively parallel and perpendicular to the pump beam polarization.
## IV Discussion.
The comparison of the observed spectra with the theoretical predictions presented above is not direct. While the calculation applies to a homogeneous ensemble of atoms at rest, the PC signal is the result of cooperative emission of atoms under different excitation conditions. In the experiment the pump field, at a given position in the sample, is due to the combined incidence of the F and B beams. The counterpropagating geometry of these beams produces a standing wave, consequently, the pump field intensity is spatially modulated. In general the pump field polarization may be spatial dependent. However, in the present study we have restricted ourselves to the case where the beams F and B have the same linear polarization and consequently no spatial variation of the polarization occurs. Also, the PC signal is sensitive to propagation effects in the atomic sample which are responsible for the phase matching condition and for possible spatial dependent amplification or depletion of the NDFWM through the medium. Finally, one should remind that the phase matching condition imposes constraints on the polarization components that may be present in the PC beam (only transverse components). However, in the two polarization cases considered in the experiment, linear parallel and linear perpendicular pump and probe polarizations, due to symmetry, the nonlinear atomic polarization should be parallel to the probe polarization (in the absence of magnetic field). In consequence, all polarization components of the NDFWM field are able to propagate along the PC beam.
In order to allow the comparison between the theoretical prediction and the experimental observation, we have incorporated into the calculation the intensity distribution of the pump field in the atomic sample. Propagation effects were not considered. Such approach, expected to be valid for an optically thin sample, is only approximative in our case since the peak probe absorption is around $`50\%`$. Assuming that the F and B beams produce a perfect standing wave, we have considered a sine wave distribution of $`\mathrm{\Omega }_1`$ in the interval $`0\mathrm{\Omega }_1\mathrm{\Omega }_{1MAX}`$. For each value of $`\mathrm{\Omega }_1`$, the NDFWM field was calculated assuming that a steady state is reached. The total NDFWM field was taken as the sum of the contributions for each $`\mathrm{\Omega }_1`$. Finally the NDFWM power correspond to the square modulus of the total field. The calculations were performed for an $`F_g=4F_e=5`$ transition with $`\mathrm{\Delta }=2\mathrm{\Gamma }`$ (corresponding to the experimental conditions) and $`\gamma =0.01\mathrm{\Gamma }`$. No magnetic field was considered. The value of $`\mathrm{\Omega }_{1MAX}`$ was adjusted to fit the experimental spectra. The best agreement was obtained for $`\mathrm{\Omega }_{1MAX}=18\mathrm{\Gamma }`$.
Figure 12b) represents the calculated spectrum for linear and parallel pump and probe polarizations with the parameters corresponding to the experimental conditions. The spectrum is dominated by the two sidebands. The width of the sidebands is mainly due to the inhomogeneity of the Rabi frequency. The largest values of $`\mathrm{\Omega }_1`$ are responsible for the central peak presenting a width of the order of $`\mathrm{\Gamma }`$ while the atoms corresponding to $`\mathrm{\Omega }_1\mathrm{\Gamma }`$ are responsible for the narrow dip present at $`\delta =0`$. The main features of the experimental spectrum are well reproduced. Some differences appear in the amplitude and shape of the narrow resonance around $`\delta =0`$. According to the theoretical considerations presented above, this narrow resonance is the consequence of the coupling between different Zeeman sublevels pairs through spontaneous emission. Its width is governed by the time of flight relaxation rate $`\gamma `$ which effectively plays the role of a ground-state decay rate. For cold atoms, the average time of flight across the exciting beams is rather long (more than $`1ms`$ for a $`1mm`$ diameter beam) and the corresponding value of $`\gamma <3\times 10^5\mathrm{\Gamma }`$ too small to account for the observed resonance. However, for typical MOT temperatures, the average time os flight across one spatial period of the stationary wave produced by the F and B beams is three orders of magnitude shorter and correspond to $`\gamma 10^2\mathrm{\Gamma }`$ as assumed in the calculation. At this point one should notice that a short travelling time across the standing wave pattern of the pump field in not compatible with our assumption of a steady state reached for each value of $`\mathrm{\Omega }_1`$. This suggests that a more sophisticated theoretical approach, incorporating the atomic motion, would be more appropriate for a precise description of the spectra. In addition to the finite interaction time, another mechanism that can affect the narrow resonance around $`\delta =0`$ is the possible leakeage out of the closed two-level transition. In our case, this may occur though non resonant excitation of other excited state hyperfine levels. However the corresponding rate can be estimated to be smaller than $`10^3\mathrm{\Gamma }`$. Also, calculations carried for open transitions give spectral shapes very different from those observed in Fig. 12.
The comparison between observed and calculated spectra in the case of linear and perpendicular pump and probe polarizations are presented in Fig. 13. The main features of the spectra are well reproduced by the calculation. Notice the significant increase in the maximum NDFWM yield with respect to the previous case. The spectrum is in the present case dominated by the narrow resonance at $`\delta =0`$. The width of this resonance is also given at low pump intensities by $`\gamma `$. As already pointed in and , this configuration provides the largest NDFWM yield. The difference in the spectral profile existing between Figs. 12 and 13 constitutes a clear demonstration of the essential role of the Zeeman degeneracy and optical polarizations in the NDFWM process.
## V Conclusions.
The process of NDFWM in a closed atomic transitions with $`F_e=F_g+1`$ has been examined both theoretically and experimentally. Large differences in the spectra are observed for different choices of the exciting fields polarizations revealing the crucial role of the internal level structure on the non-linear process. The spectra present distinct features which are determined by the different relaxation rates and characteristic frequencies of the system ($`\mathrm{\Gamma },\gamma ,\mathrm{\Omega }_1,\mathrm{\Delta }`$). NDFWM was observed in a cold sample of cesium atoms in a PC experiment for two different choices of the pump and probe polarizations. In spite of the difference existing between the experimental conditions and the assumptions of the theory a good agreement between calculated and observed spectra was obtained. Nevertheless, some features of the spectra, associated to the longest relaxation processes, indicate the need for a more detailed theoretical approach including the atomic motion and spatial field distribution.
## VI Acknowledgments.
The authors acknowledge fruitful discussions with J.R. Rios Leite. This work was supported by CNPq (PRONEX), CAPES and FINEP (Brazilian agencies) and by CSIC, CONYCIT and PEDECIBA (Uruguayan agencies).
|
warning/0005/hep-ph0005258.html
|
ar5iv
|
text
|
# CP Violation – A Brief Review 1footnote 11footnote 1Invited talk presented at 2nd Tropical Workshop in Particle Physics and Cosmology, San Juan, Puerto Rico, May 1–6, 2000, proceedings to be published by AIP. Enrico Fermi Institute Report No. EFI 2000-16, hep-ph/0005258.
## I Introduction
Fundamental discrete symmetries have provided both guidance and puzzles in our evolving understanding of elementary particle interactions. The discrete symmetries C (charge inversion), P (parity, or space reflection), and T (time reversal) are preserved by strong and electromagnetic processes, but violated by weak decays. For a brief period of several years, it was thought that the products CP and T were preserved by all processes, but that belief was shattered with the discovery of CP violation in neutral kaon decays in 1964 CCFT . The product CPT seems to be preserved, as is expected in local Lorentz-invariant quantum field theories CPT .
Since 1973 we have had a candidate theory of CP violation KM , based on phases in the coupling constants describing the weak charge-changing transitions of quarks. These couplings are described by the unitary $`3\times 3`$ Cabibbo-Kobayashi-Maskawa (CKM) KM ; Cab matrix. This theory has survived a qualitative test with the establishment of direct CP violation in neutral kaon decays E832 ; NA48 . It is well on its way to being tested in a wealth of $`B`$ decay processes. Will these tests be passed? What are the implications in either case? What will we learn about the “other” manifestation of CP asymmetry in nature, the baryon asymmetry of the Universe? This brief review is devoted to these questions.
In Section II we introduce the discrete symmetries P, T, and C by the example of Maxwell’s equations. Section III is devoted to the history and present status of CP violation and related phenomena in kaon decays, while Section IV deals with results and prospects for $`B`$ mesons. Some future measurements are discussed in Section V. The baryon number of the Universe and its relation to CP violation are treated briefly in Section VI, while Section VII concludes.
## II Discrete symmetries
Maxwell’s equations in vacuum provide a convenient framework for illustrating the action of discrete symmetries, since each term in each equation must transform similarly.
Under P, we have $`𝐄(𝐱,t)𝐄(𝐱,t)`$, $`𝐁(𝐱,t)𝐁(𝐱,t)`$, $``$, $`𝐣(𝐱,t)𝐣(𝐱,t)`$, i.e., electric fields change in sign while magnetic fields do not, and currents change in direction. Under time reversal, $`𝐄(𝐱,t)𝐄(𝐱,t)`$, $`𝐁(𝐱,t)𝐁(𝐱,t)`$, $`/t/t`$, $`𝐣(𝐱,t)𝐣(𝐱,t)`$, i.e., magnetic fields change in sign while electric fields do not, since directions of currents are reversed. Under C, $`𝐄(𝐱,t)𝐄(𝐱,t)`$, $`𝐁(𝐱,t)𝐁(𝐱,t)`$, $`\rho (𝐱,t)\rho (𝐱,t)`$, $`𝐣(𝐱,t)𝐣(𝐱,t)`$, i.e., both electric and magnetic fields change sign, since their sources $`\rho `$ and $`𝐣`$ change sign. Finally, under CPT, space and time are inverted but electric and magnetic fields retain their signs: $`𝐄(𝐱,t)𝐄(𝐱,t)`$, $`𝐁(𝐱,t)=𝐁(𝐱,t)`$.
The behavior of the Maxwell equations under P, T, C, and CPT is summarized in Table 1. Each term behaves as shown. It is interesting that a fundamental term in the Lagrangian behaving as $`𝐄𝐁`$, while Lorentz covariant, violates P and T. The strong suppression of such a term (as evidenced by the small value of the neutron electric dipole moment) is known as the strong CP problem SCPrev , and, although of fundamental importance, will not be discussed further here.
## III CP symmetry for kaons
### III.1 $`K\pi \pi `$ decays
While some neutral particles (such as $`\gamma `$, $`Z^0`$, and $`\pi ^0`$) are equal to their antiparticles, others (such as the neutron) are not. The $`K^0`$, discovered in cosmic radiation in the late 1940’s RB , is one such particle. It is characterized by an additive quantum number $`S=1`$, strangeness, introduced GN in order to explain its strong production (which conserves strangeness) and weak decay (which does not). The antiparticle of $`K^0`$, the $`\overline{K^0}`$, has $`S=1`$. Since strangeness is violated in decays, one must appeal to discrete symmetries to describe the linear combinations of $`K^0`$ and $`\overline{K^0}`$ corresponding to states of definite mass and lifetime. These states are
$$K_1=\frac{K^0+\overline{K^0}}{\sqrt{2}},K_2=\frac{K^0\overline{K^0}}{\sqrt{2}}.$$
(1)
The $`K_1`$ is permitted to decay to $`\pi \pi `$ and thus should be short-lived, while the $`K_2`$ is forbidden to decay to $`\pi \pi `$, must instead decay to $`3\pi `$, $`\pi \mathrm{}\nu _{\mathrm{}}`$, etc., and thus will be longer-lived. Indeed, the short-lived neutral kaon ($`K_1`$) lives for only 0.089 ns, while the long-lived neutral kaon ($`K_2`$) lives for 52 ns, nearly a factor of 600 longer.
The original argument by Gell-Mann and Pais GP , based in 1955 on C and P conservation, was recast in 1957 in terms of the product CP CPK , to correspond to the newly formulated CP-invariant theory of the weak interactions. The $`K^0`$ and $`\overline{K^0}`$ have spin zero. A spin-zero final state of $`\pi \pi `$ has CP eigenvalue equal to $`+1`$. Thus, if CP is conserved, it is the CP-even linear combination of $`K^0`$ and $`\overline{K^0}`$ which decays to $`\pi \pi `$. With a phase convention such that $`CP|K^0=|\overline{K^0}`$, this is just the combination $`K_1`$. The Gell-Mann–Pais proposal was soon confirmed KL by the discovery of the predicted long-lived particle corresponding to $`K_2`$.
Similar behavior is encountered in many cases of degenerate systems, such as two coupled pendula BW or a drum-head in its first excited state. In the latter case, the drum has two degenerate modes, each with one nodal line corresponding to a diameter, which will be orthogonal to one another if the corresponding nodal lines are perpendicular to each other. Consider two equally valid bases:
* (B1) Diagonal nodal lines point to the upper right ($`R`$) and the upper left ($`L`$).
* (B2) The nodal lines are horizontal ($`H`$) and vertical ($`V`$).
We can draw the analogy $`RK^0`$, $`L\overline{K^0}`$. Suppose, now, that a fly alights on the bottom edge of the drum head, such that it sits on the nodal line of the $`V`$ mode. Then the modes $`V`$ and $`H`$ are split from one another. The mode $`H=(R+L)/\sqrt{2}`$ which couples to the fly will shift in mass and lifetime. It is analogous to $`K_1`$ and the fly is analogous to the $`\pi \pi `$ system. The mode $`V`$ is unaffected by the fly. It is analogous to $`K_2`$.
In 1964, Christenson, Cronin, Fitch, and Turlay CCFT , using a spark chamber exposed to a beam of long-lived neutral kaons, found that these particles indeed did decay to $`\pi \pi `$. For many years this phenomenon could be described in terms of a single parameter $`ϵ`$, such that the states of definite mass and lifetime become
$$K_1K_S(\mathrm{`}\mathrm{`}\mathrm{short}\mathrm{"})K_1+ϵK_2,K_2K_L(\mathrm{`}\mathrm{`}\mathrm{long}\mathrm{"})K_2+ϵK_1,$$
(2)
with $`|ϵ|2\times 10^3`$, and Arg($`ϵ)\pi /4`$. Confirmation of this description was provided by the rate asymmetry in the decays $`K_L\pi ^\pm \mathrm{}^{}\nu _{\mathrm{}}`$, which measures Re $`ϵ`$. But what is the source of $`ϵ`$?
One possibility was suggested almost immediately by Wolfenstein SW : A new “superweak” $`|\mathrm{\Delta }S=2|`$ interaction could mix $`K^0=d\overline{s}`$ and $`\overline{K^0}=s\overline{d}`$ (where $`d`$ and $`s`$ denote quarks) without any other observable consequences. This theory would imply, for example, that no difference in the ratio of CP-violating and CP-conserving amplitudes would arise when comparing $`\pi ^+\pi ^{}`$ and $`\pi ^0\pi ^0`$ final states.
A new opportunity for generating not only $`ϵ`$ but other CP-violating effects as well arises when there are at least three quark families, as first proposed by Kobayashi and Maskawa KM . Loop diagrams inducing the transition $`d\overline{s}s\overline{d}`$ involving internal lines of $`W^+W^{}`$ and $`u,c,t`$ quarks and antiquarks can lead to $`ϵ0`$ when the coupling constants are complex. With three quark families, one cannot redefine phases of quarks so that all the couplings are real. Some other consequences of the Kobayashi-Maskawa theory will be mentioned presently.
The time-dependence of the two-component $`K^0`$ and $`\overline{K^0}`$ system is governed by a $`2\times 2`$ mass matrix $``$ (for reviews see Revs ):
$$i\frac{}{t}\left[\begin{array}{c}K^0\\ \overline{K^0}\end{array}\right]=\left[\begin{array}{c}K^0\\ \overline{K^0}\end{array}\right],$$
(3)
where $`=Mi\mathrm{\Gamma }/2`$, and $`M`$ and $`\mathrm{\Gamma }`$ are Hermitian matrices. The eigenstates are, approximately,
$$K_SK_1+ϵK_2,K_LK_2+ϵK_1,$$
(4)
corresponding to the eigenvalues $`\mu _{S,L}=m_{S,L}i\gamma _{S,L}/2`$, with
$$ϵ\frac{\mathrm{Im}(\mathrm{\Gamma }_{12}/2)+i\mathrm{Im}M_{12}}{\mu _S\mu _L}.$$
(5)
Using both data and the magnitude of CKM matrix elements one can show Revs that the second term dominates. Since the mass difference $`m_Lm_S`$ and width difference $`\gamma _S\gamma _L`$ are nearly equal, the phase of $`\mu _L\mu _S`$ is about $`\pi /4`$, so that the phase of $`ϵ`$ is also $`\pi /4`$ (mod $`\pi `$).
It is easy to emulate the CP-conserving neutral kaon system in table-top demonstrations of systems with two degenerate states, such as the pair of coupled pendula mentioned above BW . The demonstration of CP violation is harder, requiring systems that emulate Im($`M_{12})0`$ or Im($`\mathrm{\Gamma }_{12})0`$. One can couple two identical resonant circuits “directionally” to each other so that the energy fed from circuit 1 to circuit 2 differs from that fed in the reverse direction TTTV . Devices with this property utilize Faraday rotation of the plane of polarization of radio-frequency waves. More recently, it was realized RS that this asymmetric coupling is inherent in the equations of motion of a spherical (or “conical”) pendulum in a rotating coordinate system, giving rise to the precession of the plane of oscillation of the Foucault pendulum. In either case the analogy actually deepens the mystery of CP violation, since the CP-violating effect is imposed, so to speak, “from the outside,” using a magnetic field in the case of directional couplers or a rotating coordinate frame in the case of the Foucault pendulum.
To return to the CKM matrix, we have the following parameterization suggested by Wolfenstein WP :
$$V\left[\begin{array}{ccc}V_{ud}& V_{us}& V_{ub}\\ V_{cd}& V_{cs}& V_{cb}\\ V_{td}& V_{ts}& V_{tb}\end{array}\right]=\left[\begin{array}{ccc}1\frac{\lambda ^2}{2}& \lambda & A\lambda ^3(\rho i\eta )\\ \lambda & 1\frac{\lambda ^2}{2}& A\lambda ^2\\ A\lambda ^3(1\rho i\eta )& A\lambda ^2& 1\end{array}\right],$$
(6)
where $`\lambda =\mathrm{sin}\theta _\mathrm{C}0.22`$ describes strange particle decays. Here $`\theta _\mathrm{C}`$ is the Gell-Mann–Lévy–Cabibbo Cab ; GL angle, originally introduced to preserve the universal strength of the hadronic weak current. The unitarity of the CKM matrix, $`V^{}=V^1`$, is the modern way of implementing this requirement.
We learn $`|V_{cb}|=A\lambda ^20.039\pm 0.003`$ from the dominant decays of $`b`$ quarks, which are to charmed quarks CKMrevs . (We have expanded errors somewhat in comparison with those quoted in some reviews Parodi . The dominant source of error in many cases is theoretical.) Similarly, charmless $`b`$ decays give $`|V_{ub}/V_{cb}|=0.090\pm 0.025=\lambda (\rho ^2+\eta ^2)^{1/2}`$, leading to a constraint on $`\rho ^2+\eta ^2`$.
As a result of the unitarity of the CKM matrix, the quantities $`V_{ub}^{}/A\lambda ^3=\rho +i\eta `$, $`V_{td}/A\lambda ^3=1\rho i\eta `$, and 1 form a triangle in the $`(\rho ,\eta )`$ plane (Fig. 1). The angles opposite these sides are, respectively, $`\beta =\mathrm{Arg}(V_{td})`$, $`\gamma =\mathrm{Arg}(V_{ub}^{})`$, and $`\alpha =\pi \beta \gamma `$. We still do not have satisfactory limits on the angle $`\gamma `$ (equivalently, on the magnitude of the side $`V_{td}`$) of this “unitarity triangle.” Further information comes from the following constraints (see JRlatt for more details):
1. The magnitude of $`ϵ`$ constrains mainly the imaginary part of $`V_{td}^2`$, which is proportional to $`\eta (1\rho )`$, since the top quark dominates the loop diagram giving rise to $`K^0`$$`\overline{K^0}`$ mixing. A correction due to charmed quarks changes the 1 to 1.44, with the result $`\eta (1.44\rho )=0.51\pm 0.18`$.
2. We have taken the amplitude for mixing of the neutral $`B^0`$ meson with its antiparticle $`\overline{B^0}`$ to be $`\mathrm{\Delta }m_d=0.473\pm 0.016`$ ps<sup>-1</sup> BOSC . The subscript $`d`$ denotes the light quark in the $`B^0`$. Taking the matrix element of the four-quark operator inducing the relevant $`\overline{b}d\overline{d}b`$ transition to be $`f_B\sqrt{B_B}=200\pm 40`$ MeV, we find a constraint on $`|V_{td}|`$ which amounts to $`|1\rho i\eta |=1.01\pm 0.21`$.
3. We have used the following lower limit for mixing of the strange $`b`$ meson $`B_s=\overline{b}s`$ with its antiparticle: $`\mathrm{\Delta }m_s>14.3`$ ps<sup>-1</sup> (95% c.l.) BOSC ; Blay . Since the relevant CKM elements (including $`|V_{ts}|=A\lambda ^2`$) are fairly well known, this result serves mainly to constrain the combination of hadronic parameters $`f_{B_s}\sqrt{B_{B_s}}`$ and hence, through the assumption $`[f_{B_s}\sqrt{B_{B_s}}]/[f_B\sqrt{B_B}]<1.25`$ JRFM , yields the bound $`|V_{ts}/V_{td}|>4.3`$ or $`|1\rho i\eta |<1.05`$.
The resulting limits on $`(\rho ,\eta )`$ are a roughly rectangular region bounded on the left by $`|1\rho i\eta |<1.05`$, on the top and bottom by $`0.3<(\rho ^2+\eta ^2)^{1/2}<0.52`$, and on the right by $`|1\rho i\eta |>0.8`$. Only a small region is excluded by the bound arising from the parameter $`ϵ`$: $`\eta (1.44\rho )>0.33`$. Even without this bound, the case of real CKM matrix elements ($`\eta =0`$), i.e., a superweak origin for $`ϵ`$, is disfavored. The boundaries of this region give rise to the minimum and maximum values of $`\alpha ,\beta ,\gamma `$ shown in Table 2. These bounds imply
$$0.71<\mathrm{sin}2\alpha <0.59,0.59<\mathrm{sin}2\beta <0.89,0.54<\mathrm{sin}^2\gamma <1$$
(7)
for quantities which are measurable in $`B`$ decays (see below). The allowed values of $`(\rho ,\eta )`$ are $`(0.14\pm 0.15,0.38\pm 0.13)`$.
The Kobayashi-Maskawa theory predicts small differences in CP-violating decays to pairs of charged and neutral pions. These arise in the following way.
1. “Tree” amplitudes are governed by $`\overline{s}\overline{u}u\overline{d}`$. Since this subprocess has three nonstrange quarks in the final state, it contributes to both $`\mathrm{\Delta }I=1/2`$ and $`\mathrm{\Delta }I=3/2`$ transitions, and hence to both $`I_{\pi \pi }=0`$ and $`I_{\pi \pi }=2`$ final states. The corresponding CKM matrix elements are real, so these amplitudes do not have a weak phase.
2. “Penguin” amplitudes involve a transition $`\overline{s}\overline{d}`$ with internal $`W`$ and $`u,c,t`$ lines and emission or absorption of a gluon. The subprocess has only one nonstrange quark in the final state so it contributes only to $`\mathrm{\Delta }I=1/2`$ transitions and hence only to the $`I_{\pi \pi }=0`$ final state. Because of the presence of all three $`Q=2/3`$ quarks in internal lines, these amplitudes have a weak phase.
As a consequence of the different isospin structure and weak phases of the tree and penguin amplitudes, the $`I_{\pi \pi }=0`$ and $`I_{\pi \pi }=2`$ amplitudes thus acquire different weak phases, leading to a small difference from unity of the ratio
$$R\frac{\mathrm{\Gamma }(K_L\pi ^+\pi ^{})/\mathrm{\Gamma }(K_S\pi ^+\pi ^{})}{\mathrm{\Gamma }(K_L\pi ^0\pi ^0)/\mathrm{\Gamma }(K_S\pi ^0\pi ^0)}=1+6\mathrm{Re}\frac{ϵ^{}}{ϵ},$$
(8)
where $`ϵ^{}`$ is related to the imaginary part of the ratio of the $`I_{\pi \pi }=2`$ and $`I_{\pi \pi }=0`$ amplitudes. The ratio $`ϵ^{}/ϵ`$ acquires an important term proportional to the CKM parameter $`\eta `$ from the penguin contribution. This term is partially canceled by an “electroweak penguin” in which the gluon mentioned above is replaced by a virtual photon or $`Z`$, whose isospin-dependent couplings to quarks induce $`\mathrm{\Delta }I=3/2`$ contributions. $`ϵ^{}/ϵ`$ is expected to be nearly real. Its magnitude was estimated by one group Buras to be a few parts in $`10^4`$, with a broad and somewhat asymmetric probability distribution extending from slightly below zero to above $`2\times 10^3`$. Some other estimates, discussed in Refs. K99 , permit higher values.
The most recent experiments on Re($`ϵ^{}/ϵ`$) are summarized in Table 3. A scale factor PDG of 1.86 is included in the error of the average to account for the large spread in quoted results. The value of $`ϵ^{}/ϵ`$ is non-zero, with a magnitude in the ballpark of estimates based on the Kobayashi-Maskawa theory. The fact that it is larger than some theoretical estimates is not a serious problem, given that we still cannot account reliably for the large enhancement of $`\mathrm{\Delta }I=1/2`$ amplitudes with respect to $`\mathrm{\Delta }I=3/2`$ amplitudes in CP-conserving $`K\pi \pi `$ decays.
### III.2 Other rare kaon decays
A CP- or T-violating angular asymmetry in $`K_L\pi ^+\pi ^{}e^+e^{}`$ has recently been reported KTeVa ; NA48a . With a final state consisting of four distinct particles, using the three independent final c.m. momenta, one can construct a T-odd observable whose presence is signaled by a characteristic distribution in the angle $`\varphi `$ between the $`\pi ^+\pi ^{}`$ and $`e^+e^{}`$ planes.
The asymmetry in $`\mathrm{sin}\varphi \mathrm{cos}\varphi `$ reported in Ref. KTeVa is $`(13.6\pm 2.5\pm 1.2)\%`$. It arises from interference between two processes. (1) The $`K_L`$ decays to $`\pi ^+\pi ^{}`$ with an amplitude $`ϵ`$. This process is CP-violating. One of the pions then radiates a virtual photon which internally converts to $`e^+e^{}`$. (2) The CP-odd state $`K_2`$ can decay directly to $`\pi ^+\pi ^{}\gamma `$ via a weak magnetic dipole transition. This process is CP-conserving.
The decay $`K_L\mu ^+\mu ^{}\gamma `$ has recently been studied with sufficiently high statistics to permit a greatly improved measurement of the virtual-photon form factor in $`K_L\gamma ^{}\gamma `$ BQ . This measurement is useful in estimating the long-distance contribution to the real part of the amplitude in $`K_L\gamma ^{()}\gamma ^{()}\mu ^+\mu ^{}`$, which in turn allows one to limit the short-distance contribution to $`K_L\mu ^+\mu ^{}`$. Since this contribution involves loops with virtual $`W`$’s and $`u,c,t`$ quarks, useful bounds on CKM matrix elements can be placed. Preliminary results BQ indicate $`\rho >0.2`$, the best limit so far from any process involving kaons.
Several neutral-current processes involving $`K\pi +(\mathrm{lepton}\mathrm{pair})`$ can shed further light on the Kobayashi-Maskawa theory of CP violation BuK .
1. The decay $`K^+\pi ^+\nu \overline{\nu }`$ is sensitive primarily to $`|V_{td}|`$, with a small charm correction, and so constrains the combination $`|1.4\rho i\eta |`$. The predicted branching ratio is roughly
$$(K^+\pi ^+\nu \overline{\nu })10^{10}\left|\frac{1.4\rho i\eta }{1.4}\right|^2,$$
(9)
For $`0\rho 0.3`$ one then predicts (see BuK ) $`(K^+\pi ^+\nu \overline{\nu })=(0.8\pm 0.2)\times 10^{10}`$, with additional uncertainties associated with the charmed quark mass and the magnitude of $`V_{cb}`$. A measurement of this branching ratio with an accuracy of 10% is of high priority in constraining $`(\rho ,\eta )`$ further.
The Brookhaven E787 Collaboration has reported one event with negligible background E787 , corresponding to
$$(K^+\pi ^+\nu \overline{\nu })=(1.5_{1.2}^{+3.4})\times 10^{10}.$$
(10)
More data are expected from the final stages of analysis of this experiment, as well as from a future version (Brookhaven E949) with improved sensitivity.
2. The decays $`K_L\pi ^0\mathrm{}^+\mathrm{}^{}`$ are expected to be dominated by CP-violating contributions, both indirect ($`ϵ`$) and direct. There is also a CP-conserving “contaminant” from the intermediate state $`K_L\pi ^0\gamma \gamma `$. The direct contribution probes the CKM parameter $`\eta `$. It is expected to be comparable in magnitude to the indirect contribution, and to have a phase of about $`\pi /4`$ with respect to it. Each contribution (including the CP-conserving one) is expected to correspond to a $`\pi ^0e^+e^{}`$ branching ratio of a few parts in $`10^{12}`$. However, the decay $`K_L\pi ^0e^+e^{}`$ may be limited by backgrounds in the $`\gamma \gamma e^+e^{}`$ final state associated with radiation of a photon in $`K_L\gamma e^+e^{}`$ from one of the leptons HG . Present experimental upper limits (90% c.l.) pll are
$$(K_L\pi ^0e^+e^{})<5.64\times 10^{10},(K_L\pi ^0\mu ^+\mu ^{})<3.4\times 10^{10},$$
(11)
still significantly above theoretical expectations.
3. The decay $`K_L\pi ^0\nu \overline{\nu }`$ is expected to be due entirely to CP violation, and provides a clean probe of $`\eta `$. Its branching ratio, proportional to $`A^4\eta ^2`$, is expected to be about $`3\times 10^{11}`$. The best current experimental upper limit (90% c.l.) for this process pnn is $`(K_L\pi ^0\nu \overline{\nu })<5.9\times 10^7`$, several orders of magnitude above the expected value.
### III.3 Is the CKM picture of CP violation right?
Two key tests have been passed so far. The theory has succeeded, albeit qualitatively, in predicting the range Re$`(ϵ^{}/ϵ)=(1\mathrm{to}2)\times 10^3`$. Its prediction for the branching ratio for $`K^+\pi ^+\nu \overline{\nu }`$ is in accord with the experimental rate deduced from the one event observed so far.
One test still to be passed in the decays of neutral kaons is the measurement of the height $`\eta `$ of the unitarity triangle through the decay $`K_L\pi ^0\nu \overline{\nu }`$. Prospects for this measurement will be mentioned below. However, in the nearer term, one looks forward to a rich set of effects in decays of particles containing $`b`$ quarks, particularly the $`B`$ mesons. To this end, experiments are under way at a number of laboratories around the world.
Asymmetric $`e^+e^{}`$ collisions are being studied at two “$`B`$ factories,” the PEP-II machine at SLAC with the BaBar detector, and the KEK-B collider in Japan with the Belle detector. By end of April 2000, these detectors were recording about 100 and 60 pb<sup>-1</sup> of data per day, respectively, and had accumulated about 6 and 2 fb<sup>-1</sup> of data at the energy of the $`\mathrm{{\rm Y}}(4S)`$ resonance, which decays almost exclusively to $`B\overline{B}`$. The BaBar experiment expects to have about 100 tagged $`B^0J/\psi K_S`$ decays by this coming summer SS .
Significant further data on $`e^+e^{}`$ collisions at the $`\mathrm{{\rm Y}}(4S)`$ are expected from the Cornell Electron Storage Ring with the upgraded CLEO-III detector. The HERA-b experiment at DESY in Hamburg will study $`b`$ quark production via the collisions of 920 GeV protons with a fixed target. The CDF and D0 detectors at Fermilab will devote a significant part of their program at Run II of the Tevatron to $`B`$ physics. In the longer term, one can expect further results on $`B`$ physics from the general-purpose LHC detectors ATLAS and CMS and the dedicated LHC-b detector at CERN, and possibly the dedicated BTeV detector at Fermilab.
## IV CP violation and $`B`$ decays
In constrast to the neutral kaon system, in which the eigenstates of the mass matrix differ in lifetime by nearly a factor of 600, the eigenstates of the corresponding $`B^0`$$`\overline{B^0}`$ mass matrix are expected to differ in lifetime by at most 10–20% for strange $`B`$’s BBD , and considerably less for nonstrange $`B`$’s. Thus, instead of studying the properties of mass eigenstates like $`K_L`$, one must resort to other means. There are two main avenues of study.
* Decays to CP eigenstates $`f=\pm \mathrm{CP}(f)`$ utilize interference between direct decays $`B^0f`$ or $`\overline{B^0}f`$ and the corresponding paths involving mixing: $`B^0\overline{B^0}f`$ or $`\overline{B^0}B^0f`$. Final states such as $`f=J/\psi K_S`$ provide “clean” examples in which one quark subprocess is dominant. In this case one measures $`\mathrm{sin}2\beta `$ with negligible corrections. For the final state $`\pi ^+\pi ^{}`$, one measures $`\mathrm{sin}2\alpha `$ only to the extent that the direct decay is dominated by a “tree” amplitude (the quark subprocess $`bu\overline{u}d`$). When contamination from the penguin subprocess $`bd`$ is present (as it is expected to be at the level of several tens of percent), one must measure decays to other $`\pi \pi `$ states (such as $`\pi ^\pm \pi ^0`$ and $`\pi ^0\pi ^0`$) to sort out various decay amplitudes GrL .
* “Self-tagging” decays involve final states $`f`$ such as $`K^+\pi ^{}`$ which can be distinguished from their CP-conjugates $`\overline{f}`$. A CP-violating rate asymmetry arises if there exist two weak amplitudes $`a_i`$ with weak phases $`\varphi _i`$ and strong phases $`\delta _i`$ ($`i=1,2)`$:
$$A(Bf)=a_1e^{i(+\varphi _1+\delta _1)}+a_2e^{i(+\varphi _2+\delta _2)},$$
$$A(\overline{B}\overline{f})=a_1e^{i(\varphi _1+\delta _1)}+a_2e^{i(\varphi _2+\delta _2)}.$$
(12)
Note that the weak phase changes sign under CP-conjugation, while the strong phase does not. The rate asymmetry is then
$$𝒜(f)\frac{\mathrm{\Gamma }(f)\mathrm{\Gamma }(\overline{f})}{\mathrm{\Gamma }(f)+\mathrm{\Gamma }(\overline{f})}=\frac{2a_1a_2\mathrm{sin}(\varphi _1\varphi _2)\mathrm{sin}(\delta _1\delta _2)}{a_1^2+a_2^2+2a_1a_2\mathrm{cos}(\varphi _1\varphi _2)\mathrm{cos}(\delta _1\delta _2)}.$$
(13)
Thus the two amplitudes must have different weak and strong phases in order for a rate asymmetry to be observable. The CKM theory predicts the weak phases, but no reliable estimates of strong phases in $`B`$ decays exist. Some ways of circumventing this difficulty will be mentioned.
### IV.1 Decays to CP eigenstates
The interference between mixing and decay in decays of neutral $`B`$ mesons to CP eigenstates leads to a term which modulates the exponential decay (see, e.g., DR ):
$$\frac{d\mathrm{\Gamma }(t)}{dt}e^{\mathrm{\Gamma }t}(1\mathrm{Im}\lambda _0\mathrm{sin}\mathrm{\Delta }mt),$$
(14)
where the upper sign refers to $`B^0`$ decays and the lower to $`\overline{B^0}`$ decays. $`\mathrm{\Delta }m`$ is the mass splitting mentioned earlier, and the factor $`\lambda _0`$ expresses the interference of decay and mixing amplitudes. For $`f=J/\psi K_S`$, $`\lambda _0=e^{2i\beta }`$ to a good approximation, while for $`f=\pi ^+\pi ^{}`$, $`\lambda _0e^{2i\alpha }`$ only to the extent that the effect of penguin amplitudes can be neglected in comparison with the dominant tree contribution.
The time integral of the modulation term is
$$_0^{\mathrm{}}𝑑te^{\mathrm{\Gamma }t}\mathrm{sin}\mathrm{\Delta }mt=\frac{1}{\mathrm{\Gamma }}\frac{x}{1+x^2}\frac{1}{\mathrm{\Gamma }}\frac{1}{2},$$
(15)
where $`x\mathrm{\Delta }m/\mathrm{\Gamma }`$. This expression is maximum for $`x=1`$, and 95% of maximum for the observed value $`x0.72`$. It has been fortunate that the $`B^0`$ mixing amplitude and decay rate are so well matched to one another.
The CDF Collaboration CDFB has learned how to “tag” neutral $`B`$ mesons at the time of their production and thus to measure the decay rate asymmetry in $`B^0(\overline{B^0})J/\psi K_S`$. This asymmetry arises from the phase $`2\beta `$ characterizing the two powers of $`V_{td}`$ in the $`B^0`$$`\overline{B^0}`$ mixing amplitude. The tagging methods are of two main types. “Opposite-side” methods rely on the fact that strong interactions always produce $`b`$ and $`\overline{b}`$ in pairs, so that in order to determine the initial flavor of a decaying $`B`$ one must find out something about the “other” $`b`$-containing hadron produced in association with it, either via the charge of the jet containing it or via the charge of the lepton or kaon it emits when decaying. “Same-side” methods GNR utilize the fact that a $`B^0`$ tends to be associated more frequently with a $`\pi ^+`$, and a $`\overline{B^0}`$ with a $`\pi ^{}`$, somewhere nearby in phase space, whether through the dynamics of fragmentation or through the decays of excited $`B`$ resonances.
The CDF result is $`\mathrm{sin}2\beta =0.79_{0.44}^{+0.41}`$. An earlier result from OPAL OPALB and a newer result from ALEPH ALEPHB , both utilizing $`B`$’s produced in the decays of the $`Z^0`$, can be combined with the CDF value to obtain $`\mathrm{sin}2\beta =0.91\pm 0.35`$, which exceeds zero at the 99% confidence level ALEPHB . At the $`1\sigma `$ lower limit (0.56) this is very close to the lower bound (0.59) quoted in Table 2.
### IV.2 “Self-tagging” decays and direct CP violation
An example of direct CP violation can occur in $`B^0K^+\pi ^{}`$. One expects two types of contribution to this process: a “tree” amplitude governed by the quark subprocess $`\overline{b}\overline{u}u\overline{s}`$ with CKM factor $`V_{ub}^{}V_{us}`$, and a “penguin” amplitude governed by the quark subprocess $`\overline{b}\overline{s}`$ with dominant CKM factor $`V_{tb}^{}V_{ts}`$ (since the contribution of the top quark in the internal loop is dominant). These contributions are summarized in Table 4.
Since the tree and penguin amplitudes have a relative weak phase $`\gamma `$ (mod $`\pi `$), one can have $`\mathrm{\Gamma }(B^0K^+\pi ^{})\mathrm{\Gamma }(\overline{B^0}K^{}\pi ^+)`$ as long as the strong phases $`\delta _T`$ and $`\delta _P`$ are different in the tree and penguin amplitudes. However, even if these strong phases do not differ from one another, the ratios of rates for various charge states of $`BK\pi `$ decays can provide separate information on the weak phase $`\gamma `$ GR ; FM ; NR and the strong phase difference $`\delta _T\delta _P`$.
One must first deal with electroweak penguins which were also relevant for the interpretation of $`ϵ^{}/ϵ`$. An early suggestion (see the first of Refs. GR ) proposed a way to extract $`\gamma `$ from the rates for $`B^+(\pi ^0K^+,\pi ^+K^0,\pi ^+\pi ^0)`$ and the charge-conjugate processes. The amplitudes for the first two processes (with appropriate factors of $`\sqrt{2}`$) form a triangle with an amplitude related to the third process by flavor SU(3) as long as electroweak penguins are negligible, which they are not DH . It turns out, however NR , that the relevant electroweak penguin’s contribution to this process can be calculated, so that sufficiently precise measurements of the rates for the above processes can indeed yield useful information on $`\gamma `$.
The possibility has been raised recently NR ; GRg ; Hou that the weak phase $`\gamma `$ may exceed $`90^{}`$. Two processes whose rates hint at this constraint are $`B^0\pi ^+\pi ^{}`$ and $`B^0K^+\pi ^{}`$. The former process has a rate which is somewhat smaller than expected, while the rate for the latter is larger than expected.
The amplitudes contributing to $`B^0\pi ^+\pi ^{}`$ are summarized in Table 5. The relative phase of the tree and penguin amplitudes is $`\gamma +\beta =\pi \alpha `$. The two amplitudes will interfere destructively if the final strong phase difference is small (as expected from perturbative QCD estimates, which indeed may be risky), and if $`\alpha <\pi /2`$. This would tend to favor not-too-positive values of $`\rho `$. There is some hint that the interference is indeed destructive. The observed branching ratio CLB $`(B^0\pi ^+\pi ^{})=(4.3_{1.4}^{+1.6}\pm 0.5)\times 10^6`$ is less than the value of about $`10^5`$ which one would estimate GRg from the tree amplitude alone (e.g., using the observed $`B\pi e\nu _e`$ branching ratio and factorization).
The same types of amplitudes contributing to $`B^0K^+\pi ^{}`$ also contribute to $`B^0K^+\pi ^{}`$ (see Table 4). As in $`B^0K^+\pi ^{}`$, the relative phase between the tree and penguin amplitudes is expected to be $`\gamma \pi `$. One thus expects constructive interference between the two amplitudes if the strong phase difference is small and $`\gamma >\pi /2`$. Indeed, the branching ratio for $`B^0K^+\pi ^{}`$ appears to exceed $`2\times 10^5`$, while the pure “penguin” process $`B^+K^+\varphi `$ has a branching ratio less than $`10^5`$.
A global fit to the above two processes and many others (see the second of Refs. Hou ) finds $`\gamma =(114_{23}^{+24})^{}`$, which just grazes the allowed region quoted in Table 2. Since the upper bound on $`\gamma `$ in Table 2 is set primarily by the lower limit on $`B_s`$$`\overline{B_s}`$ mixing, such mixing should be visible in experiments of only slightly greater sensitivity than those performed up to now.
The Tevatron and the LHC will copiously produce both nonstrange and strange neutral $`B`$’s, decaying to $`\pi ^+\pi ^{}`$, $`K^\pm \pi ^{}`$, and $`K^+K^{}`$ WurtJesik . Each of these channels has particular advantages.
* The decays $`B^0K^+K^{}`$ and $`B_s\pi ^+\pi ^{}`$ should be highly suppressed unless these final states are “fed” by rescattering from other channels resc .
* The decays $`B^0\pi ^+\pi ^{}`$ and $`B_sK^+K^{}`$ can yield $`\gamma `$ when their time-dependence is measured RFKK . The kinematic peaks for these two states overlap significantly, so one must either use particle identification or utilize the vastly different oscillation frequencies for $`B^0`$$`\overline{B^0}`$ and $`B_s`$$`\overline{B_s}`$ mixing to distinguish the two final states.
* A recent proposal for measuring $`\gamma `$ bskpi utilizes the decays $`B^0K^+\pi ^{}`$, $`B^+K^0\pi ^+`$, $`B_sK^{}\pi ^+`$, and the corresponding charge-conjugate processes. The $`B^0K^+\pi ^{}`$ and $`B_sK^{}\pi ^+`$ peaks are well separated from one another and from $`B^0\pi ^+\pi ^{}`$ and $`B_sK^+K^{}`$ kinematically WurtJesik .
The proposal of Ref. bskpi is based on the observation that $`BK\pi `$ decays involve two types of amplitudes, tree ($`T`$) and penguin ($`P`$), with relative weak phase $`\gamma `$ and relative strong phase $`\delta `$. The decays $`B^+K^0\pi ^+`$ are expected to be dominated by the penguin amplitude (there is no tree contribution except through rescattering from other final states), so this channel is not expected to display any CP-violating asymmetries. One expects $`\mathrm{\Gamma }(B^+K^0\pi ^+)=\mathrm{\Gamma }(B^{}\overline{K^0}\pi ^{})`$. This will provide a check of the assumption that rescattering effects can be neglected. A typical amplitude is given by $`A(B^0K^+\pi ^{})=[P+Te^{i(\gamma +\delta )}]`$, where the signs are associated with phase conventions for states GHLR .
We now define
$$\left\{\begin{array}{c}R\\ A_0\end{array}\right\}\frac{\mathrm{\Gamma }(B^0K^+\pi ^{})\pm \mathrm{\Gamma }(\overline{B^0}K^{}\pi ^+)}{2\mathrm{\Gamma }(B^+K^0\pi ^+)},$$
(16)
$$\left\{\begin{array}{c}R_s\\ A_s\end{array}\right\}\frac{\mathrm{\Gamma }(B_sK^{}\pi ^+)\pm \mathrm{\Gamma }(\overline{B_s}K^+\pi ^{})}{2\mathrm{\Gamma }(B^+K^0\pi ^+)},$$
(17)
and $`rT/P`$, $`\stackrel{~}{\lambda }V_{us}/V_{ud}`$. Then one finds
$$R=1+r^2+2r\mathrm{cos}\delta \mathrm{cos}\gamma ,R_s=\stackrel{~}{\lambda }^2+\left(\frac{r}{\stackrel{~}{\lambda }}\right)^22r\mathrm{cos}\delta \mathrm{cos}\gamma ,$$
(18)
$$A_0=A_s=2r\mathrm{sin}\gamma \mathrm{sin}\delta .$$
(19)
The sum of $`R`$ and $`R_s`$ allows one to determine $`r`$. Then using $`R`$, $`r`$, and $`A_0`$, one can solve for both $`\delta `$ and $`\gamma `$. The prediction $`A_s=A_0`$ serves as a check of the flavor SU(3) assumption which gave these relations. An error of $`10^{}`$ on $`\gamma `$ seems feasible with forthcoming data from Run II of the Tevatron.
The CLEO Collaboration has recently presented some upper limits on CP-violating asymmetries in $`B`$ decays to light-quark systems CLEOCP , based on 9.66 million events recorded at the $`\mathrm{{\rm Y}}(4S)`$. With asymmetries defined as
$$𝒜_{CP}\frac{\mathrm{\Gamma }(\overline{B}\overline{f})\mathrm{\Gamma }(Bf)}{\mathrm{\Gamma }(\overline{B}\overline{f})+\mathrm{\Gamma }(Bf)},$$
(20)
the results are shown in Table 6. No statistically significant asymmetries have been seen yet. The sensitivity of these results is not yet adequate to probe the maximum predicted values comb $`|𝒜_{CP}^{K^+\pi }|1/3`$, but is getting close.
## V Some future measurements
The future of the experimental study of CP violation involves a broad program of experiments with kaons, charmed and $`B`$ mesons, and neutrinos. We mention just a few of the possibilities.
### V.1 Rare kaon decays
Plans are afoot for measurement of the branching ratio for $`K_L\pi ^0\nu \overline{\nu }`$ at the required sensitivity ($`3\times 10^3`$). Experiments are envisioned using both relatively slow kaons at Brookhaven National Laboratory K0pio and faster kaons at the Fermilab Main Injector KAMI . A Fermilab proposal CKM seeks to accumulate 100 events of $`K^+\pi ^+\nu \overline{\nu }`$ in order to measure $`|V_{td}|`$ to a statistical precision of 5% and an overall precision of 10%.
### V.2 Charmed mesons
Impressive strides have been taken in the measurement of mass differences and lifetime differences for CP eigenstates of the neutral charmed mesons $`D^0`$ CLEOD ; FOCUS . No significant effects have been seen yet at the level of a percent or so, but there are tantalizing hints ChTh . It would be worth while to follow up these possibilities. Electron-positron colliders, mentioned below, will devote much of their running time to the study of $`B`$ mesons, but charmed mesons are accumulated as well in such experiments, and the samples of them will increase. Hadronic experiments dedicated to producing large numbers of $`B`$’s may also have more to say about mixing, lifetime differences, and CP violation for charmed mesons.
### V.3 $`B`$ production in symmetric $`e^+e^{}`$ collisions
Although asymmetric $`e^+e^{}`$ colliders, known as “B-factories,” are now starting to take data at an impressive rate, the CLEO Collaboration at the symmetric CESR machine has recently celebrated 20 years of $`B`$ physics, and is continuing with an active program. It will be able in the CLEO-III program to probe charmless $`B`$ decays down to branching ratios of $`10^6`$. In so doing, it may be able to detect the elusive $`B^0\pi ^0\pi ^0`$ mode, whose rate will help pin down the penguin amplitude’s contribution and permit a determination of the CKM phase $`\alpha `$ GrL .
Other final states of great interest at this level include $`VP`$ and $`VV`$, where $`P,V`$ denote light pseudoscalar and vector mesons. There is a good chance that direct CP violation may show up in one or more channels if final state phase differences are sufficiently large. The detailed study of angular correlations in $`VV`$ channels may be able to provide useful information on strong final state phases.
A useful probe of rescattering effects resc , mentioned above, is the decay $`B^0K^+K^{}`$. This decay is expected to have a branching ratio of only a few parts in $`10^8`$ if rescattering is unimportant, but could be enhanced by more than an order of magnitude in the presence of rescattering from other channels.
A challenging channel of fundamental importance is $`B^+\tau ^+\overline{\nu }_\tau `$. The rate for this process will provide information on the combination $`f_B|V_{cb}|`$. Rare decays which have not yet been seen (such as $`BX\mathrm{}^+\mathrm{}^{}`$ and $`BX\nu \overline{\nu }`$) will probe the effects of new particles in loops.
### V.4 $`B`$ production in asymmetric $`e^+e^{}`$ collisions
The benchmark process for the BaBar and Belle detectors will be the measurement of $`\mathrm{sin}2\beta `$ in $`B^0J/\psi K_S`$. The PEP-II and KEK-B machines utilize asymmetric $`e^+e^{}`$ collisions in order to create a moving reference frame in which the decays of $`B^0`$ and $`\overline{B^0}`$ are separated by a large enough distance for their separation to be detectable. (Each travels only an average distance of 30 $`\mu `$m in the center of mass.) This facilitates both flavor tagging and improvement of signal with respect to background. These machines will make possible a host of time-dependent studies in such decays as $`B\pi \pi `$, $`BK\pi `$, etc., and their impressive luminosities will eventually add significantly to the world’s tally of detected $`B`$’s.
### V.5 Hadronic $`B`$ production
The strange $`B`$’s cannot be produced at the $`\mathrm{{\rm Y}}(4S)`$ which will dominate the attention of $`e^+e^{}`$ colliders for some years to come. Hadronic reactions at high energies will produce copious $`b`$’s incorporated into all sorts of hadrons: nonstrange, strange, and charmed mesons, and baryons. One looks forward to a measurement of the strange-$`B`$ mixing parameter $`x_s=\mathrm{\Delta }m_s/\mathrm{\Gamma }_s`$. The decays of $`B_s`$ can provide valueable information on CKM phases and CP violation, as in $`B_sK^+K^{}`$ RFKK . The width difference of 10–20% expected between the CP-even and CP-odd eigenstates of the $`B_s`$ system BBD should be visible in the next round of experiments.
### V.6 Neutrino studies
The origin of magnitudes and phases in the CKM matrix is intimately connected with the origin of the quark masses themselves, whose physics still eludes us. We will not understand this pattern until we have mapped out a similar pattern for the leptons, a topic to which many other talks in this Workshop are devoted. Our understanding of neutrino masses and mixings will benefit greatly from forthcoming experiments at the Sudbury Neutrino Observatory SNO , Borexino Bxo , K2K Kam , and Fermilab (BooNE and MINOS) Fnu , to name a few.
### V.7 The $`(\rho ,\eta )`$ plot in a few years
The $`(\rho ,\eta )`$ plot might appear as shown in Fig. 2 in a few years JRlatt ; NSF . We can look forward either to a reliable determination of parameters or to the possibility that one or more experiments give contradictory results, indicating the need for new physics. Such new physics most typically shows up in the form of additional contributions to $`B^0`$$`\overline{B^0}`$ mixing GLmix , though it can also show up in decays GW .
## VI Baryon asymmetry
The ratio of the number of baryons $`n_B`$ to the number of photons $`n_\gamma `$ in the Universe is a few parts in $`10^{10}`$, much larger than the corresponding ratio for antibaryons. Shortly after the discovery of CP violation in neutral kaon decays, Sakharov proposed in 1967 Sakh three ingredients needed to understand this preponderance of matter over antimatter: (1) an epoch in which the Universe was not in thermal equilibrium, (2) an interaction violating baryon number, and (3) CP (and C) violation. However, one can’t explain the observed baryon asymmetry merely by means of the CP violation contained in the CKM matrix. The effects are too small unless some new physics is introduced. Two examples are the following:
* The concept of supersymmetry, in which each particle of spin $`J`$ has a “superpartner” of spin $`J\pm 1/2`$, affords many opportunities for introducing new CP-violating phases and interactions which could affect particle-antiparticle mixing SSBrev .
* The presence of neutrino masses at the sub-eV level can signal large Majorana masses for right-hand neutrinos, exceeding $`10^{11}`$ GeV PR . Lepton number ($`L`$) is violated by such masses. The violation of $`L`$ can easily be reprocessed into baryon number ($`B`$) violation by $`BL`$ conserving interactions at the electroweak scale LepBar . New CP-violating interactions must then exist at the high mass scale if lepton number is to be generated there. It is conceivable that these interactions are related to CKM phases, but the link will be very indirect DPF . In any case, if this alternative is the correct one, it will be very important to understand the leptonic analogue of the CKM matrix!
## VII Conclusions
The CKM theory of CP violation in neutral kaon decays has passed a crucial test. The parameter $`ϵ^{}/ϵ`$ is nonzero, and has the expected order of magnitude, though exceeding some theoretical estimates. Still to come will be several tests using $`B`$ mesons, including the observation of a difference in rates between $`B^0J/\psi K_S`$ and $`\overline{B^0}J/\psi K_S`$. There will be more progress in “tagging” neutral $`B`$’s, and we can look forward to rich information from measurements of decay rates of charged and neutral $`B`$’s into a variety of final states.
I see two possibilities for our understanding of CP violation in the next few years. (1) If $`B`$ decays do not provide a consistent set of CKM phases, we will be led to examine other sources of CP violation. Most of these, in contrast to the CKM theory, predict neutron and electron dipole moments very close to their present experimental upper limits. (2) If, on the other hand, the CKM picture still hangs together after a few years, attention should naturally shift to the next “layer of the onion”: the origin of the CKM phases (and the associated quark and lepton masses). It is probably time to start anticipating this possibility, given the resilience of the CKM picture since it was first proposed nearly 30 years ago.
## VIII Acknowledgements
It is a pleasure to thank Maria Eugenia and Jose Nieves for their wonderful hospitality in San Juan. This work was supported in part by the United States Department of Energy under Grant No. DE FG02 90ER40560.
|
warning/0005/nlin0005031.html
|
ar5iv
|
text
|
# The power spectrum of geodesic divergences as an early detector of chaotic motion
## 1 Introduction
The problem of distinguishing a chaotic from an ordered trajectory in a non-integrable Hamiltonian system has been a topic of active investigation since the pioneering work of Hénon and Heiles (1964). Initially, when the study was restricted to 2-D systems, the work was done through surface of section plots. Later, when systems with more than 2-D were considered, the method of choice was the calculation of the Lyapunov Characteristic Numbers (LCNs) (Benettin et al. 1976, Froeschlé 1984). Unfortunately both the above methods suffer from the same drawback, namely they are not able to distinguish easily an ordered from a “sticky” chaotic trajectory. Various methods have been devised since then to address the above problem, namely the distinction of an ordered from a chaotic trajectory using a relatively short-time trajectory segment.
These methods generally fall into two main classes: those that use frequency or correlation analysis of a time-series, constructed by the values of generalized co-ordinates (or functions of them), and those that use the geodesic divergence of initially nearby trajectories. In the first class belong the “old” method of the rotation number (Contopoulos 1966), the frequency map analysis developed by Laskar (Laskar et al. 1992, Laskar 1993) and the power spectrum analysis of quasi-integrals developed by Voyatzis & Ichtiaroglou (1992). In the second belong the probability density function analysis of stretching numbers developed by Froeschlé et al. (1993) and Voglis & Contopoulos (1994) and the Fast Lyapunov Indicators method developed by Froeschlé et al. (1997). Each one of the above methods has its own advantages and weaknesses; in particular some of them are more suitable to test large sets of trajectories rather than single ones, some are more efficient for 2-D systems rather than for N($`>`$2)-D systems and some perform better for mappings rather than flows.
In 1997, Contopoulos & Voglis introduced a new method for distinguishing chaotic from ordered trajectories, which does not belong to any of the above mentioned two classes but, instead, may be classified as “mixed”. This new method is based on the analysis of the probability density of helicity and twist angles. Voglis & Efthymiopoulos (1998) and subsequently Froeschlé & Lega (1998) showed that the twist angles method is very efficient in testing phase space regions, at least in cases of 2-D systems, where the twist angles can be easily calculated. More recently Voglis et al. (1998, 1999) proposed two new and very efficient methods, namely the method of “Dynamical Spectral Distance” (DSD), which is particularly suitable for the characterization of single trajectories in 4-D maps, and the method of “Rotational Tori Recognizer” (ROTOR) which is very efficient for testing wide areas of the phase space of 2-D maps.
In the present paper we are introducing a new “mixed” method, which we show that it is at least as sensitive as the other methods in the literature, may be applied in a straightforward way to dynamical systems with more than two degrees of freedom and is equally efficient for single trajectories as well as large sets of them. The method consists in analyzing a time series constructed by the values of the geodesic deviation of nearby trajectories recorded at a properly selected frequency. It should be noted that a method based on a similar technique has been proposed by Lohinger and Froeschlé (1993).
The paper is organized as follows. Section 2 describes the basic features of the method, which, in Section 3, is tested upon three dynamical systems of different types. A comparison of the results of our method to those derived by various other methods is made in Section 4. Section 5 treats the application of the proposed method to one of the most important problems of solar system dynamics, the motion of asteroids. Finally in Section 6 we present our conclusions.
## 2 The method
In order to decide on the nature of a trajectory (chaotic or not), we work as follows. We integrate numerically the “main” trajectory together with a nearby one, which at a time $`t_0=0`$ starts at an infinitesimal distance in phase space, $`d_0`$, from the main, and we calculate their distance, $`d_1`$, at a time $`t=t_0+\mathrm{\Delta }t`$. Let us denote by $`q`$ the logarithm of the ratio of the two distances, $`q=\mathrm{ln}(d_1/d_0)`$. We then renormalize the nearby trajectory, so as to start from a new position in phase space, which is at distance $`d_0`$ from the main trajectory in the direction of $`d_1`$. The trajectories are followed once again for a time interval $`\mathrm{\Delta }t`$ and a new $`q`$ is calculated. After following the trajectories for a time interval $`T=N\mathrm{\Delta }t`$, we have constructed a time series consisting of the consecutive $`q`$’s
$$q(t)=\mathrm{ln}\left[\frac{d_t}{d_0}\right]_t\mathrm{or}q_k=\mathrm{ln}\left[\frac{d_k}{d_0}\right]_{t_k}$$
(1)
An ordered trajectory of a N-D conservative dynamical system will lie on an invariant torus, i.e. a N-D manifold of the 2N-D phase space. Any such trajectory is, in general, quasi-periodic and covers densely the invariant set. If a nearby trajectory is started at an infinitesimal distance from the previous one, this too would, in general, lie on an invariant torus, so that the $`q(t)`$ time series should behave in a quasi-periodic manner. On the other hand, a chaotic trajectory visits different regions of phase space in a stochastic manner and $`q(t)`$ should also be “random”.
Following the above considerations we calculate the power spectrum $`P(f)`$ of the $`q(t)`$ time series. We first calculate the discrete Fourier transform of the $`q_k`$ multiplied by a window function $`w_k`$
$$Q_j=\underset{k=0}{\overset{N1}{}}q_kw_ke^{2\pi ijk/N}j=0..N1$$
(2)
Then the power spectrum, $`P(f)`$, is defined in $`M=N/2+1`$ frequencies as
$`P(f_0)`$ $`=`$ $`{\displaystyle \frac{1}{W}}|Q_0|^2`$
$`P(f_j)`$ $`=`$ $`{\displaystyle \frac{1}{W}}(|Q_j|^2+|Q_{Nj}|^2)j=1..({\displaystyle \frac{N}{2}}1)`$ (3)
$`P(f_c)`$ $`=`$ $`{\displaystyle \frac{1}{W}}|Q_{N/2}|^2`$
where
$$W=N\underset{k=0}{\overset{N1}{}}w_k^2$$
(4)
and $`f_c=f_{N/2}`$ is the Nyquist frequency defined as
$$f_c=\frac{1}{2\mathrm{\Delta }t}.$$
(5)
The frequencies covered by the power spectrum are
$$f_j=f_c\frac{j}{M}j=0..M$$
(6)
In the present work we used the so called “Hanning” window. More details on the calculation of the power spectrum can be found in the book by Press et. al. (1992).
As a rule “mixed” methods are expected to perform better in distinguishing between ordered and chaotic trajectories. The reason for that is that time series that are constructed from the geodesic divergence of nearby orbits contain all the various characteristic frequencies that locally affect the motion in the “proper” ratio, i.e. the frequencies corresponding to the different directions (degrees of freedom) are properly weighted.In particular now, as far as our method is concerned, the Power Spectrum Of Divergences (PSOD) of an ordered trajectory is expected to posses only a few “spikes” at specific frequencies. The number of harmonics however depends on the system under consideration as well as on the values of its “controll” parameters (see below). In contrast, the PSOD of a chaotic trajectory should appear continuous, due to the random nature of the $`q_k`$ time series. However, the above considerations lie behind all methods based on time series analysis. What is really important, for the assessment of the new method with respect to the other ones appearing in the literature, is to evaluate (a) its independence from the number of degrees of freedom of the dynamical system, (b) its effectiveness with respect to the object of test (single trajectories or distributions of initial conditions covering wide phase space regions) (c) its sensitivity, i.e. the minimum length of the time series, necessary to distinguish a sticky chaotic trajectory from an ordered one and (d) its ability to produce a well-defined measure of chaos (as is the LCN).
## 3 Evaluation of the method
We proceed in the assessment of the method using three different dynamical systems, namely a 2-D mapping as well as a 2-D and a 3-D Hamiltonian system. In each one we evaluated the nature of a considerable number of trajectories. In the following subsections we present only three or four trajectories per system, which we think that are typical examples of the three different classes of trajectories, i.e. ordered, clearly chaotic and sticky. The amplitudes of the PSOD, in all figures, are normalized so that the highest has the value one, while the frequency is given in cycles per time unit.
### 3.1 2-D mapping
We first test the method in the simple 2-D mapping
$`J_{i+1}`$ $`=`$ $`J_i+k\mathrm{cos}(2\theta _i)\mathrm{mod}(2\pi )`$
$`\theta _{i+1}`$ $`=`$ $`\theta _i+J_{i+1}\mathrm{mod}(2\pi )`$ (7)
where the stochasticity parameter $`k`$ is taken equal to 0.7.
We present here the results of three trajectories, an ordered one (map1) starting at $`J_0=\pi `$, $`\theta _0=1.4\pi `$, a stochastic one (map2) starting at $`J_0=\pi `$, $`\theta _0=1.5\pi `$ and a sticky one (map3) starting at $`J_0=\pi `$, $`\theta _0=1.538\pi `$. The initial conditions of the third one place it very close to the boundary between the ordered region, surrounding the stable point $`J_0=\pi `$, $`\theta _0=5\pi /4`$, and the chaotic sea. Figure 1 shows the consequents ($`\theta `$, $`J`$) of the “sticky” trajectory at various times (iterations). As we can see, for up to 2 048 iterations the trajectory behaves like an ordered one. Around $`i=\mathrm{4\hspace{0.17em}000}`$ it starts to present some signs of irregularity and finally, after $`i=\mathrm{6\hspace{0.17em}000}`$, the chaotic nature of the trajectory becomes evident. Using our method on these three trajectories, with the length of the $`q_k`$ time series being N=1024 iterations and defining $`\mathrm{\Delta }t=1`$, we obtain the spectra shown in Fig. 2.
The upper left frame of Fig. 2 is the spectrum of the ordered trajectory. The spectrum consists of some basic frequencies while the “noise” is at a very low level. On the contrary, the spectrum of the chaotic trajectory, in the upper right frame, covers the whole frequency range with comparable amplitudes, i.e. a clearly continuous spectrum. Looking at the spectrum of the “sticky” trajectory (lower left frame), we see a pattern almost the same as that of the chaotic one. Some high-amplitude spikes are evident but the continuum noise level is again very high for the whole frequency range, denoting the stochastic nature of the trajectory. Even with much less iterations (N=256 - lower right frame of Fig. 2) we get the same result. The spectrum is less dense, since it has a smaller number of frequencies than before (see eq.(6)), but the basic features are the same. Note that for the LCN, which is the limit of the function
$$\chi (t)=\frac{1}{N\mathrm{\Delta }t}\underset{i=1}{\overset{N}{}}q_i$$
(8)
as $`N\mathrm{}`$ (see Fig. 3), or the plot of the consequents of the mapping (Fig. 1), we need much more than 1024 iterations in order to decide whether the trajectory is chaotic or not.
### 3.2 An improved criterion
The above presented results show that it is, indeed, worth to consider the new method as a useful tool in the assessment of the nature, ordered or chaotic, of a trajectory. However the method, as it is, does not entail a clear and easy to apply criterion for the classification of a trajectory as ordered or chaotic. Here we try to improve somehow the presentation of the results of our method, in order to propose such a criterion. Note that this new criterion is similar, graphically, to the criterion of the FLI proposed by Froeschlé et al (1997).
If the peaks appearing in the PSOD are plotted in descending order of amplitude, we have a graphical representation of how many strong frequencies the spectrum possesses. Figure 4 shows this representation of the PSOD for the three trajectories studied in the previous sub-section. Ordered trajectories have only a few high amplitude frequencies and the background is formed by peaks whose amplitudes are more than four orders of magnitude smaller than that of the basic frequency. On the other hand the stochastic trajectories possess only a few high-amplitude frequencies and the largest part of the spectrum consists of a “continuum” of frequencies with also considerable amplitude. In this respect one may chose to represent the results by a signle number (e.g. the number of peaks up to a certain amplitude), provided that a certain “threshold” for the noise level is chosen (10<sup>-8</sup> in Fig. 4). This value will, however, depend on the system at study.
### 3.3 Noise level
Any method of analysis of finite-sample time series is bound to suffer from noise. Thus, even in the PSOD of a regular orbit a certain noise level is expected. This is mainly due to “leakage” of power from the frequency lobes, a side effect of the calculation of the power spectrum using Discrete Fast Fourier Transform methods. When calculating the power spectrum of a “monochromatic” signal, the power contained in its basic frequency “leaks” into neighboring frequencies. The leakage depends on the windowing function used. When a signal possesses two closely spaced frequencies, all frequencies in between these two will also gain considerable amplitudes, due to this phenomenon. If the PSOD of a regular orbit has a large number of basic (strong) frequencies, then this effect can lead us to falsely identify it as chaotic. The problem can be tackled at the cost of taking more points in the sample. In this way, the number of frequencies appearing in the spectrum is increased but the frequency lobes become thinner. Thus, lobe overlapping is reduced and the noise level drops. For chaotic orbits on the other hand, the observed noisy pattern is an inherent property of the spectrum and by increasing the number of points one cannot alter the picture.
The above can be seen in Fig. 5, which shows the PSOD of a regular (left column) and a chaotic trajectory (right column) for three different values of N, i.e. N=256 (top), N=4096 (middle) and N=65 536 (bottom). As we see, the noise level in the case of the regular orbit drops significantly when N is increased, while in the case of the chaotic orbit it remains more or less unchanged. In the N=65 536 case the noise level for the ordered trajectory drops below $`10^{20}`$, becoming comparable to the accuracy of the FFT calculation (double precision).
This phenomenon is better seen if we use the amplitude-sorted PSOD. Fig. 6 shows the amplitude-sorted PSOD of the regular (left) and chaotic (right) trajectory with N=256, 1024, 4096, 16 384 and 65 536. While the noise level in the PSOD of the ordered trajectory is reduced when N is increased, it remains the same for the chaotic one.
### 3.4 2-D Hamiltonian system
We now proceed to test the method in a 2-D Hamiltonian system. We selected the Hamiltonian used by Caranicolas & Vozikis (1987),
$$H=\frac{1}{2}\left(p_x^2+p_y^2\right)+x^4+y^4+2ax^2y^2=h$$
(9)
where the parameter $`a`$ is taken equal to 2.0 and the energy constant $`h=1.0`$.
Again we study three trajectories, one ordered (ord-2) starting at $`x(0)=0.5`$, one clearly stochastic (ch-2) starting at $`x(0)=0.001`$ and one “sticky” (st-2) which starts at $`x(0)=0.1326`$. All three trajectories have also $`y(0)=0`$ and $`p_x(0)=0`$, while their $`p_y(0)`$ is given by the energy integral.
Using the PSOD with a renormalization time-step equal to $`\mathrm{\Delta }t=1.5`$, we find that the spectra of the three test trajectories present the same properties as those of the corresponding cases of the mapping. In Fig. 8 we see the PSOD of the three test trajectories. The difference between the spectrum of the chaotic trajectory ch-2 and that of the ordered trajectory ord-2 is again obvious. Moreover, the “sticky” trajectory st-2 has a spectrum similar to that of ch-2. Note that we have used only 2048 points (corresponding to $`t=3072`$) while, if we look at the evolution of the $`\chi (t)`$ (Fig. 7), the trajectory looks ordered for a time up to $`t\mathrm{7\hspace{0.17em}000}`$. The fourth frame (lower right) in Fig. 8 corresponds to another regular orbit (ord-2a) that starts at $`x=0.133`$, i.e. very close to the sticky trajectory. We see that, although the two trajectories start very close to each other and, at least for the first 3 000 times steps, span approximately the same phase-space region, their PSODs are completely different, clearly revealing the nature of each case.
We decided to test also the case where the parameter $`a`$ is taken equal to $`4.0`$. As Caranicolas & Vozikis (1987) have shown, the surface of section of this case has a completely different topology from that of the $`a=2`$ case. The equipotential curves have negative curvature along the $`y=\pm x`$ lines. This affects mainly the loop orbits which appear “squared”.
Again we study four trajectories, one ordered (ord-4) starting at $`x(0)=0.3`$, one clearly chaotic (ch-4) starting at $`x(0)=0.7`$, one “sticky” (st-4) which starts at $`x(0)=0.36282`$ and one ordered (ord-4a) starting at $`x(0)=0.362`$ very near to the sticky one. The surface of section plot for the “sticky” trajectory is shown in Fig. 9 at various times. Figures 10 and 11 present the PSOD and the amplitude-sorted PSOD for these four orbits (N=4096). An important characteristic seen in these two figures is the presence of a high number of medium-amplitude frequencies in the spectra of the regular orbits. However, a distinction between regular and chaotic orbits can still be made. Due to frequency overlapping, discussed in section 3.3, individual frequencies cannot be distinguished at an amplitude level smaller than $`10^6`$. This makes it very difficult to identify a sticky orbit with an amplitude level around $`10^610^7`$. Nevertheless the noise level of the regular orbits will be supressed if we take more points, while for a sticky chaotic orbit it will remain more or less the same.
### 3.5 3-D Hamiltonian system
We apply our method to the model Hamiltonian used by Magnenat (1982), Contopoulos & Barbanis (1989), Barbanis and Contopoulos (1995), Barbanis (1996), Varvoglis et al. (1997), Barbanis et. al. (1999) and Tsiganis et al. (2000a)
$`H`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(p_x^2+p_y^2+p_z^2\right)`$ (10)
$`+{\displaystyle \frac{1}{2}}\left(Ax^2+By^2+Cz^2\right)ϵxz^2\eta yz^2=h`$
where the parameters are taken as $`A=0.9`$, $`B=0.4`$, $`C=0.225`$, $`ϵ=0.560`$ and $`\eta =0.20`$ and the energy level is $`h=0.00765`$
We again test three trajectories, an ordered one (reg01) starting at $`\overline{x}=0.01`$, $`\overline{y}=0.032`$, a chaotic one (ch-01) starting at $`\overline{x}=0`$, $`\overline{y}=0`$ and a sticky one (st-01) with initial conditions $`\overline{x}=0.01725`$, $`\overline{y}=0.032`$, while $`\overline{z}`$, $`p_x`$, $`p_y`$ were taken equal to 0 and $`p_z`$ is calculated from the energy integral. We use the variables $`\overline{x}`$, $`\overline{y}`$, $`\overline{z}`$ instead of $`x`$, $`y`$, $`z`$ in order to be consistent with the previous publications. The barred variables are defined as $`\overline{x}=\sqrt{A}x`$, $`\overline{y}=\sqrt{B}y`$ and $`\overline{z}=\sqrt{C}z`$.
In 3-D one cannot visualize a surface of section plot, in order to check whether a particular trajectory is ordered or chaotic. Therefore, if one is using the traditional tools, he has to rely on the calculation of LCNs. It should be pointed out that a positive LCN is a proof that the trajectory under study is chaotic, while a monotonically decreasing value of $`\chi (t)`$ is not a proof of order, since this behavior could very well originate from “stickiness”. That is why we decided to test one more trajectory (reg00) for which we can be almost certain that it is ordered, as it has the same initial position as reg01 but it belongs to an almost integrable case of the model Hamiltonian, $`ϵ=0.01`$ and $`\eta =0.01`$.
Figure 12 shows the calculation of the $`\chi (t)`$ for the four trajectories. We can clearly see that $`\chi (t)`$ of st-01 is decreasing up to $`t=\mathrm{7\hspace{0.17em}10}^4`$ and then begins to saturate to a non-zero LCN value. The stochasticity is even more evident after $`t=\mathrm{2.2\hspace{0.17em}10}^5`$, where we have a “jump” to a higher LCN value.
We calculate the PSOD using $`\mathrm{\Delta }t=15`$, a value approximately equal to the time interval between two consecutive crossings of the $`\overline{x}\overline{y}`$ plane by the trajectory. Figure 13 shows the PSOD of the four trajectories using 512 points, i.e. for $`t=\mathrm{7\hspace{0.17em}680}`$. Note once again that we can decide that the “sticky” trajectory st-01 is actually stochastic well in advance of the LCN method. The LCN shows the stochastic behaviour of the trajectory only after $`t=\mathrm{70\hspace{0.17em}000}`$, while with the PSOD we need a modest $`t=\mathrm{7\hspace{0.17em}680}`$.
## 4 Comparison with other methods
As we already mentioned in the introduction, three of the most recent methods for distinguishing ordered from chaotic trajectories are the Fast Lyapunov Indicators (FLI) (Froeschlé et al. 1997), the “spectra” of stretching numbers and/or twist angles (Froeschlé et al. 1993, Voglis & Contopoulos 1994, Contopoulos & Voglis 1997) and the “spectral distance” (DSD) (Voglis et al. 1998, 1999). Out of these three methods the fastest ones are the FLI and the DSD methods. In this section we compare the PSOD with the FLI and the DSD methods by applying it to the test models presented in the above mentioned papers. A discussion concerning the spectra of stretching numbers follows.
### 4.1 Single trajectories
#### 4.1.1 FLI
In the paper by Froeschlé et al. (1997) the authors tested the FLI method on two trajectories of the standard map
$`x_{i+1}`$ $`=`$ $`x_i+k\mathrm{sin}(x_i+y_i)\mathrm{mod}(2\pi )`$
$`y_{i+1}`$ $`=`$ $`x_i+y_i\mathrm{mod}(2\pi )`$ (11)
with $`k=0.3`$; one stochastic starting at $`x=0.001`$, $`y=0.001`$ and one ordered starting at $`x=1`$, $`y=0`$. They found that for the stochastic trajectory the FLI’s drop very quickly down to $`10^{20}`$ (Fig. 2 in their paper - 200 iterations). On the contrary, the function $`\chi (t)`$ levels only after about 10 000 iterations. For the ordered trajectory the FLI’s are slowly decreasing, following $`\chi (t)`$. In Fig. 15 we present the PSOD’s of the stochastic (left frame) and the ordered (right frame) trajectory, calculated using only 256 iterations. It is obvious that the two spectra clearly differentiate between the two types of trajectory.
#### 4.1.2 Spectral distance (the 4D map)
In a recent paper Voglis et al. (1999) proposed, as a tool for the distinction between chaotic and regular orbits in 4D maps, the use of the “spectral distance” $`D^2`$. The method is based on the property that the “spectrum” of stretching numbers (as well as that of the helicity angles) of a chaotic trajectory is independent of the initial orientation of the deviation vector, while the spectrum of a regular trajectory is not.
The “spectral distance” $`D^2`$ is a norm defined as
$$D^2=\underset{q}{}\left[S_1(q)S_2(q)\right]^2$$
(12)
where the summation is for all $`q`$’s and $`S_1`$, $`S_2`$ are two spectra of the same orbit but with two different initial deviation vectors.
Voglis et al. (1999) applied their method to a 4-D mapping consisting of two coupled 2-D standard maps, i.e.
$`x_1^{}`$ $`=`$ $`x_1+x_2^{}`$
$`x_2^{}`$ $`=`$ $`x_2+{\displaystyle \frac{k}{2\pi }}\mathrm{sin}(2\pi x_1){\displaystyle \frac{\beta }{\pi }}\mathrm{sin}(2\pi (x_1x_3))`$ (13)
$`x_3^{}`$ $`=`$ $`x_3+x_4^{}`$
$`x_4^{}`$ $`=`$ $`x_4+{\displaystyle \frac{k}{2\pi }}\mathrm{sin}(2\pi x_3){\displaystyle \frac{\beta }{\pi }}\mathrm{sin}(2\pi (x_3x_1))`$
where the $`x_i`$’s are defined in the interval \[0,1) (i.e.$`\mathrm{mod1}`$).
We chose to test our method upon the two most interesting cases shown in Voglis et al. (1999), namely trajectories A2 and A3, in their notation. The A2 case has initial conditions ($`x_1,x_2,x_3,x_4`$)=(0.55,0.1,0.62,0.2), $`\beta =0.1`$ and is a regular orbit, while the A3 case has the same initial $`x_i`$’s, $`\beta =0.3051`$ and is a chaotic orbit but with a very low value of LCN (around $`\mathrm{4\hspace{0.17em}10}^7`$).
Fig.16 shows the PSOD of the two test trajectories. The left panel corresponds to the regular orbit (A2) and the right panel corresponds to the chaotic one (A3). The spectra were calculated using 4096 iterations for each orbit. The distinction between the regular and the chaotic orbits is apparent. Note that our method gave the same result as the $`D^2`$ method with almost the same computational effort. Of course, both methods are much faster than the traditional LCN method.
#### 4.1.3 “Spectrum” of stretching numbers
The “spectrum” of stretching numbers, $`S(q)`$, (Froeschlé et al. 1993, Voglis & Contopoulos 1994, Contopoulos & Voglis 1997, Dvorak et al. 1998) is a method also based on the divergence of nearby trajectories. It consists of the calculation of the probability density of $`q_k`$ (eq.(̃1)), i.e.
$$S(q)=\frac{\mathrm{\Delta }N(q)}{Ndq}$$
(14)
where $`\mathrm{\Delta }N(q)`$ is the number of $`q_k`$ with values between $`q`$ and $`q+dq`$. A quasi-periodic trajectory, which lies very close to a periodic trajectory, has a “U” shaped distribution of stretching numbers which is also symmetric around $`q=0`$ (Contopoulos et al. 1997). As we move away from the periodic trajectory, this symmetry is destroyed (Caranicolas & Vozikis 1999) and the spectrum starts to develop a greater number of maxima. On the other hand if the trajectory is chaotic, the spectra have different shapes and are not symmetric at all. It should be pointed out, however, that, in order to obtain a well defined spectrum, one needs to account for a large number of iterations (typically $`N=10^5`$ or more). Therefore, we did not attempt to compare this method to our own.
### 4.2 Sets of trajectories
In order to circumvent the problem of calculating a trajectory for long times, Contopoulos and Voglis (1997) proposed the use of the average value of $`q`$,
$$<q>=\frac{1}{N}\underset{k=1}{\overset{N}{}}q_k$$
(15)
If we keep N small, we can scan a wide area of initial conditions and map its dynamical behavior. Trajectories in chaotic domains will have $`<q>`$ scattered around the value of the LCN of this domain, while trajectories in the ordered domain will have $`<q>`$ near zero. The method is very fast in distinguishing between ordered and chaotic domains. Fig. 9 of Contopoulos & Voglis (1997) is a very good example of the results obtained by this method with only 10 iterations.
However, although the method is very good in scanning wide areas of phase space for locating islands of order, it cannot give reliable results with so few iterations for a particular trajectory. In the case of stochastic trajectories, $`<q>`$ for $`N=10`$ varies so much, that it may yield a number as small as the one given for ordered trajectories. The situation is even worse in the case of sticky trajectories (i.e. in the borders of islands). In order to decide on the character of such trajectories one needs considerably longer calculations.
Froeschlé & Lega (1998) tested the FLI method along with the method of twist angles (Contopoulos & Voglis 1997), the frequency map analysis method (Laskar et al. 1992, Laskar 1993) and the sup-map method (Laskar 1990, Froeschlé & Lega 1996). Figs. 8a-d of their paper shows the results of the four methods on a cross section of 1 000 trajectories near the hyperbolic point of the 1/6 resonance of the standard map (Eq. 11) with $`k=1.3`$. For the FLI method they used 2 000 iterations while for the other three methods 20 000 iterations. In order to produce unambiguous results, the trajectories in this test are classified as ordered or chaotic by an appropriately selected number/indicator. For the FLI method the authors have used as an indicator the time necessary for the FLI to reach a value lower than $`10^{10}`$.
In order to compare our method with these results we need also an one-number indicator derived from the PSOD. As such we have selected here the average value of $`<P/P_{max}>`$. The averaging is performed not over all values but by ignoring the highest 1/6th and the lowest 1/6th of the amplitudes, the former being probably due to periodicities in sticky trajectories and the latter probably coming from numerical errors in the integration and the calculation of the power spectrum. Fig. 17 shows the same cross section with Fig. 8a-d of Froeschlé & Lega (1998), using the above indicator taken after 8 192 iterations. As we can see it gives essentially the same information as the other four methods. Note that the y-axis in Fig. 17 is inverted for easier comparison with Fig. 8a-d of Froeschlé & Lega (1998).
## 5 Application to asteroidal motion
As an application of our method to a problem of physical importance, we shall use it in order to assess the chaotic or not nature of asteroidal trajectories. We use a simplified model of the solar system, namely the planar restricted three body problem where the Sun and Jupiter move on elliptic trajectories around their center of mass and an asteroid of infinitesimal mass moves in the gravitational field of the two bodies. We calculate the PSOD and the LCN using as renormalization time, $`\mathrm{\Delta }t`$, the period of Jupiter, i.e. $`\mathrm{\Delta }t=T_J11.86`$ years.
Fig. 18 shows the $`\chi (t)`$ of the four trajectories tested. The upper left frame belongs to an ordered trajectory starting with semi-major axis $`a=3`$ and eccentricity $`e=0`$, the upper right to a stochastic trajectory with initial elements $`a=4`$, $`e=0.2`$. The other two frames correspond to trajectories with initial $`a=3.64`$, $`e=0.08`$ (lower left) and $`a=3.63`$, $`e=0.08`$ (lower right) representing “clones” of the asteroid 522 - Helga, which is a well-known example of “stable chaos” (Milani & Nobili 1992).
The PSOD of the four trajectories (N=512) is shown in Fig. 19. As we can see, a few Jovian periods are sufficient to decide if the motion of a particular asteroid is stochastic or ordered. In the case of the Helga clone with $`a=3.64`$ (lower left of Fig. 19) the PSOD shows clearly a chaotic nature after only 512 Jovian periods i.e. 6 072 years, while even a rough calculation of the LCN needs at least $`10^5`$ years (see lower right frame of Fig. 18). The PSOD of the $`a=3.63`$ Helga clone is rather peculiar. Although it differs from that of the ordered orbit (upper left), it is not clearly chaotic, unlike the PSOD of the other Helga clone (lower left). Nevertheless, if we take more points in our time series the chaotic nature of the orbit becomes apparent, as we can see in Fig. 20. However, the spectrum can still be described as having a strong quasi-periodic component, something which is related to the peculiar dynamical nature of this orbit as discussed in Tsiganis et al. (2000b).
## 6 Conclusions – Discussion
In the present paper we propose an alternative tool, which we call PSOD, for the characterization of the chaotic or not nature of trajectories in conservative dynamical systems. The method is based on the frequency analysis of a time series, constructed by successive records of the amplitude of the deviation vector of nearby trajectories. As discussed in Section 2, such a “mixed”<sup>1</sup><sup>1</sup>1Since the method uses both the deviation vectors and frequency analysis it may be classified as “mixed” method is expected to have certain advantages. The reason is that the power spectrum of such a time series will contain all the characteristic frequencies of the motion in a properly “weighted” ratio.
The basic characteristic of the PSOD, seen in all three test cases (a 2-D mapping, a 2-D and a 3-D Hamiltonian system), is that
* for ordered trajectories the spetrum possesses only a few high-amplitude peaks, the exact number of which depends not only on the system but also on the particular orbit. Of course a small amplitude noise level, due to the numerical procedure, is superimposed on the spectrum, which diminishes as the length of the time series is increased;
* for chaotic trajectories the spectrum has a noisy pattern. For weakly chaotic orbits a few high-amplitude peaks are also present. Increasing the length of the time series, the spectrum tends to a white noise spectrum which remains practically unchanged for any (large enough) number of points (see Figs. 5 and 6).
Like most methods existing in the literature, it seems that the method performs better for maps than for flows. However, the results found for the three Hamiltonian flows tested (including the three-boody problem) show that the method can be applied to any system, no matter how many degrees of freedom are involved. We believe that the results may be significantly improved for flows, provided that proper analysis concerning the renormalization time is conducted. The difference between maps and flows is that, in the former case, isochronous records of dynamical quantities also mean studying the system on a surface of section. This is not the case for flows and, thus, a more refined analysis on how to select a proper renormalization time has to be made.
We have shown that the sensitivity of the PSOD in testing single trajectories in maps is comparable to the FLI and the DSD methods. This also makes the PSOD an efficient tool for scanning wide areas of the phase space (see Section 4). For such purposes one would like to have an one-number indicator for measuring chaos. The only uniquely defined measure of chaos is of course the LCN. Any other indicator should be in a one-to-one correspondence with the LCN, in order to give the same information. There is no guarantee that such an indicator can be based on the frequency content of the PSOD, as chaotic orbits with similar LCNs may have a different frequency distribution. Further analysis of the properties of the PSOD has to be performed in order to decide whether such an indicator can be found.
###### Acknowledgements.
The authors would like to acknowledge the constructive comments of the anonymous referee.
|
warning/0005/hep-lat0005011.html
|
ar5iv
|
text
|
# Non-Gaussian numerical errors versus mass hierarchy
## I Introduction
In the standard model of electro-weak and strong interactions, all the masses are generated by couplings to a scalar Higgs particle. However, one might be inclined to think that, ultimately, gravitational interactions should be responsible for mass generation. If there is no new physics between the electro-weak scale and the Planck scale and if we know nothing about the Planck scale interactions, the best we can do is to use an effective scalar Lagrangian with a cut-off of the order of the Planck scale. In this approach, one is confronted with the hierarchy problem, which means that in order to keep a physical mass very small in cut-off units, one needs to fine-tune some parameter of the theory (usually the bare mass) with an incredible precision. In four dimensions, this fine-tuning can be understood in terms of perturbative quadratic divergences.
In the renormalization group (RG) approach of field theory, the need for fine-tuning can be related to the existence of unstable directions in a way which is independent of perturbation theory in a particular dimension. In this approach, the renormalized mass expressed in cut-off units decreases exponentially with the number of iterations spent near an unstable fixed point. In order to keep the mass small, one needs to fine-tune the initial parameter in order to start close to the critical hypersurface. This hypersurface in the space of bare Lagrangians separates the symmetric phase from the symmetry broken phase and contains the unstable fixed point.
In the following, we use the statistical mechanics language where criticality is approached by tuning the inverse temperature $`\beta `$ close to its critical value $`\beta _c`$. In terms of this parameter, the ratio of the physical mass $`m`$ and the ultra-violet (UV) cutoff $`\mathrm{\Lambda }`$ is
$$m/\mathrm{\Lambda }(\beta _c\beta )^{\gamma /2},$$
(1)
where $`\gamma `$ is the critical exponent associated with the magnetic susceptibility. In four dimensions, $`\gamma =1`$. If we take $`m=100GeV`$, a typical electroweak scale, and $`\mathrm{\Lambda }=10^{19}`$GeV of the order of the Planck mass, we need to fine-tune $`\beta `$ with 34 digits.
The main virtue of the RG approach is to introduce some hierarchy in the information contained in the partition function. At each iteration, the information relevant to understand the large distance behavior is amplified, while the rest of the information is discarded according to its degree of irrelevance. However if some “noise” is introduced in this process, for instance as numerical errors in the calculation, the error in the relevant direction will be amplified too. This may lead to situations where the amplified errors wipe out the careful fine-tuning and one obtains meaningless results.
In this article, we discuss the numerical errors with examples where the RG transformation can be calculated for many iterations. The models used for our calculations are Dyson’s hierarchical model , for which very accurate methods of calculation have been recently designed, and a simplified version of this model for which the nonlinear aspects of the interpolation between fixed points is understood in great detail. For comparison, we also considered a random number generator designed to produce a Gaussian distribution.
The question of numerical errors in RG calculation has been briefly discussed in recent publications. The mechanism of error amplification was identified in section IV of Ref. . In these calculations, we probed the numerical errors by slightly varying the parameter (denoted $`s`$ hereafter) used to rescale the fields at each iteration of the RG transformation and exploiting the fact that the physical quantities such as the magnetic susceptibilty $`\chi `$ are independent of the choice of rescaling. This procedure is reviewed in section III. We predicted that if $`\delta `$ represents a typical round-off error (of the order of $`10^{16}`$ in double precision calculations), the relative difference between the numerical values obtained for two slightly different values of $`s`$ should be of the order
$$\mathrm{\Delta }\chi /\chi \delta (\beta _c\beta )^1.$$
(2)
This behavior was observed in good approximation for a wide range of $`\beta `$ (see Fig. 4 of Ref. ). A consequence of this result is that at fixed $`\delta `$ and $`m`$, there is a maximum UV cutoff of the order of $`m\delta ^{\gamma /2}`$ beyond which the result of a calculation becomes totally meaningless.
In the models we have considered, the exact rescaling factor $`s`$ for which the RG transformation has a non-trivial fixed point (denoted $`s_{exact}`$ hereafter) is exactly the same as for a free theory. In models with nearest neighbor interactions, this is only approximately the case. The free value is corrected whenever the critical exponent $`\eta `$ is non-zero. Since in most cases, $`\eta `$ is only known with a finite accuracy, one is naturally led to consider a range of values for $`s`$. In section III.b of Ref. we considered the distribution of values of the susceptibility for 2000 values of $`s`$. To our surprise, we found that the errors did not average out to the correct value which was calculated using higher precision arithmetic. In addition, we found that the distribution of errors was not Gaussian. In the present paper, we readdress these questions with larger statistics and we compare the results with other models. The technical details of our analysis are outlined in section II.
Our first result is that the average error is of the same order of magnitude as the standard deviation of the distribution. There is a systematic bias in the calculations and one does not gain anything by increasing the statistics. This bias depends on the peculiarities of arithmetic round-offs and we have not attempted to model it from a “microscopic” point of view. It should however be noticed that the average error is “compiler robust” and that it should in principle be understandable at least for the simplified model for which only a few arithmetic operations are needed at each iteration of the RG transformation.
Another robust result is that all our RG calculations show significant departures from Gaussian behavior. These departures can be estimated in terms of the skewness and kurtosis coefficients. In all the RG calculations these coefficients are at least one order of magnitude larger than the corresponding quantities for the Gaussian random number generator.
How do the repeated errors fail to erase the details of the individual distribution and produce a Gaussian distribution as in the central limit theorem? We have considered two possible explanations. The first one is that the distribution of errors changes from step to step. A detailed study shows that there are indeed small variations in these distributions, however this might not be the most important effect. The main reason why the central limit does not apply seem to be that the errors are added with unequal weight. The errors made during the initial iterations are more amplified than errors made at later stage. As a consequence, the “weighted average” inherits the skewness and kurtosis of the individual distribution. Since these have “short tails”, so does the distribution of errors, unlike non-Gaussian distributions found in the study of turbulence which have “long tails”.
Our results impose limitations on numerical approaches of scalar field theory, however these are not drastic. The main result is that as far as the average of low powers of the total field is concerned, a hierarchy $`\mathrm{\Lambda }/m`$ requires a number of significant digits proportional to Log($`\mathrm{\Lambda }/m`$), which is not prohibitive for $`\mathrm{\Lambda }/m10^{17}`$.
## II Technical outline of the paper
The renormalization group (RG) approach of scalar field theory, the hierarchy problem and our strategy to probe the numerical errors are reviewed in generic terms in section III. The specific models used for our calculations are presented in section IV.
In section V, we consider seven samples with $`10^5`$ data points, six from RG calculations and one designed to produce a Gaussian distribution. It should be noted that in all cases randomness is due to sensitive dependence on the initial conditions. We show that the departure from Gaussian behavior of the distributions of RG values are significantly larger than in the case of the random number generator designed to produce a Gaussian distribution. We also show that the errors spread differently when $`s<s_{exact}`$ and $`s>s_{exact}`$, indicating that the sample should be “resolved” into subsamples with different distributions.
Subsamples are analyzed quantitatively in section VI in terms of “uniformity indicators” designed to establish correlations between the values of $`s`$ and the moments of the distribution. These indicators take values close to 1 when the sample is obtained from independent and identically distributed random variables. It is found that samples with only $`s<s_{exact}`$ or only $`s>s_{exact}`$ have much better uniformity than the the samples combining both sets of values. In the case of the simplified model, the samples with only $`s<s_{exact}`$ or only $`s>s_{exact}`$ have values of the indicators which seem consistent with being samples coming from independent and identically distributed random variables. For the hierarchical model, the analogous samples need further resolution.
It should be noted that in general one expects fluctuations of the order of (spread of the parent distribution)/((size of the sample)<sup>1/2</sup>). Since we use large samples, these fluctuations are expected to be small. Compared to this small scale, the variations of the estimators for $`s<s_{exact}`$ and $`s>s_{exact}`$ are large. However, they are still relatively small when compared to the spread itself. Consequently, it still makes sense to talk in an approximate way about, for instance, the mean of the distribution. To take an example, one can look at Table I and see that the average errors for $`s<s_{exact}`$ (803) and $`s>s_{exact}`$ (1297) differ by approximately one quart of the standard deviation of the any of the two samples. Consequently, we can still say that for any value of $`s`$, the error is approximately 1000. These remark should be kept in mind while we use expressions such as “the moments of a distribution”.
A specific model where the errors made during the initial iterations are more amplified than errors made at later stage is provided in section VII. In the conclusions, we discuss the limitations that our results impose on numerical approaches of scalar field theory. The situation is compared to what happens in chaotic dynamical systems and in the study of turbulence.
## III General strategy to probe the numerical errors
Before getting into the specifics of model calculations, we would like to describe in general terms, our method to probe the numerical errors. The main points of this section are the following. First, the RG transformation involves a rescaling of the sum of the fields in boxes of increasing sizes. In general, this rescaling factor is only known with a finite accuracy and so some range of values should in general be considered. Second, the physical quantities are independent of the choice of this rescaling factor. However in practical numerical calculations, the arithmetic errors are different for different choices of rescaling factor. Consequently, we can use large samples of rescaling factors to obtain a statistical distribution of these errors. We now proceed to discuss these points in the case of a generic scalar field theory.
Let us consider a scalar model in $`D`$-dimensions with a lattice spacing $`a_0`$. We first integrate the fields in blocks of side $`ba_0`$ while keeping the sum of the fileds in the block constant. We then divide the sums of the fields by a factor $`b^{(2+D\eta )/2}`$ and treat them as our new field variables. The procedure defines a discrete renormalization group transformation provided that the scale factor $`b`$ is real number strictly larger than 1.
The critical hypersurface (in the space of bare Lagrangians) is given as the stable manifold (e.g. the basin of attraction) of a non-trivial fixed point of this transformation. The stable manifold can be reached by considering a family of models indexed by a parameter which can be tuned in order to cross the stable manifold. In field theory context, one usually pick the bare mass to accomplish this purpose. In the statistical mechanics formulation, the inverse temperature $`\beta `$ can be tuned to its critical value $`\beta _c`$ which is a function of the other interactions. This notation will be used later.
The information that we are keeping during the renormalization group transformation is encoded in the average values of all the integer powers of the sum of the fields in the blocks. We call these average values the “zero momentum Green’s functions at finite volume”. This set of values can be thought of as an element of an infinite vector space. Near the fixed point, we can use the eigenvectors of the linearized transformation as a basis. As far as we are close to the fixed point, the average values of the powers of the rescaled total field stay approximately unchanged after one transformation. However at each iteration, the components in the eigendirections are multiplied by the corresponding eigenvalue. In the following, we discuss the case where there is only one relevant direction, in other words, only one eigenvalue larger than 1. We call this eigenvalue $`\lambda `$.
After repeating the renormalization group transformation $`n`$ times, we have replaced $`b^{Dn}`$ sites by one site and associated a block variable with it. If at this point we neglect the interactions among the blocks of size $`b^{Dn}`$ and larger, we can calculate the finite volume volume Green’s functions. For the sake of definiteness, we only discuss the case of the two point function. In the statistical mechanics language, the zero-momentum two point function is called the magnetic susceptibility. We define the finite volume susceptibility $`\chi _n`$ as the average value of the square of the sum of all the (unrescaled) fields divided by the number of sites $`b^{Dn}`$.
From the above discussion, we estimate
$$\chi _nb^{n(2\eta )}(K_1+K_2\lambda ^n(\beta _c\beta ))$$
(3)
The power of $`b`$ comes from rescaling back two powers of the fields to their original values, together with the division by the number of sites. The constant $`K_1`$ is the constant value of the average of the square of the rescaled sum of the fields at the fixed point. The constant $`K_2`$ depends on the way the critical hypersurface is approached when $`\beta `$ is varied close to $`\beta _c`$.
We want to emphasize that Eq. (3) is valid only if the linearization procedure is applicable, in other words if $`\lambda ^n(\beta _c\beta )<<1`$. On the other hand, when $`n`$ reaches some critical value $`n^{}`$ such that
$$\lambda ^n^{}(\beta _c\beta )1,$$
(4)
non-linear effects become important and the sign of $`(\beta _c\beta )`$ becomes important. In the following we will consider exclusively the case of the symmetric phase which is simpler. For a discussion in the case of the broken symmetry phase ($`\beta >\beta _c`$), the reader may consult Ref. . If $`\beta <\beta _c`$, the value of $`\chi `$ starts stabilizing when $`n`$ gets of the order of $`n^{}`$. As a consequence,
$$\chi _{\mathrm{}}b^{n^{}(2\eta )}(\beta _c\beta )^\gamma ,$$
(5)
with
$$\gamma =(2\eta )\mathrm{ln}b/\mathrm{ln}\lambda .$$
(6)
When $`n`$ gets larger, the high-temperature fixed point is reached rapidly. This fixed point is completely attractive. The irrelevant directions are manifested by volume effects decreasing like $`b^{n2}`$. A model calculation of these effects can be found in Ref. .
For most models studied in the literature, the exact rescaling factor $`s=b^{(2+D\eta )/2}`$ is not known with perfect accuracy. For instance in a Monte Carlo calculation for values of the couplings minimizing the subleading corrections, the values of $`\eta `$ obtained (for two-components in 3 dimensions) is 0.0381 with errors of order 2 in the last quoted digit. Since at the end of the calculation we are rescaling back to the original field variables, this uncertainty would not affect the physical results provided that we were able to carry the integrations exactly. If the integrations are carried numerically, the arithmetic operations are performed differently and choosing a sample of values for the rescaling factor $`s`$ close to the approximate values can be used to obtain a statistical sample of the numerical errors.
We will thus consider values of the rescaling factor $`s`$ of the form $`s_{exact}+\delta `$, where $`s_{exact}`$ is the value for which there is an exact fixed point. In some sense $`\delta `$ can be seen as the “seed” of a random number generator. We will now perform sample calculations for two models and compare our results with the results obtained from a set of seeds for a random number generator designed to produce approximate Gaussian distributions.
## IV Model Calculations
In this section, we describe in detail the three procedures used to obtain the data analyzed in the following sections. We discuss the hierarchical model (subsection A), a simplified version of it (subsection B) and a random number generator which represents our naive expectations for the two other cases (subsection C).
### A The hierarchical model
The choice of the hierarchical model allows easy calculations with a controllable accuracy. In order to avoid repetitions, we refer the reader to Ref. for a more systematic presentation. The main interest of this model is that only the local potential (or equivalently, the part of the measure which factorizes into a product of identical local functions which are called “the local measure” later) is affected by the RG transformation while the “kinetic” part is left invariant.
The block-spin procedure can then be summarized by a simple integral formula (Eq. (2.2) in Ref. ) for the local measure. Taking the Fourier transform and rescaling the sum of the fields in the block by an arbitrary rescaling factor $`1/s`$ one obtains the recursion relation
$$R_{n+1}(k)=C_{n+1}exp(\frac{1}{2}\beta (\frac{c}{4}s^2)^{n+1}\frac{^2}{k^2})(R_n(\frac{k}{s}))^2.$$
(7)
The parameter $`c`$ is set equal to $`2^{12/D}`$ in order to approximate a $`D`$-dimensional model with nearest neighbor interactions. In the following the value $`D=3`$ will be used.
We fix the normalization constant $`C_n`$ is such way that $`R_n(0)=1`$. $`R_n(k)`$ has then a direct probabilistic interpretation. If we call $`M_n`$ the total field $`\varphi _x`$ inside blocks of side $`2^n`$ and $`<\mathrm{}>_n`$ the average calculated without taking into account the interactions among different blocks of this size find
$$R_n(k)=\underset{q=0}{\overset{\mathrm{}}{}}\frac{(ik)^{2q}}{2q!}\frac{<(M_n)^{2q}>_n}{s^{2qn}}$$
(8)
We see that the Fourier transform of the local measure after $`n`$ iterations generates the zero-momentum Green’s functions calculated with $`2^n`$ sites. In particular, we are interested in calculating the finite volume susceptibility
$$\chi _n=\frac{<(M_n)^2>_n}{2^n}=2a_{n,1}(\frac{s^2}{2})^n.$$
(9)
As far as we are only interested in the calculation of $`<(M_n)^{2q}>_n`$, the choice of $`s`$ is a matter of convenience. For the calculations in the high temperature phase (symmetric phase) not too close to the critical points, or high temperature expansions the choice $`s=\sqrt{2}`$ is natural . On the other hand, the the choice of rescaling factor $`s=2c^{1/2}`$ makes the explicit dependence on $`n`$ in Eq. (7) disappear. For any other value of $`s`$, the map is somehow analogous to a differential equation with explicitly time-dependent coefficients. What we call “the RG transformation” in section III, is Eq. (7) with $`s=2c^{1/2}`$. It corresponds to the values $`b=2^{1/D}`$, $`\eta =0`$ and $`s=b^{(2+D)/2}`$ of the parameters defined in section III. For this value of $`s`$, Eq. (7) has non-trivial fixed point which seems to be unique.
We have calculated the susceptibility for the Ising measure ($`R_0(k)=cos(k)`$) with $`D=3`$ (i.e. $`c=2^{1/3}`$) and $`\beta =\beta _c10^8`$. The calculations have been performed using double precision Fortran. Using the approximate error given in Eq. (2), we see that 8 out of the 16 significant digits of $`\chi `$ should be correct for $`\beta =\beta _c10^8`$. Since we are in the high-temperature phase, $`\chi _n`$ stops growing after approximately $`n^{}52`$ iterations (see section III and Ref. for details). The 16 digits of $`\chi _n`$ are completely stabilized after $`n=140`$ iterations. We call this stable value $`\chi (s)`$ where $`s`$ refers to the value of $`s`$ in Eq. (7) used for the numerical calculation. The calculations have performed using dimensional approximations of degree $`l_{max}`$:
$$R_n(k)=1+a_{n,1}k^2+a_{n,2}k^4+\mathrm{}+a_{n,l_{max}}k^{2l_{max}},$$
(10)
for which the recursion formula for the $`a_{n,m}`$ reads :
$$a_{n+1,m}=\frac{\underset{l=m}{\overset{l_{max}}{}}(\underset{p+q=l}{}a_{n,p}a_{n,q})[(2l)!/(lm)!(2m)!](c/4)^l[(1/2)\beta ]^{lm}}{_{l=0}^{l_{max}}(_{p+q=l}a_{n,p}a_{n,q})[(2l)!/l!](c/4)^l[(1/2)\beta ]^l}$$
(11)
We found that the 16 digits of $`\chi _n`$ were completely stabilized for $`l_{max}=50`$.
We have used the set of values $`s=2/\sqrt{c}\pm m\times 10^8`$ with $`m=1,\mathrm{}10^5`$. We recorded the difference of $`\chi (s)`$ defined above with respect to the accurate value $`\chi =5.2316268857268\times 10^{10}`$. This value was obtained by performing the calculation using higher precision arithmetic (namely 30 digits). We have checked that this accurate result is insensitive to changes in $`s`$. The results are discussed in the next sections.
### B A simplified model
As noticed in Ref. , for $`\beta <\beta _c`$ and $`n>>n^{}`$ Eq. (7) imples the approximate behavior
$$\chi _{n+1}\chi _n+(\beta /4)(c/2)^{n+1}\chi _n^2.$$
(12)
This map has been studied on its own in the rescaled form
$$h_{n+1}=(c/2)h_n+(1c/2)h_n^2.$$
(13)
This map can be seen as a drastically simplified version of the hierarchical model. One can check that if $`0h_01`$, lim$`{}_{n>\mathrm{}}{}^{}(2/c)_{}^{n}h_n`$ is finite. This limit plays the role of the susceptibility in the following and can be calculated accurately by combining dual expansions .
Eq. (13) has an unstable fixed point at $`h=1`$ with eigenvalue $`\lambda =2c/2`$. We have required $`\lambda =1.427`$, approximately as for the hierarchical model with $`D=3`$, in order to keep the value of $`n^{}`$ the same. Consequently, the value of $`c`$ used in Eq. (13) is not the same as the value of $`c`$ for the hierarchical model with $`D=3`$.
We have introduced the rescaling factor $`s`$ through the redefinition
$$a_n(s^2c/4)^nh_n,$$
(14)
in terms of which the map becomes
$$a_{n+1}=(2/s^2)a_n+(1c/2)(s^2c/4)^{n1}a_n^2$$
(15)
After calculating, $`a_n`$ one can always return to $`h_n`$ using Eq. (14). If we had the chance to use exact arithmetic the expression would be independent of $`s`$.
We have performed calculations for $`h_0=a_0=110^8`$. We found that 150 iterations were sufficient to stabilize lim$`{}_{n>\mathrm{}}{}^{}(2/c)_{}^{n}h_n`$. We have calculated this value for $`s=2/\sqrt{c}\pm m\times 10^8`$ with $`m=1,\mathrm{}10^5`$, as in the previous case. We have then subtracted the more accurate value
$$\mathrm{lim}_{n>\mathrm{}}(2/c)^nh_n=3.842965603774557\times 10^{12}.$$
(16)
obtained and checked exactly with the same procedure as in the previous subsection.
### C A model with gaussian distribution
In order to provide a comparison of the errors distributions of the two previous models, we have also generated $`10^5`$ numbers using a method designed to give a gaussian distribution. Since in the two previous cases we have approximately $`n^{}=52`$ random processes before the value recorded starts stabilizing, we have added 52 random numbers. These random numbers have been produced by using repeated multiplication by a large number ($`7^5`$) followed by a reduction modulo 1 (in other words, we drop the integer part). This procedure is inspired by results reviewed in Ref. . Iterating this procedure, we generate a sequence of random numbers which we expect to be uniformly distributed between 0 and 1. In order to get numbers distributed between -1 and 1, we multiply each number by 2 and subtract 1. Finally, in order to get numbers approximately of the same order as the numbers of the other two sets, we have multiplied the final sum of 52 random numbers by 1000. The final results depends only on the initial number provided (the “seed”). We have repeated this calculation for $`10^5`$ values of the seed between 0 and 1.
The above discussion can be summarized as follows. We iterate 52 times the map
$$\alpha _{n+1}=7^5\alpha _n\mathrm{Modulo}1,$$
(17)
and then calculate
$$X_j=1000\times \underset{n=1}{\overset{52}{}}(2\alpha _n1).$$
(18)
The sub-index $`j`$ corresponds to different values of the seed $`\alpha _0`$. The data analyzed in the next sections corresponds the choice
$$\alpha _0(j)=j/100005.23,$$
(19)
with $`j=1,\mathrm{}10^5`$. The choice of the denominator is motivated by the fact that we want a spacing between successive initial values slightly smaller than $`10^5`$. The decimal points (.23) have been adjusted empirically in order to get a good uniformity of in subsamples (this question is discussed in section VI B).
The map of Eq. (17) is designed to provide a sample from a variable $`\alpha `$ which we expect to be uniformly distributed between 0 and 1. If this is the case, we should have $`(2\alpha 1)=0`$ and Var$`(2\alpha 1)=1/3`$. We use the common notation Var$`(A)=(AA)^2`$. If we now define, $`X=1000\times _{n=1}^{52}(2\alpha _n1)`$, we expect
$`X`$ $`=`$ $`0;`$ (20)
$`\mathrm{Var}(X)`$ $`=`$ $`(1000)^2\times (52/3)(4163)^2.`$ (21)
These predictions will be tested in section VI.
## V Error distributions
In order to get a first idea about what can be done to characterize the distributions of values obtained by the procedures described in the previous section, we first consider the hierarchical model and display a subsample of 1000 data points for $`s`$ below $`s_{exact}=2/\sqrt{c}`$ and 1000 data points for $`s`$ above this value. The selection of the subsample was done by taking one out of every 100 values out of the original sample. More precisely, the integer $`m`$ used in the parametrization of $`s`$ given in subsection IV A, takes only values which are multiples of 100. The distribution of values is shown in Fig. 1.
One immediately realizes that the distribution is not symmetric about $`s=2/\sqrt{c}`$. If $`s>2/\sqrt{c}`$, the values of $`\chi `$ are more spread than if $`s<2/\sqrt{c}`$. An histogram can be obtained by dividing the data points into “horizontal bins” of equal vertical height and counting the number of data points in each bin. This procedure has been followed for 50,000 data points with $`s>2/\sqrt{c}`$ and 50,000 data points with $`s<2/\sqrt{c}`$. Each of the two sets is obtained by taking every other point out of the data for the hierarchical model described in the previous section. With this procedure the total number of data points is still $`10^5`$ and the statistical properties can be easily compared with the other samples of $`10^5`$ data points. The histogram is displayed in Fig. 2.
The solid line is obtained by plugging the estimated mean and standard deviation (discussed in the next section) in a gaussian probability distribution. In the rest of this section, the terminology “Gaussian fit” refers to this procedure. One sees that there are large deviations from the Gaussian fit. Given the size of the sample it is very unlikely that the deviations can be interpreted as statistical fluctuations.
For comparison, we give in Fig. 3 an histogram of the distribution of $`10^5`$ data points obtained from the random number generator discussed in section IV C. The features of the distribution can be seen better by plotting the logarithm of the number of data points in each bin. In this semi-log plot, the fit corresponding to a Gaussian distribution is simply a parabola. Also, we will study separately, the data points with $`s<2/\sqrt{c}`$ and $`s>2/\sqrt{c}`$ since from Fig. 1 they have manifestly different standard deviations. The results for the two sets of $`10^5`$ data points described in subsection IV A for the hierarchical model are shown in Fig. 4.
These distributions are roughly of rectangular shape with slow modulations on the “flat” part. They can be compared with a similar graph for the random number generator (Fig. 5). On this graph, the departure from the parabolic behavior is barely visible except in the tails. Obviously, bins with zero data points cannot be shown in such a graph.
Finally, we have displayed the results for the two sets of $`10^5`$ data points described in subsection IV B for the toy model (Fig. 6). One sees that significant deviations from the Gaussian fit appear in the tails. The fact that the probability distribution falls more rapidly in this region is indicative of a probability distribution of the form exp$`(ax^2bx^4)`$ with $`a`$ and $`b`$ positive. This relative simplicity suggests that we have a better chance to fully understand this simplified example.
## VI Moment Analysis
### A Moment estimators
A probability distribution is characterized by its moments. We would like to estimate the first moments of the distributions described in the previous sections assuming that they are sample of a unique probability distribution. Let $`X_i`$ with $`i,\mathrm{}N`$ be a set of independent and identically distributed (i.i.d.) random variables with moments
$`X_i`$ $`=`$ $`\mu ;`$ (22)
$`(X_i\mu )^r`$ $`=`$ $`\mu _r,r2,`$ (23)
identical for any $`i`$. In particular, $`\mu _2=\sigma ^2`$ is the variance or the square of the standard deviation $`\sigma `$. We define the estimators
$`\widehat{\mu }`$ $`=`$ $`(1/N){\displaystyle \underset{1}{\overset{N}{}}}X_i;`$ (24)
$`\widehat{\mu _r}`$ $`=`$ $`(1/N){\displaystyle \underset{1}{\overset{N}{}}}(X_i\widehat{\mu })^r,r2.`$ (25)
Using the hypothesis that the random variables are independent, we can factor the powers of a given $`X_i`$ and calculate its average independently. For instance, if $`ij`$, $`X_iX_j=X_iX_j`$. Using the hypothesis that the random variables are identically distributed, we can use Eq. (23) to express these expectations in terms of the common values of their moments. Using these rules, one obtains
$`\widehat{\mu }`$ $`=`$ $`\mu ;`$ (26)
$`\widehat{\mu _r}`$ $`=`$ $`\mu _r+O(1/N),r2.`$ (27)
The biases of the $`\mu _r`$ are of order $`1/N`$. It is not difficult to remove the bias, however since we will work with large samples, the corrections are very small.
In Table I, we give the estimated values of $`\mu `$ and $`\sigma `$ for the distributions discussed before. The abbreviation a. is short for “above” which means the $`10^5`$ points with $`s>2/\sqrt{c}`$ for the hierarchical model (H.M.) ot the toy model (T.M.). Similarly b. is short for “below” ($`s<2/\sqrt{c}`$), while a.+b. means “above and below” which is short for $`10^5`$ points obtained by taking every other value in the “above” and “below” set as explained at the beginning of section V. Finally, “R.N.G.” is short for the random number generator described in subsection IV C. These notations will be used again in the following. We emphasize that a.+b. is not the union of a. and b. (so $`\mu _{a.+b.}(1/2)(\mu _{a.}+\mu _{b.}`$)) and that all the samples have $`10^5`$ data points.
Initially we expected that $`\mu `$ would be of order $`\sigma /\sqrt{N}`$. However, this is only the case for the random number generator. In all the other cases, $`\mu `$, which we remind is the average difference with respect to the accurate value, is of the same order of magnitude as $`\sigma `$. In other words, it seems impossible to get rid of the errors by using large statistics!
In order to check the compiler-dependence of our results, we have repeated the calculations of the toy model in C, Mathematica and a different version of Fortran. We found that $`\mu `$ was only affected by less than 10 percent. On the other hand, $`\sigma `$ varied much more significantly.
Note also that in the case of the R.N.G., the estimated value of $`\sigma `$ is close to the result (4163) obtained in Eq. (21) by assuming that the $`\alpha _l`$ are uniform over the interval $`[0,1]`$.
As expected, the $`\sigma `$ “above” and “below” are significantly different. It is very unlikely that the data “above and below” is a sample from a unique distribution. More generally, one could study subsamples and check that the fluctuations are compatible with their size. This is the topic of the next subsection.
### B Fluctuations and uniformity
In the previous subsection, we have used moments estimators which are in good approximations unbiased. However, it is not clear that the sample comes from a unique distribution. One way to tackle this question is to consider the estimators in many subsamples and decide if the the fluctuations of the estimated values of the moments in the subsamples are compatible with the variance of the estimator which we now proceed to estimate.
Using again the hypothesis of independent and identical distributions, we find in leading order in $`1/N`$ (in other words, up to $`O(1/N^2)`$ corrections) that:
$`Var(\widehat{\mu })`$ $`=`$ $`(1/N)(\mu _2);`$ (28)
$`Var(\widehat{\mu _2})`$ $`=`$ $`(1/N)(\mu _4\mu _2^2);`$ (29)
$`Var(\widehat{\mu _3})`$ $`=`$ $`(1/N)(\mu _6\mu _3^2+9\mu _2^36\mu _2\mu _4);`$ (30)
$`Var(\widehat{\mu _4})`$ $`=`$ $`(1/N)(\mu _8\mu _4^2+16\mu _2\mu _3^28\mu _3\mu _5).`$ (31)
The proof of these results can be found for instance in Ref. .
We considered the seven sets of $`10^5`$ data points discussed in the previous subsection. Each set has been divided into $`m`$ subsets with $`N_B`$ data points. Obviously, $`N=m\times N_B`$. The partition has been done by putting together successive values of $`s`$. If one refers to Fig. 1, the subsets are vertical partitions. We call $`\widehat{\mu _{r,S}}`$ (or $`\widehat{\mu _S}`$) the estimator of $`\mu _r`$ (or $`\mu `$) in the $`S`$-th subsample. These are defined as in Eq. (25), except for the fact that $`N`$ needs to be replaced by $`N_B`$. In order to avoid confusion, we have used a subscript $`T`$ (short for Total) to designate the estimators in the whole sample. We have studied the differences between the subsample estimates and the whole sample estimates. In order to get comparable answers, we have expressed these differences in “natural units”. These units should be such that if we had a sample from a unique distribution, all the fluctuations would be of order 1. From the expressions of the variances, ones sees that $`\sigma ^r/\sqrt{N_B}`$ are natural units for the fluctuations of $`\widehat{\mu _{r,S}}`$. In practice, we have to replace $`\sigma `$ by its estimated value. As an illustration, we have taken three sets previously abbreviated as H.M.a.+b., T.M.a.+b. and R.N.G. and divided them into $`m=400`$ sub-bins. These sub-bins are ordered according to increasing values of $`s`$. For illustration, we have displayed the fluctuations of the second moment in natural units. The results are shown in Fig. 7.
One sees that for the first two sets, there is a “jump” at $`s=2/\sqrt{c}`$ that is significantly larger than the other fluctuations. Note that we have checked that both a.+b. sets had their sub-bins ordered in the same way. Besides these jumps, the fluctuations appear to be of much more “normal” size. In order to make this visual impression quantitative, we have defined the average of the square of the fluctuations in the subsamples:
$`U_1`$ $`=`$ $`(1/m){\displaystyle \underset{S=1}{\overset{m}{}}}(\widehat{\mu }_S\widehat{\mu _T})^2;`$ (32)
$`U_r`$ $`=`$ $`(1/m){\displaystyle \underset{S=1}{\overset{m}{}}}(\widehat{\mu _{r,S}}\widehat{\mu _{r,T}})^2.`$ (33)
If the sample comes from a single distribution, we expect these quantities to be equal to the variance of the corresponding estimator with fluctuations of order $`1/\sqrt{m}`$. We have thus defined some “indicators of uniformity” as
$`p_1=U_1/\mathrm{Var}(\widehat{\mu }_S);`$ (34)
$`p_r=U_r/\mathrm{Var}(\widehat{\mu _{r,S}}).`$ (35)
If the sample comes from a single distributions, the $`p_i`$ should be close to 1, while a large value indicates the presence of several distributions. The values of these indicators are given for the seven sets of data points previously described in Table II.
First, one notices that the R.N.G. data has all the $`p_i`$ within less than ten percent of 1. On the other hand, both a.+b. sets have large values. For the toy model, the separation into above and below is sufficient to get rid of these large values. The values for T.M.a. and T.M.b are within 20 percent of 1. This seems in good approximation compatible with a single distribution. This quantitative analysis confirms the visual impression that one gets from Figs. 4 and 6. On the other hand, it is possible to further “resolve” the a. and b. data for the hierarchical model.
In the case of the hierarchical model, the local analysis of the sub-samples shows that the rapid variations of the mean has “large-scale” tendencies. Namely we observed regions of $`s`$ where in average, the mean grows linearly when $`s`$ increases followed by regions where it decreases. This behavior is presently under study.
### C Departure from Gaussian behavior
In section V, we noticed that the distributions for the toy model a. and b. were more sharply peaked than a Gaussian distribution. This feature can be analyzed more quantitatively by estimating the skewness coefficient $`\mu _3/\mu _2^{3/2}`$ and the coefficient of kurtosis $`\mu _4/\mu _2^23`$. For a Gaussian distribution, both coefficients are zero. These estimations of these coefficients for the data sets previously considered are given in Table III.
For the T.M. a. and b., both coefficients are an order of magnitude larger than the corresponding coefficients for the R.N.G. . Larger deviations are observed for the other sets reflecting the lack of uniformity of these sets of data.
In summary, the above table supports the general statement that the numerical errors appearing in RG calculations are non-Gaussian. We will now attempt to explain this general feature with a simple model.
## VII A model for non-Gaussian errors
One reason to be inclined to believe a-priori that errors distributions should be Gaussian is the famous central-limit theorem which asserts that the average of $`N`$ i.i.d. random variables approaches a Gaussian distribution when $`N`$ becomes infinite. In this section, we give an heuristic derivation of this theorem and then present a mechanism by which we can avoid ending up with Gaussian distributions.
### A The central limit theorem
We first review the central-limit theorem in terms of a moment analysis. Let $`\alpha _j`$ $`j=1,\mathrm{}n`$ be a set of i.i.d. random variables with arbitrary moments ($`\mu ,\mu _2=\sigma ^2,\mu _3\mathrm{}`$). We define
$$Z=[(1/n)\underset{j=1}{\overset{n}{}}(\alpha _i\mu )]/(\sigma /\sqrt{n}).$$
(36)
It is clear that $`Z`$ has $`\mu =0`$ and $`\sigma =1`$. The central-limit asserts that
$$\mathrm{Lim}_{n>\mathrm{}}P(Z)=(1/\sqrt{2\pi })e^{Z^2/2}.$$
(37)
This theorem can be proven by showing that the moments of $`Z`$ coincide with the moments of a normal Gaussian :
$`\mu _{2r}`$ $`=`$ $`(2l1)!!;`$ (38)
$`\mu _{2r+1}`$ $`=`$ $`0.`$ (39)
The calculation of the expected values of $`Z^q`$ goes as follows. Each term of $`Z^q`$ is a product of $`q`$ terms of the form $`(\alpha _i\mu )`$. The expected value can be factored into products involving the same $`i`$ and then expressed in terms of the moments. Since $`\alpha _i\mu =0`$, each $`i`$ should be repeated at least twice. For $`Z^3`$, the three $`i`$ should be the same or one will be “left alone” (which would imply a vanishing average). Since the three indices have to be the same and since there are $`n`$ indices we get a combinatoric factor $`n`$ which is insufficient to overcome the $`n^{3/2}`$ appearing in $`Z^3`$. A similar reasoning in the case of $`Z^4`$, shows there are three ways to arrange four indices into two pairs having the same index (assuming that the index of the two pairs are distinct). If we require that the four indexes are the same, it costs an additional $`1/n`$ suppression. In summary, when $`n`$ becomes large
$`<Z^3>=n\mu _3/(\sigma ^3n^{3/2})>0;`$ (40)
$`<Z^4>=(n\mu _4+3n(n1)\sigma ^4)/(\sigma ^4n^2)>3.`$ (41)
The argument generalizes easily, and one realizes that the odd moments vanish and the even moment reduce to the number of unordered pairs, in agreement with Eq. (38).
### B Evading the central limit
In order to see how the central-limit can be evaded, we have constructed a simplified model for the numerical errors. In the rest of this section, when we say “the quantity we are calculating numerically” this means the $`R_n(k)`$ function defined in Eq. (7) for the hierarchical model or the quantity $`a_n`$ of Eq. (15) for the toy model. While we are near the fixed point, the numerical values appearing in these quantities which are relevant for our discussion are roughly of order 1.
We first assume that the initial data ($`R_0`$ or $`a_0`$) has an error $`_0`$ in the unstable direction. Using the linearized model of section III, we see from Eq. (3) that as far as we are near the fixed point, the quantity we are calculating numerically is of order $`K_1`$. After one iteration, $`_0`$ is amplified by a factor $`\lambda `$ In addition, an error of order $`K_110^{16}`$ is made. We express our ignorance about the details of the round-off errors in terms of a random variable $`\alpha _1`$ which takes positive or negative values of order 1 and write the new error as $`K_1\alpha _110^{16}`$. Putting the two terms we get
$$_1=\lambda _0+K_1\alpha _110^{16}.$$
(42)
Iterating $`n`$ times, we get
$$_n=K_1(\underset{j=0}{\overset{n}{}}\lambda ^{nj}\alpha _j)10^{16}.$$
(43)
In order to have compact notations, we have introduced a random variable $`\alpha _0`$ such that $`_0=K_1\alpha _010^{16}`$. This term is not very important for the rest of the discussion, we might have set $`_0=0`$ and considered errors for a given initial data expressed with a finite precision. In that case, the sum in Eq. (43) would run from 1 to $`n`$.
The main difference between the expression of the error of Eq. (43) and the variable $`Z`$ introduced in the discussion of the central-limit theorem, is that the $`\alpha _i`$ appear with different weights. One can then escape the “pair dominance” mechanism which puts all the individual errors on the same footing. Instead, the short-distance errors (small $`n`$) are more amplified than the large-distance errors (large $`n`$). Consequently, the total error is likely to inherit some of the features of the individual errors. This intuitive picture is confirmed by an explicit calculation that we now proceed to explain.
We will calculate the skewness and kurtosis associated with
$$Y=\underset{j=0}{\overset{n}{}}\lambda ^{nj}\alpha _j,$$
(44)
assuming that all the $`\alpha _i`$ are independent and identically distributed. A detailed study of the errors associated with one iteration of the RG transformation for the toy model shows that this assumption is not totally correct but that it is a reasonable first-order approximation. A simple calculation shows that after neglecting terms of order 1 compared to $`\lambda ^n`$, we obtain
$`<(Y<Y>)^3>/(<(Y<Y>)^2>)^{3/2}`$ $``$ $`[(\lambda ^21)^{3/2}/(\lambda ^31)](\mu ^3/\sigma ^{3/2});`$ (45)
$`[<(Y<Y>)^4>/(<(Y<Y>)^2>)^2]3`$ $``$ $`[(\lambda ^21)^2/(\lambda ^41)][(\mu ^4/\sigma ^4)3].`$ (46)
One sees that $`Y`$ inherits the non-Gaussian behavior of the $`\alpha `$’s, in the sense that the skewness and kurtosis coefficients of $`Y`$ are proportional to those of $`\alpha `$.
For $`\lambda =1.427`$ and $`\alpha `$ uniformly distributed between -1 and 1 (corresponding to an error of $`\pm 1`$ in the last significant digit) and which is roughly what we observed for the toy model, the skewness and kurtosis coefficients of $`Y`$ are 0 and -0.41 respectively. These numbers are consistent with the values obtained for the toy model. A more detailed study shows that if we replace the average in Eq. (18) for the random number generator by a weighted average as in Eq. (43), one obtains histograms very similar to those of T.M. a. or b. (see Fig. 6).
## VIII Conclusions
We have shown that in RG calculations, perturbations occurring at different scales fail to average each other out and that the final result is most affected by the peculiarities of the short-distance perturbations. In our study the perturbations are “anthropomorphic” in the sense that they are due to our calculation procedure rather then to natural phenomena. We have also shown that it is not possible to get rid of the effects by averaging over calculations at slightly different values of the rescaling parameter $`s`$ (which is not known exactly in most realistic situations). The only way the errors can be reduced to an acceptable level is by using enough significant digits in the arithmetic used in the calculations.
The situations is somehow similar to what occurs in chaotic dynamical systems. There, the sensitive dependence on the initial conditions, implies that for time scales large compared to the inverse Lyapounov’s exponents, one needs more significant digits than possibly achievable in order to keep track of a particular orbit. In the calculations performed here, the product of the inverse Lyapounov exponent (log($`\lambda `$)) and the time of calculation ($`n^{}`$) grows only like $`(2/\gamma )\mathrm{log}(\mathrm{\Lambda }/m)`$ and one can obtain reasonably precise calculations of the renormalized mass with $`(\mathrm{\Lambda }/m)10^{17}`$ by using only 40 significant digits (for $`\gamma =1`$). If we are interested in calculating the connected higher point functions, more digits are necessary because some digits are lost in the subtraction procedure as explained in Ref. . In conclusions, we see that if we are interested in the lowest point connected functions and if we have hierarchies of only 17 order of magnitudes, there is no serious limitation in our capability to calculate using the RG method.
Finally, it should be noteds that non-Gaussian distributions are a common feature in the study turbulence . A simple example is the distribution of transverse velocity increments between two points separated by a distance $`l`$ in a turbulent jet, denoted $`\delta v_{}(l)`$. Typically, the distribution starts developing “long tails” when $`l`$ becomes sufficiently small. In the case studied here, we have the opposite effect: our distribution have “short tails”. This is due to the fact that the “microscopic” errors are the individual round-offs which have no tails at all.
###### Acknowledgements.
This research was supported in part by the Department of Energy under Contract No. FG02-91ER40664. We thank A. Bhattacharjee and J. Scudder for discussions on non-Gaussian distributions.
|
warning/0005/nlin0005009.html
|
ar5iv
|
text
|
# Composition of Chaotic Maps with an Invariant Measure
## 1 INTRODUCTION
In the past twenty years dynamical systems, particularly one dimensional iterative maps have attracted much attention and have become an important area of research activity. One of the landmarks in it was introduction of the concept of Sinai-Ruelle-Bowen (SRB) measure or natural invariant measure. This is, roughly speaking, a measure that is supported on an attractor and also describe the statistics of the long time behavior of the orbits for almost every initial condition in the corresponding basin of attractor . This measure can be obtained by computing the fixed density of the so called Frobenius-Perron operator which can be viewed as a differential-integral operator, hence, exact determination of invariant measure of dynamical systems is rather a nontrivial task, such that invariant measure of few dynamical systems such as one-parameter family one-dimensional piecewise linear maps including Baker and tent maps or unimodal maps such as logistic map for certain values of its parameter, can be derived analytically. In most cases only numerical algorithms, as an example Ulam’s method are used for computation of fixed densities of Frobenius-Perron operator.
Authors in reference have given hierarchy of one parameter family of nonlinear maps of interval $`[0,1]`$ with an invariant measure. Here in this paper we generate new hierarchy of many-parameter family of maps of the interval $`[0,1]`$ with an invariant measure, by composition of the chaotic maps of reference . These maps are also defined as ratio of polynomials, where we have derived analytically their invariant measure for arbitrary values of the parameters. Using this measure, we have calculated analytically, Kolmogorov-Sinai entropy or equivalently positive Lyapunov characteristic exponent of these maps, where the numerical simulation approve the analytic calculation. Also it is shown that just like the maps of reference , they possess very peculiar property, that is, contrary to the usual, these maps do not possess period doubling or period-n-tupling cascade bifurcation to chaos, but instead they have single fixed point attractor at certain region of parameters values, where they bifurcate directly to chaos without having period-n-tupling scenario exactly at these values of parameter whose Lyapunov characteristic exponent begins to be positive.
The paper is organized as follows: In section II we introduce new hierarchy of family of many-parameters maps by composition of the chaotic maps of refernce . In Section III we show that the proposed anzats for the invariant measure of these maps are eigenfuntion of Ferobenios-Perron operator with largest eigenvalue $`1`$. Then in section IV using this measure we calculate Kolmogorov-Sinai entropy of these maps for an arbitrary value of parameters. In section V we compare analytic calculation with the numerical simulation. Paper ends with a brief conclusion.
## 2 Many-Parameter Families of Chaotic Maps
Let us consider the one-parameter families of chaotic maps of the interval $`[0,1]`$ given in reference , defined as the ratio of polynomials of degree $`N`$:
$$\mathrm{\Phi }_N(x,\alpha )=\frac{\alpha ^2\left(1+(1)^N{}_{2}{}^{}F_{1}^{}(N,N,\frac{1}{2},x)\right)}{(\alpha ^2+1)+(\alpha ^21)(1)^N{}_{2}{}^{}F_{1}^{}(N,N,\frac{1}{2},x)}$$
$$=\frac{\alpha ^2(T_N(\sqrt{x}))^2}{1+(\alpha ^21)(T_N(\sqrt{x})^2)},$$
(2-1)
where $`N`$ is an integer greater than one. Also
$${}_{2}{}^{}F_{1}^{}(N,N,\frac{1}{2},x)=(1)^N\mathrm{cos}(2N\mathrm{arccos}\sqrt{x})=(1)^NT_{2N}(\sqrt{x})$$
is hypergeometric polynomials of degree $`N`$ and $`T_N(x)(U_N(x))`$ are Chebyshev polynomials of type I (type II), respectively. Obviously these map the unit interval $`[0,1]`$ into itself. $`\mathrm{\Phi }_N(x,\alpha )`$ is (N-1)-model map, that is it has $`(N1)`$ critical points in unit interval $`[0,1]`$,(see Figure 4) since its derivative is proportional to derivative of hypergeometric polynomial $`{}_{2}{}^{}F_{1}^{}(N,N,\frac{1}{2},x)`$ which is itself a hypergeometric polynomial of degree $`(N1)`$, hence it has $`(N1)`$ real roots in unit interval $`[0,1]`$. Defining Shwarzian derivative $`S\mathrm{\Phi }_N(x,\alpha )`$ as:
$$S\left(\mathrm{\Phi }_N(x,\alpha )\right)=\frac{\mathrm{\Phi }_N^{\prime \prime \prime }(x,\alpha )}{\mathrm{\Phi }_N^{}(x,\alpha )}\frac{3}{2}\left(\frac{\mathrm{\Phi }_N^{\prime \prime }(x,\alpha )}{\mathrm{\Phi }_N^{}(x,\alpha )}\right)^2=\left(\frac{\mathrm{\Phi }_N^{\prime \prime }(x,\alpha )}{\mathrm{\Phi }_N^{}(x,\alpha )}\right)^{}\frac{1}{2}\left(\frac{\mathrm{\Phi }_N^{\prime \prime }(x,\alpha )}{\mathrm{\Phi }_N^{}(x,\alpha )}\right)^2,$$
with a prime denoting a single differentiation with respect to variable $`x`$, one can show that:
$$S\left(\mathrm{\Phi }_N(x,\alpha )\right)=S\left({}_{2}{}^{}F_{1}^{}(N,N,\frac{1}{2},x)\right)0.$$
Therefore, the maps $`\mathrm{\Phi }_N(x)`$ have at most $`N+1`$ attracting periodic orbits. As it is shown in reference , these maps have only single period one stable fixed points.
Using the above hierarchy of family one-parameter of maps we can generate new hierarchy of family many-parameters chaotic maps with an invariant measure simply from the composition of these maps. Hence considering the functions $`\mathrm{\Phi }_{N_k}(x,\alpha _k),k=1,2,\mathrm{},n`$ we denote their composition by: $`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x)`$ which can be written in terms of them in the following form:
$`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x)=\stackrel{n}{\stackrel{}{\left(\mathrm{\Phi }_{N_1}\mathrm{\Phi }_{N_2}\mathrm{}\mathrm{\Phi }_{N_n}(x)\right)}}=`$
$`\mathrm{\Phi }_{N_1}(\mathrm{\Phi }_{N_2}(\mathrm{}(\mathrm{\Phi }_{N_n}(x,\alpha _n),\alpha _{(n1)})\mathrm{},\alpha _2),\alpha _1)`$ (2-2)
Since these maps consist of the composition of the $`(N_k1)`$-modals $`(k=1,2,\mathrm{},n)`$ maps with negative Shwarzian derivative, therefore, they are $`(N_1N_2\mathrm{}N_n1)`$ modals map and their Shwarzian derivative is negative too . Therefore these maps have at most $`N_1N_2\mathrm{}N_n+1`$ attracting periodic orbits. As we will show below in this section, these maps have only a single period one stable fixed points. Since, denoting m-composition of these functions by $`\mathrm{\Phi }^{(m)}`$, it is straightforward to show that the derivative of $`\mathrm{\Phi }^{(m)}`$ at its possible $`m\times n`$ periodic points of an m-cycle: $`x_{\mu ,k+1}=\mathrm{\Phi }_{N_k}(x_{\mu ,k},\alpha _k),x_{1,\mu +1}=\mathrm{\Phi }_{N_n}(x_{n,\mu },\alpha _N),\mu =1,2\mathrm{},m\text{and}k=1,2,\mathrm{},n`$ and $`x_{1,1}=\mathrm{\Phi }_{N_n}(x_{m,n},\alpha _n)`$ is
$$\frac{d}{dx}\mathrm{\Phi }^{(m)}==\underset{\mu =1}{\overset{m}{}}(\underset{k=1}{\overset{n}{}}\frac{N_k}{\alpha _k}(\alpha _k^2+(1\alpha _k^2)x_{\mu ,k}),$$
(2-3)
since for $`x_{\mu ,k}[0,1]`$ we have:
$$min(\alpha _k^2+(1\alpha _k^2x_{\mu ,k}))=min(1,\alpha _k^2),$$
therefore,
$$min\frac{d}{dx}\mathrm{\Phi }^{(m)}=\underset{k=1}{\overset{n}{}}\left(\frac{N_k}{\alpha _k}min(1,\alpha _k^2)\right)^m.$$
Hence the above expression is definitely greater than one for $`_{k=1}^n\frac{1}{N_k}<_{k=1}^n\alpha _k<_{k=1}^nN_k`$, that is, these maps do not have any kind of m-cycle or periodic orbits in the region of the parameters space defined by $`_{k=1}^n\frac{1}{N_k}<_{k=1}^n\alpha _k<_{k=1}^nN_k`$, actually they are ergodic in this region of the parameters space. From (2-3) it follows that $`\frac{d}{dx}\mathrm{\Phi }^{(m)}`$ at $`m\times n`$ periodic points of the m-cycle belonging to interval , varies between $`_{k=1}^n(N_k\alpha _k)^m`$ and $`_{k=1}^n(\frac{N_k}{\alpha _k})^m`$ for $`_{k=1}^n\alpha _k<_{k=1}^n\frac{1}{N_k}`$ and between $`_{k=1}^n(\frac{N_k}{\alpha _k})^m`$ and $`_{k=1}^n(N_k\alpha _k)^m`$ for $`_{k=1}^n\alpha _k>_{k=1}^nN_k`$, respectively.
From the definition of these maps, we see that definitely $`x=1`$ and $`x=0`$ (in special case of odd integer values of $`N1,N_2,\mathrm{},N_n`$ ) belong to one of the m-cycles.
For $`_{k=1}^n\alpha _k<_{k=1}^n\frac{1}{N_k}(_{k=1}^n\alpha _k>_{k=1}^nN_k)`$, the formula $`(23)`$ implies that for those cases in which $`x=1(x=0)`$ belongs to one of m-cycles we will have $`\frac{d}{dx}\mathrm{\Phi }^{(m)}<1`$, hence the curve of $`\mathrm{\Phi }^{(m)}`$ starts at $`x=1(x=0)`$ beneath the bisector and then crosses it at the previous (next) periodic point with slope greater than one (see Fig. 1), since the formula $`(23)`$ implies that the slope of fixed points increases with the decreasing (increasing) of $`x_{\mu ,k}`$, therefore at all periodic points of n-cycles except for $`x=1(x=0)`$ the slope is greater than one that is they are unstable, this is possible only if $`x=1(x=0)`$ is the only period one fixed point of these maps.
Hence all m-cycles except for possible period one fixed points $`x=1`$ and $`x=0`$ are unstable.
Actually, the fixed point $`x=0`$ is the stable fixed point of these maps in the regions of the parameters spaces defined by $`\alpha _k>0,k=1,2,\mathrm{},n`$ and $`_{k=1}^n\alpha _k<_{k=1}^n\frac{1}{N_k}`$ only for odd integer values of $`N_1,N_2,\mathrm{},N_n`$, however, if one of the integers $`N_k,k=1.2,\mathrm{},n`$ happens to be even, then the $`x=0`$ will not be a stable fixed point anymore. But, the fixed point $`x=1`$ is stable fixed point of these maps in the regions of the parameters spaces defined by $`_{k=1}^n\alpha _k>_{k=1}^nN_k`$ and $`\alpha _k<\mathrm{},k=1,2,\mathrm{},n`$ for all integer values of $`N_1,N_2,\mathrm{},N_n`$.
As an example we give below some of these maps:
$$\varphi _{2,2}^{\alpha _1,\alpha _2}(x)=\frac{\alpha _{1}^{}{}_{}{}^{2}\left(4x\left(x1\right)+\left(2x1\right)^2\alpha _{2}^{}{}_{}{}^{2}\right)^2}{\alpha _{1}^{}{}_{}{}^{2}\left(4x\left(x1\right)+\left(2x1\right)^2\alpha _{2}^{}{}_{}{}^{2}\right)^2+h1}$$
(2-4)
$$\varphi _{2,3}^{\alpha _1,\alpha _2}(x)=\frac{\alpha _{1}^{}{}_{}{}^{2}\left(\left(x1\right)\left(4x1\right)^2+x\left(4x3\right)^2\alpha _{2}^{}{}_{}{}^{2}\right)^2}{\alpha _{1}^{}{}_{}{}^{2}\left(\left(x1\right)\left(4x1\right)^2+x\left(4x3\right)^2\alpha _{2}^{}{}_{}{}^{2}\right)^2+h2}$$
(2-5)
$$\varphi _{3,2}^{\alpha _1,\alpha _2}(x)=\frac{\alpha _{2}^{}{}_{}{}^{2}\left((x1)\left(4x1\right)^2+x\left(4x3\right)^2\alpha _{1}^{}{}_{}{}^{2}\right)^2}{\alpha _{2}^{}{}_{}{}^{2}\left(\left(x1\right)\left(4x1\right)^2+x\left(4x3\right)^2\alpha _{1}^{}{}_{}{}^{2}\right)^2+h3}$$
(2-6)
$$\varphi _{3,3}^{\alpha _1,\alpha _2}(x)=\frac{\alpha _{1}^{}{}_{}{}^{2}\alpha _{2}^{}{}_{}{}^{2}x\left(4x3\right)^2\left(3\left(x1\right)\left(4x1\right)^2+x\left(4x3\right)^2\alpha _{2}^{}{}_{}{}^{2}\right)}{\left(x1\right)^3\left(4x1\right)^6+3x\left(3\alpha _{1}^{}{}_{}{}^{2}2\right)\left(4x3\right)^2\left(x1\right)^2\left(4x1\right)^4\alpha _{2}^{}{}_{}{}^{2}+h4}$$
(2-7)
where :
$$h1=16\alpha _{2}^{}{}_{}{}^{2}(2x1)^2x(x1)$$
$$h2=4x(x1)(4x1)^2(4x3)^2\alpha _{2}^{}{}_{}{}^{2}$$
$$h3=4x(x1)(4x1)^2(4x3)^2\alpha _{1}^{}{}_{}{}^{2}$$
$$h4=3x^2\left(3+2\alpha _{1}^{}{}_{}{}^{2}\right)\left(x1\right)\left(4x1\right)^2\left(4x3\right)^4\alpha _{2}^{}{}_{}{}^{4}+\alpha _{1}^{}{}_{}{}^{2}x^3\left(4x3\right)^6\alpha _{2}^{}{}_{}{}^{6}$$
Below we also introduce their conjugate or isomorphic maps which will be very useful in derivation of their invariant measure and calculation of their KS-entropy in the next section. Conjugacy means that the invertible map $`h(x)=\frac{1x}{x}`$ maps $`I=[0,1]`$ into $`[0,\mathrm{})`$ and transforms maps $`\mathrm{\Phi }_{N_k}(x,\alpha _k)`$ into $`\stackrel{~}{\mathrm{\Phi }}_{N_k}(x,\alpha _k)`$ defined as:
$$\stackrel{~}{\mathrm{\Phi }}_{N_k}(x,\alpha _k)=h\mathrm{\Phi }_{N_k}(x,\alpha _k)h^{(1)}=\frac{1}{\alpha _k^2}\mathrm{tan}^2(N_k\mathrm{arctan}\sqrt{x})$$
Hence this transforms the maps $`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x)`$ into $`\stackrel{~}{\mathrm{\Phi }}_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x)`$ defined as:
$$\stackrel{~}{\mathrm{\Phi }}_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x)$$
$$\frac{1}{\alpha _1^2}\mathrm{tan}^2(N_1\mathrm{arctan}\sqrt{}\frac{1}{\alpha _2^2}\mathrm{tan}^2(N_2\mathrm{arctan}\sqrt{}\mathrm{}\frac{1}{\alpha _n^2}\mathrm{tan}^2(N_n\mathrm{arctan}\sqrt{x})))$$
$$\frac{1}{\alpha _1^2}\mathrm{tan}^2(N_1\mathrm{arctan}\sqrt{\frac{1}{\alpha _2^2}\mathrm{tan}^2(N_2\mathrm{arctan}\sqrt{\mathrm{}\frac{1}{\alpha _n^2}\mathrm{tan}^2(N_n\mathrm{arctan}\sqrt{x})}\mathrm{})})$$
(2-8)
## 3 INVARIANT MEASURE
Dynamical systems, even apparently simple dynamical systems as those described by maps of an interval, can display a rich variety of different asymptotic behavior. On measure theoretical level these types of behavior are described by SRB or invariant measure describing statistically stationary states of the system. The probability measure $`\mu `$ on $`[0,1]`$ is called an SRB or invariant measure of the maps $`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x)`$ given in $`(22)`$, if it is $`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x)`$-invariant and absolutely continuous with respect to Lebesgue measure. For deterministic system such as these composed maps, the $`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x)`$-invariance means that its invariant measure $`\mu _{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}(x)`$ fulfills the following formal Ferbenius-Perron integral equation
$$\mu _{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}(y)=_0^1\delta (y\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x))\mu _{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}(x)𝑑x.$$
This is equivalent to:
$$\mu _{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}(y)=\underset{x\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(y)}{}\mu _{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}(x)\frac{dx}{dy},$$
(3-1)
defining the action of standard Ferobenius-Perron operator for the map $`\mathrm{\Phi }(x)`$ over a function as:
$$P_{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}f(y)=\underset{x\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(y)}{}f(x)\frac{dx}{dy}.$$
(3-2)
We see that, the invariant measure $`\mu _{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}(x)`$ is given as the eigenstate of the Frobenius-Perron operator $`P_{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}`$ corresponding to the largest eigenvalue 1.
As we will prove below the measure $`\mu _{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}(x,\beta )`$ defined as:
$$\frac{1}{\pi }\frac{\sqrt{\beta }}{\sqrt{x(1x)}(\beta +(1\beta )x)},$$
(3-3)
is the invariant measure of the maps $`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x)`$ provided that the parameter $`\beta `$ is positive and fulfills the following relation:
$$\underset{k=1}{\overset{n}{}}\alpha _k\times \frac{A_{N_n}(\frac{1}{\beta })}{B_{N_n}(\frac{1}{\beta })}\times \frac{A_{N_{n1}}(\frac{1}{\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })}}{B_{N_{n1}}(\frac{1}{\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })})}\times \frac{A_{N_{n2}}(\frac{1}{\eta _{N_{n1},N_n}^{\alpha _{n1},\alpha _n}(\frac{1}{\beta })})}{B_{N_{n2}}(\frac{1}{\eta _{N_{n1},N_n}^{\alpha _{n1},\alpha _n}(\frac{1}{\beta })})}\times \times \frac{A_{N_1}(\frac{1}{\eta _{N_2,N_3,\mathrm{},N_n}^{\alpha _2,\alpha _3,\mathrm{},a_n}(\frac{1}{\beta })})}{B_{N_1}(\frac{1}{\eta _{N_2,N_3,\mathrm{},N_n}^{\alpha _2,\alpha _3,\mathrm{},\alpha _n}(\frac{1}{\beta })})}=1$$
(3-4)
where the polynomials $`A_{N_k}(x)`$ and $`B_{N_k}(x)`$ $`(k=1,2,\mathrm{},n)`$ are defined as:
$$A_{N_k}(x)=\underset{l=0}{\overset{[\frac{N_k}{2}]}{}}C_{2l}^{N_k}x^l,$$
$$B_{N_k}(x)=\underset{l=0}{\overset{[\frac{N_k1}{2}]}{}}C_{2l+1}^{N_k}x^l,$$
where $`[]`$ means greatest integer part. Also the functions $`\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })`$, $`\eta _{N_{n1},N_n}^{\alpha _{n1},\alpha _n}(\frac{1}{\beta }),\mathrm{}`$ and $`\eta _{N_2,N_3,\mathrm{},N_n}^{\alpha _2,\alpha _3,\mathrm{},\alpha _n}(\frac{1}{\beta })`$ are defined in the following form:
$$\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })=\beta (\frac{\alpha _nA_{N_n}(\frac{1}{\beta })}{B_{N_n}(\frac{1}{\beta })})^2$$
$$\eta _{N_{n1},N_n}^{\alpha _{n1},\alpha _n}(\frac{1}{\beta })=\beta (\frac{\alpha _{n1}A_{N_{n1}}(\frac{1}{\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })})}{B_{N_{n1}}(\frac{1}{\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })})})^2$$
$$\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}$$
$$\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}$$
$$\eta _{N_2,N_3,\mathrm{},N_n}^{\alpha _2,\alpha _3,\mathrm{},\alpha _n}(\frac{1}{\beta })=\beta (\frac{\alpha _2A_{N_2}(\frac{1}{\eta _{N_3,N_4,\mathrm{},N_n}^{\alpha _3,\alpha _4,\mathrm{},\alpha _n}(\frac{1}{\beta })})}{B_{N_2}(\frac{1}{\eta _{N_3,N_4,\mathrm{},N_n}^{\alpha _3,\alpha _4,\mathrm{},\alpha _n}(\frac{1}{\beta })})})^2$$
As we see the above measure is defined only for $`\beta >0`$ hence, from the relations $`(34)`$, it follows that these maps are ergodic in the region of the parameter space which leads to positive solution of $`\beta `$. Taking the limits of $`\beta 0_+`$ and $`\beta \mathrm{}`$ in the relation (3-4),respectively, one can show that the ergodic regions are : $`_{k=1}^n\frac{1}{N_k}<_{k=1}^n\alpha _k<_{k=1}^nN_k`$ for odd integer values of $`N_1,N_2,\mathrm{},N_n`$ and $`\alpha _k>0,\text{for}k=1,2,\mathrm{},n\&_{k=1}^n\alpha _k<_{k=1}^nN_k`$ if one of the integers happens to become even, respectively. Out of these regions they have only single stable fixed points.
In order to prove that measure $`(33)`$ satisfies equation $`(31)`$, with $`\beta `$ given by relation $`(34)`$, it is rather convenient to consider the conjugate map:
$$\stackrel{~}{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}}\alpha _1,\alpha _2,\mathrm{},\alpha _n(x),$$
with measure $`\stackrel{~}{\mu }_{\stackrel{~}{\mathrm{\Phi }}_{N_1,N_2\mathrm{},N_n}^{\alpha _1,\alpha _2\mathrm{},\alpha _n}}`$ denoted by $`\stackrel{~}{\mu }_{\stackrel{~}{\varphi }}`$ related to the measure $`\mu _{\mathrm{\Phi }_{N_1,N_2\mathrm{},N_n}^{\alpha _1,\alpha _2\mathrm{},\alpha _n}}`$ denoted by $`\mu _\varphi `$ through the following relation:
$$\stackrel{~}{\mu _{\stackrel{~}{\mathrm{\Phi }}}}(x)=\frac{1}{(1+x)^2}\mu _\mathrm{\Phi }(\frac{1}{1+x}).$$
Denoting $`\stackrel{~}{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}}\alpha _1,\alpha _2,\mathrm{},\alpha _n(x)`$ by $`y`$ and inverting it, we get :
$$x_{k_1}=\mathrm{tan}^2(\frac{1}{N_1}\mathrm{arctan}\sqrt{y\alpha _1^2}+\frac{k_1\pi }{N_1})k_1=1,..,N_1.$$
$$x_{k_1,k_2}=\mathrm{tan}^2(\frac{1}{N_2}\mathrm{arctan}\sqrt{x_{k_1}\alpha _2^2}+\frac{k_2\pi }{N_2})k_2=1,..,N_2.$$
$$\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}$$
$$\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}$$
$$x_{k_1,k_2,\mathrm{},k_n}=\mathrm{tan}^2(\frac{1}{N_n}\mathrm{arctan}\sqrt{x_{k_1,k_2,\mathrm{},k_{n1}}\alpha _n^2}+\frac{k_n\pi }{N_n})k_1=1,..,N_1.$$
Then, taking derivative of $`x_{k_1,k_2,\mathrm{},k_n}`$ with respect to $`y`$, we obtain:
$$\frac{dx_{k1,k_2,\mathrm{},k_n}}{dy}=(\underset{k=1}{\overset{n}{}}\frac{\alpha _k}{N_k})\sqrt{\frac{x_{k_1,k_2,\mathrm{},k_n}}{y}}$$
$$\frac{(1+x_{k_1,k_2,\mathrm{},k_n})(1+x_{k_2,k_3,\mathrm{},k_n})\mathrm{}(1+x_{k_{n1},k_n})(1+x_{k_n})}{(1+\alpha _n^2x_{k_2,k_3,\mathrm{},k_n})(1+\alpha _{n1}^2x_{k_3,k_4,\mathrm{},k_n})\mathrm{}(1+\alpha _3^2x_{k_{n1},k_n})(1+\alpha _2^2x_{k_n})(1+\alpha _1^2y)}.$$
(3-5)
In derivation of the formula we have used the chain rule properties of the derivative of composite functions.
Substituting the above result in equation $`(31)`$, we have:
$$\stackrel{~}{\mu }_{\stackrel{~}{\mathrm{\Phi }}}(y)\sqrt{y}(1+\alpha _1^2y)=(\underset{k=1}{\overset{n}{}}\frac{\alpha _k}{N_k})\underset{k_1}{}\underset{k_2}{}\mathrm{}\underset{k_n}{}\sqrt{x_{k_1,k_2,\mathrm{},k_n}}$$
$$\times \frac{(1+x_{k_1,k_2,\mathrm{},k_n})(1+x_{k_2,k_3,\mathrm{},k_n})\mathrm{}(1+x_{k_{n1},k_n})(1+x_{k_n})}{(1+\alpha _n^2x_{k_2,k_3,\mathrm{},k_n})(1+\alpha _{n1}^2x_{k_3,k_4,\mathrm{},k_n})\mathrm{}(1+\alpha _3^2x_{k_{n1},k_n})(1+\alpha _2^2x_{k_n})}\stackrel{~}{\mu }_{\stackrel{~}{\mathrm{\Phi }}}(x_{k_1,k_2,\mathrm{},k_n}).$$
Now,considering the following anzatz for the invariant measure $`\stackrel{~}{\mu }_{\stackrel{~}{\mathrm{\Phi }}}(y)`$:
$$\stackrel{~}{\mu }_{\stackrel{~}{\mathrm{\Phi }}}(y)=\frac{1}{\sqrt{y}(1+\beta y)},$$
(3-6)
then the above equation reduces to:
$$\frac{1+\alpha _1^2y}{1+\beta y}=(\underset{k=1}{\overset{n}{}}\frac{\alpha _k}{N_k})$$
$$\times \underset{k_1=1}{\overset{N_1}{}}\underset{k_2=1}{\overset{n_2}{}}\mathrm{}\underset{k_n=1}{\overset{N_n}{}}\left(\frac{(1+x_{k_1,k_2,\mathrm{},k_n})(1+x_{k_2,k_3,\mathrm{},k_n})\mathrm{}(1+x_{k_{n1},k_n})(1+x_{k_n})}{(1+\alpha _n^2x_{k_2,k_3,\mathrm{},k_n})(1+\alpha _{n1}^2x_{k_3,k_4,\mathrm{},k_n})\mathrm{}(1+\alpha _3^2x_{k_{n1},k_n})(1+\alpha _2^2x_{k_n})}\right).$$
Now, using the formula (3-5) of reference we obtain:
$$\frac{\alpha _n}{N_n}\underset{k_n=1}{\overset{N_n}{}}\frac{1+x_{k_1,k_2,\mathrm{},k_n}}{1+\beta x_{k_1,k_2,\mathrm{},k_n}}=\frac{\alpha _nA_{N_n}(\frac{1}{\beta })}{B_{N_n}(\frac{1}{\beta })}\frac{1+\alpha _n^2x_{k_1,k_2,\mathrm{},k_{n1}}}{1+\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })x_{k_1,k_2,\mathrm{},k_{n1}}},$$
$$\frac{\alpha _{n1}\alpha _n}{N_{n1}N_n}\underset{k_{n1}=1}{\overset{N_{n1}}{}}\underset{k_n=1}{\overset{N_n}{}}\frac{(1+x_{k_1,k_2,\mathrm{},k_{n1}})(1+x_{k_1,k_2,\mathrm{},k_n})}{(1+\alpha _{n1}x_{k_1,k_2,\mathrm{},k_{n1}})(1+\beta x_{k_1,k_2,\mathrm{},k_n})}=$$
$$\frac{\alpha _n\alpha _{n1}A_{N_n}(\frac{1}{\beta })A_{N_{n1}}(\frac{1}{\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })})}{B_{N_n}(\frac{1}{\beta })B_{N_{n1}}(\frac{1}{\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })})}\times \frac{1+\alpha _{n1}^2x_{k_1,k_2,\mathrm{},k_{n2}}}{1+\eta _{N_{n1},N_n}^{\alpha _{n1},\alpha _n}(\frac{1}{\beta })x_{k_1,k_2,\mathrm{},k_{n2}}},$$
$$\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}$$
$$\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}$$
$$(\underset{k=1}{\overset{n}{}}\frac{\alpha _k}{N_k})\underset{k_1=1}{\overset{N_1}{}}\underset{k_2=1}{\overset{n_2}{}}\mathrm{}$$
$$\times \underset{k_n=1}{\overset{N_n}{}}\left(\frac{(1+x_{k_1,k_2,\mathrm{},k_n})(1+x_{k_2,k_3,\mathrm{},k_n})\mathrm{}(1+x_{k_{n1},k_n})(1+x_{k_n})}{(1+\alpha _n^2x_{k_2,k_3,\mathrm{},k_n})(1+\alpha _{n1}^2x_{k_3,k_4,\mathrm{},k_n})\mathrm{}(1+\alpha _3^2x_{k_{n1},k_n})(1+\alpha _2^2x_k{}_{n}{}^{})}\right)=$$
$$\underset{k=1}{\overset{n}{}}\alpha _k\times \frac{A_{N_n}(\frac{1}{\beta })}{B_{N_n}(\frac{1}{\beta })}\times \frac{A_{N_{n1}}(\frac{1}{\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })})}{B_{N_{n1}}(\frac{1}{\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })})}$$
$$\times \frac{A_{N_{n2}}(\frac{1}{\eta _{N_{n1},N_n}^{\alpha _{n1},\alpha _n}(\frac{1}{\beta })})}{B_{N_{n2}}(\frac{1}{\eta _{N_{n1},N_n}^{\alpha _{n1},\alpha _n}(\frac{1}{\beta })})}\times \times \frac{A_{N_1}(\frac{1}{\eta _{N_2,N_3,\mathrm{},N_n}^{\alpha _2,\alpha _3,\mathrm{},a_n}(\frac{1}{\beta })})}{B_{N_1}(\frac{1}{\eta _{N_2,N_3,\mathrm{},N_n}^{\alpha _2,\alpha _3,\mathrm{},\alpha _n}(\frac{1}{\beta })})}\frac{1+\alpha _1^2y}{1+\eta _{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(\frac{1}{\beta })y}.$$
Now, inserting the right hand side of last relation in $`(35)`$, we get:
$$\frac{1+\alpha _1^2y}{1+\beta y}=\underset{k=1}{\overset{n}{}}\alpha _k\frac{A_{N_n}(\frac{1}{\beta })}{B_{N_n}(\frac{1}{\beta })}\times \frac{A_{N_{n1}}(\frac{1}{\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })})}{B_{N_{n1}}(\frac{1}{\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })})}\times $$
$$\frac{A_{N_{n2}}(\frac{1}{\eta _{N_{n1},N_n}^{\alpha _{n1},\alpha _n}(\frac{1}{\beta })})}{B_{N_{n2}}(\frac{1}{\eta _{N_{n1},N_n}^{\alpha _{n1},\alpha _n}(\frac{1}{\beta })})}\times \times \frac{A_{N_1}(\frac{1}{\eta _{N_2,N_3,\mathrm{},N_n}^{\alpha _2,\alpha _3,\mathrm{},a_n}(\frac{1}{\beta })})}{B_{N_1}(\frac{1}{\eta _{N_2,N_3,\mathrm{},N_n}^{\alpha _2,\alpha _3,\mathrm{},\alpha _n}(\frac{1}{\beta })})}\frac{1+\alpha _1^2y}{1+\eta _{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(\frac{1}{\beta })y}.$$
We see that the above relation will hold true provided that the parameter $`\beta `$ fulfills the relation (3-4).
## 4 KOLMOGROV-SINAI ENTROPY
Kolomogrov-Sinai entropy (KS) or metric entropy measure how chaotic a dynamical system is and it is proportional to the rate at which information about the state of dynamical system is lost in the course of time or iteration. Therefore, it can also be defined as the average rate of information loss for a discrete measurable dynamical system $`(\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x),\mu )`$, by introducing a partition $`\alpha =A_c(n_1,\mathrm{}..n_\gamma )`$ of the interval $`[0,1]`$ into individual laps $`A_i`$ one can define the usual entropy associated with the partition by:
$$H(\mu ,\gamma )=\underset{i=1}{\overset{n(\gamma )}{}}m(A_c)\mathrm{ln}m(A_c),$$
where $`m(A_c)={}_{nA_i}{}^{}\mu (x)𝑑x`$ is the invariant measure of $`A_i`$. Defining n-th refining $`\gamma (n)`$ of $`\gamma `$:
$$\gamma ^n=\underset{k=0}{\overset{n1}{}}(\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x))^{(k)}(\gamma ),$$
and defining an entropy per unit step of refining by :
$$h(\mu ,\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x),\gamma )=lim{}_{n\mathrm{}}{}^{}(\frac{1}{n}H(\mu ,\gamma )),$$
if the size of individual laps of $`\gamma (N)`$ tends to zero as n increases, then the above entropy is known as Kolmogorov-Sinai entropy, that is:
$$h(\mu ,\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x))=h(\mu ,\mathrm{\Phi }N_1,N_2,\mathrm{},N_{n}^{}{}_{(}{}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}x),\gamma ).$$
KS-entropy , which is a quantitative measure of the rate of information loss with the refining, may also be written as:
$$h(\mu ,\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x))=\mu (x)𝑑x\mathrm{ln}\frac{d}{dx}\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x),$$
(4-1)
which is also a statistical mechanical expression for the Lyapunov characteristic exponent, that is, the mean divergence rate of two nearby orbits. The measurable dynamical system $`(\mathrm{\Phi }N_1,N_2,\mathrm{},N_{n}^{}{}_{(}{}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}x),\mu )`$ is chaotic for $`h>0`$ and predictive for $`h=0`$.
In order to calculate the KS-entropy of the maps $`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x)`$, it is rather convenient to consider their conjugate maps given by $`(28)`$, since it can be shown that KS-entropy is a kind of topological invariant, that is, it is preserved under conjugacy map. Hence we have:
$$h(\mu ,\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x))=h(\stackrel{~}{\mu },\stackrel{~}{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}(x)).$$
Using the integral $`(41)`$, the KS-entropy of $`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x)`$ can be written as
$$h(\mu ,\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x))=h(\stackrel{~}{\mu },\stackrel{~}{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}(x))=$$
$$\frac{1}{\pi }_0^{\mathrm{}}\frac{\sqrt{\beta }dx}{\sqrt{x}(1+\beta x)}\mathrm{ln}(\frac{d}{dy_{N_2,N_3,,N_n}}(\frac{1}{\alpha _1^2}\mathrm{tan}^2(N_1\mathrm{arctan}\sqrt{y_{N_2,N_3,,N_n}}))\times $$
$$\frac{d}{dy_{N_3,N_4,,N_n}}(\frac{1}{\alpha _2^2}\mathrm{tan}^2(N_2\mathrm{arctan}\sqrt{y_{N_3,N_4,,N_n}}))\mathrm{}\frac{d}{dx}(\frac{1}{\alpha _n^2}\mathrm{tan}^2(N_n\mathrm{arctan}\sqrt{x})))$$
or
$$h(\mu ,\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x))=$$
$$\frac{1}{\pi }_0^{\mathrm{}}\frac{\sqrt{\beta }dx}{\sqrt{x}(1+\beta x)}\mathrm{ln}(\frac{d}{dy_{N_2,N_3,,N_n}}(\frac{1}{\alpha _1^2}\mathrm{tan}^2(N_1\mathrm{arctan}\sqrt{y_{N_2,N_3,,N_n}})))+$$
$$\frac{1}{\pi }_0^{\mathrm{}}\frac{\sqrt{\beta }dx}{\sqrt{x}(1+\beta x)}\mathrm{ln}(\frac{d}{dy_{N_3,N_4,,N_n}}(\frac{1}{\alpha _2^2}\mathrm{tan}^2(N_2\mathrm{arctan}\sqrt{y_{N_3,N_4,,N_n}})))+\mathrm{}+$$
$$\frac{1}{\pi }_0^{\mathrm{}}\frac{\sqrt{\beta }dx}{\sqrt{x}(1+\beta x)}\mathrm{ln}(\frac{d}{dy_{N_n}}(\frac{1}{\alpha _{n1}^2}\mathrm{tan}^2(N_{n1}\mathrm{arctan}\sqrt{y_{N_n}})))+$$
$$\frac{1}{\pi }_0^{\mathrm{}}\frac{\sqrt{\beta }dx}{\sqrt{x}(1+\beta x)}\mathrm{ln}(\frac{d}{dx}(\frac{1}{\alpha _n^2}\mathrm{tan}^2(N_n\mathrm{arctan}(\sqrt{x})))))$$
(4-2)
where
$$y_{N_n}=\frac{1}{\alpha _n^2}\mathrm{tan}^2(N_n\mathrm{arctan}(\sqrt{x}))$$
(4-3)
$$y_{N_{n1},N_n}=\frac{1}{\alpha _{n1}}\mathrm{tan}^2(N_{n1}\mathrm{arctan}(\sqrt{y_{N_n}})))$$
(4-4)
$$\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}$$
$$\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}$$
$$y_{N_2,N_3,,N_n}=\frac{1}{\alpha _1^2}\mathrm{tan}^2(N_1\mathrm{arctan}(\sqrt{y_{N_3,N_4,,N_n}})).$$
(4-5)
. Now, we calculate the integrals appearing above in the expression for the entropy, separately. The last integral in (4-2) can be calculated with the prescription of section IV of the reference and it reads:
$$\frac{1}{\pi }_0^{\mathrm{}}\frac{\sqrt{\beta }dx}{\sqrt{x}(1+\beta x)}\mathrm{ln}(\frac{d}{dx}(\frac{1}{\alpha _n^2}\mathrm{tan}^2(N_n\mathrm{arctan}(\sqrt{x})))))=\mathrm{ln}(\frac{N_n(1+\beta +2\sqrt{\beta })^{N_n1}}{A_{N_n}(\frac{1}{\beta })B_{N_n}(\frac{1}{\beta })}).$$
(4-6)
In order to calculate the integral one before last in (4-2), that is:
$$\frac{1}{\pi }_0^{\mathrm{}}\frac{\sqrt{\beta }dx}{\sqrt{x}(1+\beta x)}\mathrm{ln}(\frac{d}{dy_{N_n}}(\frac{1}{\alpha _{n1}^2}\mathrm{tan}^2(N_{n1}\mathrm{arctan}\sqrt{y_{N_n}})))$$
(4-7)
first we make the following change of variable by inverting the relation(4-3):
$$x_{k_{n1}}=\mathrm{tan}^2(\frac{1}{N_{n1}}\mathrm{arctan}(\sqrt{y_{N_n}}\alpha _{n1}^2)+\frac{k_{n1}\pi }{N_{n1}})k_{n1}=1,..,N_{n1}.$$
then the integral (4-7) reduces to:
$$\underset{k_{n1}=1}{\overset{N_{n1}}{}}\frac{1}{\pi }_{x_{k_{n1}}^i}^{x_{k_{n1}}^f}\frac{\sqrt{\beta }dx_{k_{n1}}}{\sqrt{x_{k_{n1}}}(1+\beta x_{k_{n1}})}\mathrm{ln}(\frac{d}{dy_{N_n}}(\frac{1}{\alpha _{n1}^2}\mathrm{tan}^2(N_{n1}\mathrm{arctan}\sqrt{y_{N_n}}))$$
where $`x_{k_{n1}}^i`$ and $`x_{k_{n1}}^f`$ ($`k_{n1}=1,2,\mathrm{},N_{n1}`$) denote the initial and end points of $`k`$-th branch of the inversion of function $`y_{N_n}=(\frac{1}{\alpha _n^2}\mathrm{tan}^2(N_n\mathrm{arctan}\sqrt{x}))`$, respectively. Now, inserting the derivative of $`x_{k_{n1}}`$ with respect to $`y_{N_n}`$ in the above relation and changing the order of sum and integration, we get:
$$\frac{1}{\pi }_0^{\mathrm{}}\underset{k_{n1}=1}{\overset{N_{n1}}{}}\sqrt{\beta }dy_{N_n}\frac{\alpha _{n1}\sqrt{x_{k_{n1}}}(1+x_{k_{n1}})}{N_{n1}\sqrt{y_{n_n}}(1+\alpha _{n1}^2y_{N_n})\sqrt{x_{k_{n1}}}(1+\beta x_{k_{n1}})}$$
$$\times \mathrm{ln}(\frac{d}{dy_{N_n}}(\frac{1}{\alpha _{n1}^2}\mathrm{tan}^2(N_{n1}\mathrm{arctan}\sqrt{y_{N_n}}))).$$
Using the formula (3-5) of reference , it reduces to
$$\frac{1}{\pi }_0^{\mathrm{}}\frac{\sqrt{\beta }dy_n}{\sqrt{y_n}}(\frac{B_{N_n}(\frac{1}{\beta })}{\alpha _nA_{N_n}(\frac{1}{\beta })}+\beta \frac{\alpha _nA_{N_n}(\frac{1}{\beta })}{B_{N_n}(\frac{1}{\beta })}y_{N_n})\mathrm{ln}(\frac{d}{dy_{N_n}}(\frac{1}{\alpha _{n1}^2}\mathrm{tan}^2(N_{n1}\mathrm{arctan}\sqrt{y_{N_n}}))).$$
Finally, calculating the above integral with the prescription of reference we obtain:
$$\mathrm{ln}\left(\frac{N_{n1}(1+\eta _{N_n}^{\alpha _n}+2\sqrt{\eta _{N_n}^{\alpha _n}})^{N_{n1}1}}{A_{N_{n1}}(\eta _{N_n}^{\alpha _n})B_{N_{n1}}(\eta _{N_n}^{\alpha _n})}\right).$$
Similarly, we can calculate the other integrals appearing in the expression for the entropy of the composed maps given in (4-2):
$$=\frac{1}{\pi }_0^{\mathrm{}}\frac{\sqrt{\beta }dx}{\sqrt{x}(1+\beta x)}\mathrm{ln}(\frac{d}{dy_{N_k,N_{k+1},,N_n}}(\frac{1}{\alpha _{k1}^2}\mathrm{tan}^2(N_{k1}\mathrm{arctan}\sqrt{y_{N_k,N_{k+1},,N_n}}))=$$
$$\mathrm{ln}\left(\frac{N_{k1}(1+\eta _{N_k,N_{k1},\mathrm{},N_n}^{\alpha _k,\alpha _{k+1},\mathrm{},\alpha _n}(\frac{1}{\beta })+2\sqrt{\eta _{N_k,N_{k1},\mathrm{},N_n}^{\alpha _k,\alpha _{k+1},\mathrm{},\alpha _n}(\frac{1}{\beta })})^{N_{k1}1}}{A_{N_{k1}}(\eta _{N_k,N_{k1},\mathrm{},N_n}^{\alpha _k,\alpha _{k+1},\mathrm{},\alpha _n}(\frac{1}{\beta }))B_{N_{k1}}(\eta _{N_k,N_{k1},\mathrm{},N_n}^{\alpha _k,\alpha _{k+1},\mathrm{},\alpha _n})(\frac{1}{\beta })}\right).\text{for}k=1,2,\mathrm{},n$$
Finally summing the above integral we get the following expression for the entropy of these maps:
$$h(\mu ,\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x))=$$
$$\mathrm{ln}(\frac{(N_1N_2\mathrm{}N_n)(1+\sqrt{\beta })^{2(N_n1)}(1+\sqrt{\eta _{N_n}^{\alpha _n}(\frac{1}{\beta })})^{2(N_{n1}1)}\mathrm{}(1+\sqrt{\eta _{N_2,N_3,\mathrm{},N_n}^{\alpha _2,\alpha _3,\mathrm{},\alpha _n}(\frac{1}{\beta })})^{2(N_11)}}{A_{N_n}(\beta )B_{N_n}(\beta )A_{N_{n1}}(\eta _{N_n}^{\alpha _n}(\frac{1}{\beta }))B_{N_{n1}}(\eta _{N_n}^{\alpha _n}(\frac{1}{\beta }))\mathrm{}A_{N_1}(\eta _{N_2,N_3,\mathrm{},N_n}^{\alpha _2,\alpha _3,\mathrm{},\alpha _n}(\frac{1}{\beta }))B_{N_1}(\eta _{N_2,N_3,\mathrm{},N_n}^{\alpha _2,\alpha _3,\mathrm{},\alpha _n}(\frac{1}{\beta }))}).$$
(4-8)
Imposing the relations between the parameters $`\alpha _k,k=1,2,\mathrm{},n`$ which are consistent with the relation (3-4), reduces these maps to other maps of many-parameters family of the maps with number of the parameters less than $`n`$. Particularly by imposing enough relations we can reduce them to one-parameter family of chaotic maps with an arbitrary asymptotic behavior as the parameter takes the limiting values. Hence we can construct chaotic maps with arbitrary universality class. As an illustration we consider the chaotic map $`\mathrm{\Phi }_{2,2}^{\alpha _1,\alpha _2}(x)`$. Using the formula (4-8) we have
$$h(\mu ,\mathrm{\Phi }_{2,2}^{\alpha _1,\alpha _2}(x)=\mathrm{ln}\frac{(1+\sqrt{\beta })^2(2\sqrt{\beta }+\alpha _2(1+\beta ))^2}{(1+\beta )(4\beta +\alpha _2^2(1+\beta )^2)}$$
with the following relation among the parameters $`\alpha _1,\alpha _2`$ and $`\beta `$:
$$\alpha _1(4\beta +\alpha _2^2(1+\beta )^2)=4\alpha _2\beta (1+\beta )$$
which is obtained from the relation(3-4). Now choosing $`\beta =\alpha _2^\nu ,,0<\nu <2`$, the above relation reduces to:
$$a_1=\frac{4\alpha _2^{1+\nu }(1+\alpha _2^\nu )}{\alpha _2^2(1+\alpha _2^\nu )^2+4\alpha _2^\nu }$$
and entropy given by (4-8) reads:
$$h(\mu ,\mathrm{\Phi }_{2,2}^{\alpha _2}(x)=\mathrm{ln}\frac{(1+\alpha _2^{\frac{\nu }{2}})^2(2\alpha _2^{\frac{\nu }{2}}+\alpha _2(1+\alpha _2^\nu ))^2}{(1+\alpha _2^\nu )(4\alpha _2^\nu +\alpha _2^2(1+\alpha _2^\nu )^2)}$$
which has the following asymptotic behavior near $`\alpha _20`$ and $`\alpha _2\mathrm{}`$:
$$\{\begin{array}{c}h(\mu ,\mathrm{\Phi }_{2,2}^{\alpha _2}(x)\alpha _2^{\frac{\nu }{2}}\text{as}\alpha _20\hfill \\ h(\mu ,\mathrm{\Phi }_{2,2}^{\alpha _2}(x)(\frac{1}{\alpha _2})^{\frac{\nu }{2}}\text{as}\alpha _2\mathrm{}.\hfill \end{array}$$
The above asymptotic form indicates that, for an arbitrary value of $`0<\nu <2`$, the maps $`\mathrm{\Phi }_{2,2}^{\alpha _2}(x)`$ belong to the universality class which is different from the universality class of chaotic maps of the reference or the universality class of pitch fork bifurcating maps.
Here in this section we try to calculate Lyapunov characteristic exponent of maps $`\mathrm{\Phi }_N^{(1,2)}(x,\alpha )`$, $`N=1,2,\mathrm{}.,5`$ in order to investigate these maps numerically. In fact, Lyapunov characteristic exponent is the characteristic exponent of the rate of average magnificent of the neighborhood of an arbitrary point $`x_0`$ and it is denoted by $`\mathrm{\Lambda }(x_0)`$ which is written as:
$$\mathrm{\Lambda }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x_0)=lim_n\mathrm{}\mathrm{ln}(\stackrel{n}{\stackrel{}{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x,\alpha )\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}\mathrm{}.\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}}$$
$$=lim_n\mathrm{}\underset{k=0}{\overset{n1}{}}\mathrm{ln}\frac{d\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x_k,\alpha )}{dx},$$
(4-1)
where $`x_k=\stackrel{}{\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}\mathrm{}.\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}}`$ . It is obvious that $`\mathrm{\Lambda }^{(1,2)}(x_0)<0`$ for an attractor, $`\mathrm{\Lambda }N_1,N_2,\mathrm{},N_{n}^{}{}_{}{}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x_0)>0`$ for a repeller and $`\mathrm{\Lambda }N_1,N_2,\mathrm{},N_{n}^{}{}_{}{}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x_0)=0`$ for marginal situation. Also the Liapunov number is independent of initial point $`x_0`$, provided that the motion inside the invariant manifold is ergodic, thus $`\mathrm{\Lambda }N_1,N_2,\mathrm{},N_{n}^{}{}_{}{}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x_0)`$ characterizes the invariant manifold of $`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}`$ as a whole. For values of parameter $`\alpha _k,k=1,2,\mathrm{},n`$, such that the map $`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}`$ be measurable, Birkohf ergodic theorem implies equality of KS-entropy and Liapunov characteristic exponent, that is:
$$h(\mu ,\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n})=\mathrm{\Lambda }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x_0,\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}).$$
(4-2)
Comparison of analytically calculated KS-entropy of maps $`\mathrm{\Phi }_{N_1,N_2,\mathrm{},N_n}^{\alpha _1,\alpha _2,\mathrm{},\alpha _n}(x,\alpha )`$ for $`N_1=2,3`$ and $`N_2=2,3`$, (see Figures $`5,6`$ and $`7`$ ) with the corresponding Lyapunov characteristic exponent obtained by simulation, indicate that in chaotic region, these maps are ergodic as predicted by Birkohf ergodic theorem. In non chaotic region of the parameter, Lyapunov characteristic exponent is negative, since in this region we have only stable period one fixed points without bifurcation. In summary, combining the analytic discussion of section II with the numerical simulation we deduce that these maps are ergodic in certain region of their parameters space as explained above and in the complementary region of the parameters space they have only a single period one attractive fixed point, such that in contrary to the most of usual one-dimensional one-parameter or many-parameters family of maps they have only a bifurcation from a period one attractive fixed point to chaotic state or vice-versa.
## 5 Conclusion
We have given hierarchy of exactly solvable many-parameter family of one-dimensional chaotic maps with an invariant measure, that is measurable dynamical system with an interesting property of being either chaotic (proper to say ergodic ) or having stable period one fixed point and they bifurcate from a stable single periodic state to chaotic one and vice-versa without having usual period doubling or period-n-tupling scenario.
Again this interesting property is due to existence of invariant measure for a region of the parameters space of these maps. Hence, to approve this conjecture, it would be interesting to find other measurable one parameter maps, specially higher dimensional maps, which is under investigation.
|
warning/0005/hep-ph0005018.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
These are exciting times for $`B`$ physics. The CDF collaboration has made a measurement of the CP-violating phase $`\mathrm{sin}2\beta =0.79_{0.44}^{+0.41}`$ , leading to the nontrivial constraint $`\mathrm{sin}2\beta >0`$ at 93% C.L. The asymmetric $`e^+e^{}`$ $`B`$-factories BaBar and Belle are now running and will hopefully make measurements of CP-violating rate asymmetries in the $`B`$ system before too long. And in the near future, data from HERA-B and hadron colliders will add to our knowledge of CP violation in the $`B`$ system.
The purpose of all this activity is to test the standard model (SM) explanation of CP violation. In the SM, CP violation, which to date has been only seen in the kaon system, is due to the presence of a nonzero complex phase in the Cabibbo-Kobayashi-Maskawa (CKM) quark mixing matrix $`V`$. In this scenario, one expects large CP-violating effects in $`B`$ decays, and the above experiments are searching for such signals.
The CP-violating signals which have been the most extensively studied are rate asymmetries in $`B`$ decays . Measurements of such asymmetries will allow one to cleanly probe the interior angles $`\alpha `$, $`\beta `$ and $`\gamma `$ of the so-called unitarity triangle , which will in turn provide important tests of the SM.
However, there is another class of CP-violating signals which has received relatively little attention: triple-product correlations . In a given decay, it may be possible to measure the momenta and/or spins of the particles involved. From these one can construct triple products of the form $`\stackrel{}{v}_1(\stackrel{}{v}_2\times \stackrel{}{v}_3)`$, where each $`v_i`$ is a spin or momentum. Such triple products are odd under time reversal (T) and hence, by the CPT theorem, are also potential signals of CP violation. (Note that there is a technical distinction to be made here: although the action of T changes the sign of a triple product, if a triple product changes sign, it is not necessarily due to the T transformation. This is because, in addition to reversing spins and momenta, the time reversal symmetry T also exchanges the initial and final states. Thus, in a particular decay, a nonzero triple product is not necessarily a signal of T (and CP) violation. For this reason, in what follows we refer to triple-product asymmetries as T-odd effects. We also show how to establish the presence of a true signal of T violation.)
To establish the presence of a nonzero triple-product correlation, one constructs a T-odd asymmetry of the form
$$A_T\frac{\mathrm{\Gamma }(\stackrel{}{v}_1(\stackrel{}{v}_2\times \stackrel{}{v}_3)>0)\mathrm{\Gamma }(\stackrel{}{v}_1(\stackrel{}{v}_2\times \stackrel{}{v}_3)<0)}{\mathrm{\Gamma }(\stackrel{}{v}_1(\stackrel{}{v}_2\times \stackrel{}{v}_3)>0)+\mathrm{\Gamma }(\stackrel{}{v}_1(\stackrel{}{v}_2\times \stackrel{}{v}_3)<0)},$$
(1)
where $`\mathrm{\Gamma }`$ is the decay rate for the process in question. Unfortunately, triple-product correlations suffer from a well-known complication: their signals can be faked by the presence of strong phases, even if there is no CP violation. (As noted above, this is because such correlations are not true T-violating signals.) That is, one typically finds that
$$A_T\mathrm{sin}(\varphi +\delta ),$$
(2)
where $`\varphi `$ is a weak, CP-violating phase and $`\delta `$ is a strong phase. From this we see that if $`\delta 0`$, a triple-product correlation will appear, even in the absence of CP violation (i.e. if $`\varphi =0`$).
To remedy this, one can construct the T-violating asymmetry:
$$𝒜_T\frac{1}{2}(A_T\overline{A}_T),$$
(3)
where $`\overline{A}_T`$ is the T-odd asymmetry measured in the CP-conjugate decay process. This is a true T-violating signal in that it is nonzero only if $`\varphi 0`$ (i.e. if CP violation is present). Furthermore, unlike decay-rate asymmetries in direct CP violation, a nonzero $`𝒜_T`$ does not require the presence of a nonzero strong phase. Indeed:
$$𝒜_T\mathrm{sin}\varphi \mathrm{cos}\delta ,$$
(4)
so that the signal is maximized when the strong phase is zero.
As with all CP-violating signals, (at least) two decay amplitudes are necessary to produce a triple-product correlation. Such correlations have been studied in semileptonic $`B`$ decays . However, since there is only a single amplitude in the SM, any such signal can occur only in the presence of new physics.
To our knowledge, the only study of triple products in the SM has been made by Valencia , who examined the decay $`BV_1V_2`$, where $`V_1`$ and $`V_2`$ are vector mesons. He looked at triple products of the form $`\stackrel{}{k}(\stackrel{}{ϵ}_1\times \stackrel{}{ϵ}_2)`$, where $`\stackrel{}{ϵ}_1`$ and $`\stackrel{}{ϵ}_2`$ are the polarizations of $`V_1`$ and $`V_2`$, respectively, and $`\stackrel{}{k}`$ is the momentum of one of the vector mesons. Since the calculation was done at the meson level, estimates of the various form factors were needed. The conclusion of this study was that, within the SM, one could expect a T-violating asymmetry at the level of several percent.
In this paper we re-examine the question of triple products in the SM using a complementary approach. In particular, we search for triple-product correlations at the quark level. The motivation is the following: if a significant triple-product correlation exists at the hadron level, it must also exist at the quark level. After all, given that QCD (which is responsible for hadronization) is CP-conserving, it is difficult to see how one can generate a large T-violating asymmetry at the hadron level if it is absent at the quark level.
Of course, the converse is not necessarily true: a large T-violating effect at the quark level might be “washed out” during hadronization, since the spins and momenta of the quarks may not correlate well with the spins and momenta of the hadrons. (The most obvious example of this is if spin-0 mesons are involved. In this case no information about the spins of the constituent quarks can be obtained.) Thus, there may be considerable hadronic uncertainty in taking a nonzero quark-level signal and applying it at the hadron level.
With this in mind, in this paper we examine the inclusive decay $`bsu\overline{u}`$ within the SM. If there is a large $`\stackrel{}{k}(\stackrel{}{ϵ}_1\times \stackrel{}{ϵ}_2)`$ triple product in $`BV_1V_2`$, there should also be a large triple product at the quark level of the form $`\stackrel{}{p}(\stackrel{}{s}\times \stackrel{}{s}^{})`$, where $`\stackrel{}{p}`$ is the momentum of one of the quarks, and $`\stackrel{}{s}`$ and $`\stackrel{}{s}^{}`$ are the spins of two of the light quarks. And indeed, we find that the quark-level T-violating asymmetry due to the triple product $`\stackrel{}{p}_u(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})`$ or $`\stackrel{}{p}_{\overline{u}}(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})`$ can be as large as about 5%. This strongly supports Valencia’s conclusion that the SM predicts a measurable T-violating asymmetry in $`BV_1V_2`$.
However, we also find another significant T-violating signal in $`bsu\overline{u}`$. It is due to the triple-product $`\stackrel{}{s}_b(\stackrel{}{p}_u\times \stackrel{}{p}_s)`$, which involves the $`b`$-quark spin and the momenta of the $`s`$ and $`u`$ quarks. In the SM, this signal turns out to be in the range of 1% to 3% of the total rate, which may be measurable. It might be observable in decays such as $`\mathrm{\Lambda }_b\mathrm{\Lambda }\pi ^+\pi ^{}`$ (though the usual caveats about large hadronic uncertainties apply).
Finally, it is also important to note which T-violating signals are not present. For example, we find that there are no significant T-violating asymmetries in the SM which involve the spin of the $`s`$-quark. Thus, should such an asymmetry be measured, it would be a clear sign of new physics.
In Sec. 2, we compute the triple products present in the decay $`bsu\overline{u}`$, and estimate their sizes. We discuss possible hadron-level applications in Sec. 3. We conclude in Sec. 4.
## 2 Triple Products in $`bsu\overline{u}`$
In the inclusive decay $`bsu\overline{u}`$, the amplitude has two dominant contributions: the tree diagram ($`T`$) due to $`W`$-boson exchange and the loop-level strong penguin diagram ($`P`$). Furthermore, the penguin amplitude contains two dominant terms, $`P_1`$ and $`P_2`$ . These various contributions are given by:
$`T`$ $`=`$ $`{\displaystyle \frac{4G_F}{\sqrt{2}}}V_{ub}V_{us}^{}\left[\overline{u}\gamma _\mu \gamma _Lb\right]\left[\overline{s}\gamma ^\mu \gamma _Lv_u\right]e^{i\delta _t},`$
$`P_1`$ $`=`$ $`{\displaystyle \frac{\alpha _sG_F}{\sqrt{2}\pi }}F_1^cV_{cb}V_{cs}^{}\left[\overline{s}t^\alpha \gamma _\mu \gamma _Lb\right]\left[\overline{u}t_\alpha \gamma ^\mu v_u\right]e^{i\delta _1},`$
$`P_2`$ $`=`$ $`{\displaystyle \frac{\alpha _sG_F}{\sqrt{2}\pi }}\left[{\displaystyle \frac{im_b}{q^2}}F_2\right]V_{tb}V_{ts}^{}\left[\overline{s}t^\alpha \sigma _{\mu \nu }q^\nu \gamma _Rb\right]\left[\overline{u}t_\alpha \gamma ^\mu v_u\right]e^{i\delta _2}.`$ (5)
In the above, $`\gamma _{L(R)}=(1\gamma _5)/2`$, the $`t^\alpha `$ are the Gell-Mann matrices, and the $`\delta _i`$ are the strong phases. In $`P_2`$, $`q`$ is the momentum of the internal gluon. The factors $`F_1^c`$ and $`F_2`$ are functions of $`\left(m_c^2/M_W^2\right)`$ and $`\left(m_t^2/M_W^2\right)`$, respectively, and take the values $`F_1^c5.0`$ and $`F_20.2`$ for $`m_t=160`$ GeV . $`P_1`$ and $`P_2`$ are often called the chromoelectric dipole moment term and chromomagnetic dipole moment term, respectively.
The next step is the calculation of the square of the decay amplitude. We have:
$$||^2=|T|^2+|P_1|^2+|P_2|^2+2Re\left(T^{}P_1\right)+2Re\left(T^{}P_2\right)+2Re\left(P_1^{}P_2\right).$$
(6)
We will see below that the dominant term here is $`|P_1|^2`$.
We find triple products in all of the interference terms above (i.e. the last three terms of $`||^2`$). Before giving the specific forms of these triple products, we make the following general remarks:
* In the calculation, we neglect the masses of the light quarks $`s`$, $`u`$ and $`\overline{u}`$, but we keep the spins (i.e. polarization four-vectors) of these particles (at least to begin with). It turns out that there are no triple products involving the polarization of the $`s`$ quark. (In other words, such terms are suppressed by at least $`m_s/m_b`$.) In light of this, in our results below, we automatically sum over the $`s`$-quark spin states.
This is an interesting result: it suggests that if a triple product involving the $`s`$-quark polarization is observed experimentally, it is probably due to physics beyond the SM.
* Since the $`B`$ meson has spin 0, triple products in $`BV_1V_2`$ cannot involve the spin of the $`b`$-quark. If one sums over the spin of the $`b`$-quark, the only term which contains triple products is the $`TP_1`$ interference term. We will therefore use this term to estimate the size of the T-violating asymmetry in $`BV_1V_2`$ .
* If the spins of the $`u`$ and $`\overline{u}`$ quarks cannot be measured, one can then take them to be unpolarized, i.e. we sum over their polarizations. In this case, only the $`TP_2`$ and $`P_1P_2`$ interferences contain a triple product. This unique signal takes the form $`\stackrel{}{s}_b(\stackrel{}{p}_u\times \stackrel{}{p}_s)`$.
* In all interference terms, there are triple products which involve the three polarizations $`\stackrel{}{s}_b`$, $`\stackrel{}{s}_u`$ and $`\stackrel{}{s}_{\overline{u}}`$. Experimentally, such signals will be extremely difficult to measure, and so are of less interest than the others described here.
### 2.1 $`TP_1`$ interference
Keeping explicit the spins of the $`b`$-, $`u`$\- and $`\overline{u}`$-quarks, the T-odd piece of the $`TP_1`$ interference term is
$`\left[{\displaystyle \underset{sspins}{}}2Re\left(T^{}P_1\right)\right]_{Todd}`$ $`=`$ $`{\displaystyle \frac{16\alpha _sG_F^2F_1^c}{3\pi }}Im\left[V_{cs}^{}V_{cb}V_{us}V_{ub}^{}e^{i\left(\delta _1\delta _t\right)}\right]`$ (7)
$`\times \{2(p_bs_u)ϵ_{\mu \nu \rho \xi }p_b^\mu p_u^\nu p_{\overline{u}}^\rho s_{\overline{u}}^\xi \mathrm{\hspace{0.17em}2}(p_bp_u)ϵ_{\mu \nu \rho \xi }p_b^\mu s_u^\nu p_{\overline{u}}^\rho s_{\overline{u}}^\xi +m_b^2ϵ_{\mu \nu \rho \xi }p_u^\mu s_u^\nu p_{\overline{u}}^\rho s_{\overline{u}}^\xi `$
$`+m_b[(s_bs_u)ϵ_{\mu \nu \rho \xi }p_s^\mu p_u^\nu p_{\overline{u}}^\rho s_{\overline{u}}^\xi (s_bp_u)ϵ_{\mu \nu \rho \xi }p_s^\mu p_{\overline{u}}^\nu s_{\overline{u}}^\rho s_u^\xi `$
$`(p_sp_{\overline{u}})ϵ_{\mu \nu \rho \xi }p_u^\mu s_b^\nu s_u^\rho s_{\overline{u}}^\xi +(p_ss_{\overline{u}})ϵ_{\mu \nu \rho \xi }p_u^\mu p_{\overline{u}}^\nu s_b^\rho s_u^\xi ]\}.`$
Here, $`p_i`$ is the 4-momentum of the $`i`$-quark and $`s_i`$ is its polarization four-vector. Triple products<sup>3</sup><sup>3</sup>3Note that, due to the identity $`g_{\alpha \beta }ϵ_{\mu \nu \rho \xi }g_{\alpha \mu }ϵ_{\beta \nu \rho \xi }g_{\alpha \nu }ϵ_{\mu \beta \rho \xi }g_{\alpha \rho }ϵ_{\mu \nu \beta \xi }g_{\alpha \xi }ϵ_{\mu \nu \rho \beta }=0`$, not all terms of the form $`v_1v_2ϵ_{\mu \nu \rho \xi }v_3^\mu v_4^\nu v_5^\rho v_6^\xi `$ are necessarily independent. are found in the terms $`ϵ_{\mu \nu \rho \xi }v_1^\mu v_2^\nu v_3^\rho v_4^\xi `$.
In the above expression, we see that there are two categories of triple products: those which involve $`s_b`$, the $`b`$-quark polarization, and those which do not. Those terms which include $`s_b`$ (the last four terms in Eq. 7) also include the polarizations of the $`u`$\- and $`\overline{u}`$-quarks ($`s_u`$ and $`s_{\overline{u}}`$). Since all three spins must be measured, these triple products will be extremely difficult to observe experimentally. Because of this, it is the first three terms of Eq. 7 which most interest us, and we therefore isolate them by averaging over $`s_b`$.
Of course, as written, the terms $`ϵ_{\mu \nu \rho \xi }v_1^\mu v_2^\nu v_3^\rho v_4^\xi `$ involve only four-vectors, and therefore do not look like triple products. In order to identify the triple products implicit in these terms, we have to choose a particular frame of reference. The most natural choice is the rest frame of the $`b`$-quark, in which case Eq. 7 then takes the form
$`\left[{\displaystyle \frac{1}{2}}{\displaystyle \underset{b,sspins}{}}2Re\left(T^{}P_1\right)\right]_{Todd}`$ $`=`$ $`{\displaystyle \frac{16\alpha _sG_F^2F_1^c}{3\pi }}Im\left[V_{cs}^{}V_{cb}V_{us}V_{ub}^{}e^{i\mathrm{\Delta }_{1t}}\right]m_b^2`$ (8)
$`\times \left\{s_u^0\stackrel{}{p}_u(\stackrel{}{p}_{\overline{u}}\times \stackrel{}{s}_{\overline{u}})+E_u\stackrel{}{p}_{\overline{u}}(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})+s_{\overline{u}}^0\stackrel{}{p}_u(\stackrel{}{p}_{\overline{u}}\times \stackrel{}{s}_u)+E_{\overline{u}}\stackrel{}{p}_u(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})\right\},`$
where $`\mathrm{\Delta }_{1t}\delta _1\delta _t`$. We therefore see that there are, in fact, four distinct triple products in the $`TP_1`$ interference term.
These triple products depend on the polarization four-vectors of the $`u`$\- and $`\overline{u}`$-quarks, whose most general form is
$$s_i^\mu =(\frac{\stackrel{}{n}_i\stackrel{}{p}_i}{m_i},\stackrel{}{n}_i+\frac{\stackrel{}{n}_i\stackrel{}{p}_i}{m_i(E_i+m_i)}\stackrel{}{p}_i),$$
(9)
for $`i=u,\overline{u}`$. In the above, $`\stackrel{}{n}_i`$ is the polarization vector of the $`i`$-quark in its rest frame, and satisfies $`|\stackrel{}{n}_i|=1`$.
The triple products in Eq. (8) all involve two spins, which makes their evaluation somewhat problematic. Since one has to choose directions for $`\stackrel{}{s}_u`$ and $`\stackrel{}{s}_{\overline{u}}`$, there are an infinite number of possibilities. We have studied three cases:
1. The polarizations of the $`u`$\- and $`\overline{u}`$-quarks in their respective centre-of-mass frames both point in a particular direction, say $`\stackrel{}{n}_u=\stackrel{}{n}_{\overline{u}}=\widehat{z}`$.
2. The polarizations of the $`u`$\- and $`\overline{u}`$-quarks are perpendicular, say $`\stackrel{}{n}_u=\widehat{z}`$ and $`\stackrel{}{n}_{\overline{u}}=\widehat{x}`$.
3. The polarization of the $`u`$-quark is longitudinal, $`\stackrel{}{n}_u=\widehat{p}_u`$ while $`\stackrel{}{n}_{\overline{u}}=\widehat{z}`$.
In the following we will refer to these three scenarios as Case I, Case II and Case III, respectively.
In order to compute the size of these triple-product asymmetries, in addition to integrating over phase space, we also need estimates of the sizes of the weak and strong phases. In the Wolfenstein parametrization , we can write the T-odd combination of CKM and strong phases as
$$Im\left[V_{cs}^{}V_{cb}V_{us}V_{ub}^{}e^{i\mathrm{\Delta }_{1t}}\right]=A^2\lambda ^6\left[\eta \mathrm{cos}\mathrm{\Delta }_{1t}+\rho \mathrm{sin}\mathrm{\Delta }_{1t}\right].$$
(10)
CP violation in the CKM matrix is parametrized by the parameter $`\eta `$. As discussed in the introduction, nonzero strong phases can fake a T-violating signal. The term $`\rho \mathrm{sin}\mathrm{\Delta }_{1t}`$ in the above expression is an example of such a fake signal. However, by forming a true T-violating asymmetry $`𝒜_T`$ (Eq. 3), one can eliminate this fake signal. In this case $`𝒜_T\eta \mathrm{cos}\mathrm{\Delta }_{1t}`$.
At present, $`\eta `$ is constrained to lie in the range $`0.23\eta 0.50`$ . Turning to the strong phase, the tree-level phase $`\delta _t`$ is usually assumed to be small: the logic is that, roughly speaking, the quarks will hadronize before having time to exchange gluons. On the other hand, for the $`bsu\overline{u}`$ penguin amplitude, it is often assumed that strong phases come from the absorptive part of the penguin contribution . Since $`P_1`$ involves an internal $`c`$-quark, it is possible that $`\delta _10`$, which of course implies that $`\mathrm{\Delta }_{1t}0`$. Even so, for simplicity, in our calculation we assume that $`\mathrm{\Delta }_{1t}`$ is small enough that $`\mathrm{cos}\mathrm{\Delta }_{1t}1`$ is a good approximation. However, the reader should be aware that the asymmetries may be reduced should this strong phase be large. (Note that the T-violating signal is maximal when $`\mathrm{cos}\mathrm{\Delta }_{1t}=1`$. For comparison, direct CP-violating rate asymmetries require the strong phase to be nonzero.)
We have performed the phase-space integration using the computer program RAMBO. For each of the three cases above we have calculated the four T-violating asymmetries \[see Eq. (3)\] $`𝒜_T^1`$, $`𝒜_T^2`$, $`𝒜_T^3`$, and $`𝒜_T^4`$, which correspond respectively to the four triple products of Eq. (8): $`s_u^0\stackrel{}{p}_u(\stackrel{}{p}_{\overline{u}}\times \stackrel{}{s}_{\overline{u}})`$, $`E_u\stackrel{}{p}_{\overline{u}}(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})`$, $`s_{\overline{u}}^0\stackrel{}{p}_u(\stackrel{}{p}_{\overline{u}}\times \stackrel{}{s}_u)`$, and $`E_{\overline{u}}\stackrel{}{p}_u(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})`$. We have taken $`\eta =0.4`$. Our results are as follows.
1. Case I: The asymmetries $`𝒜_T^1`$ and $`𝒜_T^3`$ are both negligible. However, we find that $`𝒜_T^2=𝒜_T^44.6\%`$.
2. Case II: $`𝒜_T^1`$ and $`𝒜_T^3`$ are negligible, while $`𝒜_T^2=𝒜_T^43.9\%`$.
3. Case III: Here the triple products $`\stackrel{}{p}_u(\stackrel{}{p}_{\overline{u}}\times \stackrel{}{s}_u)`$ and $`\stackrel{}{p}_u(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})`$ vanish identically, so that $`𝒜_T^3=𝒜_T^4=0`$. We find that the other two asymmetries are tiny: $`𝒜_T^2=𝒜_T^10.09\%`$.
From the above, we conclude that the asymmetries $`𝒜_T^1`$ and $`𝒜_T^3`$ are both very small in the SM, and that $`𝒜_T^2`$ and $`𝒜_T^4`$ can be as large as about 5%.
Note that the triple product in $`BV_1V_2`$ discussed by Valencia is of the form $`\stackrel{}{k}(\stackrel{}{ϵ}_1\times \stackrel{}{ϵ}_2)`$. In Eq. (8), it is the terms $`\stackrel{}{p}_{\overline{u}}(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})`$ and $`\stackrel{}{p}_u(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})`$ which could potentially give such a triple-product signal. We have found that the asymmetries $`𝒜_T^2`$ and $`𝒜_T^4`$, which correspond to these triple products, can be reasonably big ($`<5\%`$). This is consistent with the results found by Valencia at the meson level, and suggests that the SM does indeed predict a measurable T-violating asymmetry in $`BV_1V_2`$ decays. (Of course, due to our difficulties in understanding hadronization, it is not possible to use the quark-level result to predict the size of the asymmetry for a specific meson-level decay.)
Finally, for comparison, consider the decay-rate asymmetry, calculated by Hou for the same process :
$$a_{CP}(bsu\overline{u})1.4\%$$
(11)
We therefore see that one expects T-violating triple-product asymmetries in $`bsu\overline{u}`$ to be considerably larger than the decay rate asymmetry.
### 2.2 $`P_1P_2`$ interference
The T-odd piece of the $`P_1P_2`$ interference term is
$`\left[{\displaystyle \underset{sspins}{}}2Re\left(P_1^{}P_2\right)\right]_{Todd}`$ $`=`$ $`{\displaystyle \frac{4\alpha _s^2G_F^2F_1^cF_2m_b}{3\pi ^2q^2}}Im\left[V_{ts}^{}V_{tb}V_{cs}V_{cb}^{}e^{i\left(\delta _2\delta _1\right)}\right]`$ (12)
$`\times \{[p_b(p_up_{\overline{u}})(1s_us_{\overline{u}})(s_{\overline{u}}p_s)(s_up_{\overline{u}})+(s_up_s)(s_{\overline{u}}p_u)]ϵ_{\mu \nu \rho \xi }p_b^\mu s_b^\nu p_u^\rho p_s^\xi `$
$`+\left[(s_up_{\overline{u}})(p_up_s){\displaystyle \frac{q^2}{2}}(s_up_s)\right]ϵ_{\mu \nu \rho \xi }p_b^\mu s_b^\nu p_s^\rho s_{\overline{u}}^\xi `$
$`+[(s_{\overline{u}}p_u)(p_{\overline{u}}p_s){\displaystyle \frac{q^2}{2}}(s_{\overline{u}}p_s)]ϵ_{\mu \nu \rho \xi }p_b^\mu s_b^\nu p_s^\rho s_u^\xi \}.`$
Here, if we average over the $`b`$-quark spin states, there is no T-violating signal at all.
We note that most of the terms in Eq. 12 correspond to triple products in which three spins must be measured. As we have already discussed, such signals are very difficult to observe experimentally, and so do not interest us. There is one term, however, which does not involve three spins, and it can be isolated by summing over the $`u`$\- and $`\overline{u}`$-quark spin states:
$`\left[{\displaystyle \underset{u,\overline{u},sspins}{}}2Re\left(P_1^{}P_2\right)\right]_{Todd}`$ $`=`$ $`{\displaystyle \frac{16\alpha _s^2G_F^2F_1^cF_2m_b}{3\pi ^2q^2}}Im\left[V_{ts}^{}V_{tb}V_{cs}V_{cb}^{}e^{i\left(\delta _2\delta _1\right)}\right]`$ (13)
$`\times p_b\left(p_up_{\overline{u}}\right)ϵ_{\mu \nu \rho \xi }p_b^\mu s_b^\nu p_u^\rho p_s^\xi .`$
In the rest frame of the $`b`$-quark, the triple product takes the form $`m_b^2(E_uE_{\overline{u}})\stackrel{}{s}_b(\stackrel{}{p}_u\times \stackrel{}{p}_s)`$. Integrating over phase space with RAMBO, we find that the $`P_1P_2`$ T-violating asymmetry is $`O(10^5)`$, which is negligible.
### 2.3 $`TP_2`$ interference
Like $`P_1P_2`$ interference, the $`TP_2`$ interference term contains two types of triple produts: (i) those involving a single quark polarization, $`s_b`$, and (ii) those involving the three polarization four-vectors $`s_b`$, $`s_u`$ and $`s_{\overline{u}}`$. As usual, we are not interested in triple products involving three spins, and so we can therefore sum over $`s_u`$ and $`s_{\overline{u}}`$. The T-odd piece of the $`TP_2`$ interference term is then given by
$`\left[{\displaystyle \underset{u,\overline{u},sspins}{}}2Re\left(T^{}P_2\right)\right]_{Todd}`$ $`=`$ $`{\displaystyle \frac{128\alpha _sG_F^2F_2m_b}{3\pi q^2}}Im\left[V_{ts}^{}V_{tb}V_{us}V_{ub}^{}e^{i\mathrm{\Delta }_{2t}}\right]`$ (14)
$`\times p_sp_uϵ_{\mu \nu \rho \xi }p_b^\mu s_b^\nu p_u^\rho p_s^\xi ,`$
where $`\mathrm{\Delta }_{2t}\delta _2\delta _t`$.
As was the case for the $`TP_1`$ interference term, the T-violating asymmetry $`𝒜_T`$ is proportional to $`\eta \mathrm{cos}\mathrm{\Delta }_{2t}`$. And, as before, we expect the $`\delta _t`$ piece of the strong phase $`\mathrm{\Delta }_{2t}`$ to be small. However, there is a difference here compared to the $`TP_1`$ case: previously, the penguin amplitude $`P_1`$ involved an internal $`c`$-quark, and so it was possible that the strong phase $`\delta _{1t}`$, which is related to the absorptive part of the amplitude, could be sizeable. Here, the triple product involves only the $`t`$-quark penguin contribution $`P_2`$, which is purely dispersive, and so leads to $`\delta _2=0`$. Thus, it is an excellent approximation to set $`\mathrm{\Delta }_{2t}0`$.
In the rest frame of the $`b`$-quark, the triple-product of Eq. 14 is $`m_bp_sp_u\stackrel{}{s}_b(\stackrel{}{p}_u\times \stackrel{}{p}_s)`$. Integrating over phase space using RAMBO, and using the allowed range for $`\eta `$, we find that this T-violating triple-product asymmetry can be of the order of several percent:
$$1.3\%<𝒜_T(bsu\overline{u})<3.2\%.$$
(15)
This could conceivably be measured at a future experiment.
Furthermore, if it is found that this asymmetry is considerably larger than the above values, it is probably a signal of new physics. For example, in some models of new physics, the chromomagnetic dipole moment $`F_2`$ can be enhanced up to ten times its SM value . This will clearly have an enormous affect on the above asymmetry.
## 3 Applications
In the previous section, in our study of the quark-level decay $`bsu\overline{u}`$ within the SM, we found two classes of triple products whose T-violating asymmetry is large. They are: (i) $`E_u\stackrel{}{p}_{\overline{u}}(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})`$ and $`E_{\overline{u}}\stackrel{}{p}_u(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})`$, and (ii) $`m_bp_sp_u\stackrel{}{s}_b(\stackrel{}{p}_u\times \stackrel{}{p}_s)`$. The next obvious question is then: how can one test these results?
The ideal way would be to make triple-product measurements inclusively. If this were possible, then it would be straightforward to compare the experimental values with the theoretical predictions. However, this may not be experimentally feasible, in which case we must turn to exclusive $`B`$ decays.
The first class of triple-product asymmetries can be studied in $`BV_1V_2`$ decays which are dominated by the quark-level process $`bsu\overline{u}`$. Examples of such decays include $`\overline{B_d^0}\rho K^{}`$, $`\overline{B_s^0}K^+K^{}`$, $`B_c^{}D^{}K^{}`$, etc. These have been examined by Valencia, and we refer the reader to Ref. for details.
Turning to the second class of triple products, it is clear that we cannot use decays of $`B`$ mesons to obtain these asymmetries: since the $`B`$-meson spin is zero, there is no way to measure the spin of the $`b`$-quark (which is the only spin contributing to the triple product). However, one possibility would be to use the $`\mathrm{\Lambda }_b`$ baryon, whose spin is largely that of the $`b`$ quark. For example, we can consider the process $`\mathrm{\Lambda }_b\mathrm{\Lambda }\pi ^+\pi ^{}`$. The triple product $`\stackrel{}{s}_b(\stackrel{}{p}_u\times \stackrel{}{p}_s)`$ can be roughly equated to $`\stackrel{}{s}_{\mathrm{\Lambda }_b}(\stackrel{}{p}_{\pi ^+}\times \stackrel{}{p}_\mathrm{\Lambda })`$.
Another possibility is to consider a $`B^{}`$ meson decaying into any two mesons $`X_sX`$, where $`X_s`$ and $`X`$ then decay respectively into mesons $`\mathrm{\Phi }_1\mathrm{\Phi }_2`$ and $`\mathrm{\Phi }_3\mathrm{\Phi }_4`$ (since with only one spin, we need three independent momenta). The triple product $`\stackrel{}{s}_b(\stackrel{}{p}_u\times \stackrel{}{p}_s)`$ may then be roughly related to $`\stackrel{}{s}_B^{}(\stackrel{}{p}_{\mathrm{\Phi }_1}\times \stackrel{}{p}_{\mathrm{\Phi }_3})`$.
In the above, we have been deliberately vague about the relationship between the triple products at the quark and hadron levels. We do not understand hadronization of quarks into mesons all that well, and when one adds the complication of relating the quark spins to the hadron spins, things are even more uncertain. Given that there is a large quark-level triple-product asymmetry, there may be one at the hadron level. However, if experiment does not find such an asymmetry, this may not be a sign of new physics – it may simply mean that the asymmetry has been washed out during hadronization. Regardless of the results, studies of this kind are likely to help us understand how quarks hadronize into mesons and baryons.
Finally, we note that certain quark-level triple products are predicted to be small in the SM. For example, triple products involving the spin of the $`s`$-quark are suppressed by powers of its mass. Hence, if a T-violating asymmetry due to a triple product involving the $`s`$-quark spin were found to be sizeable, this would probably indicate the presence of new physics. The decay $`\mathrm{\Lambda }_b\mathrm{\Lambda }\pi ^+\pi ^{}`$, which was mentioned above, can be used to test this. The spin of the $`\mathrm{\Lambda }`$ is due mostly to the $`s`$-quark spin. So any T-violating asymmetry involving the spin of the $`\mathrm{\Lambda }`$, such as $`\stackrel{}{s}_{\mathrm{\Lambda }_b}(\stackrel{}{s}_\mathrm{\Lambda }\times \stackrel{}{p}_\mathrm{\Lambda })`$, should be tiny in the SM.
As another example, recall that we found that $`TP_1`$ interference produced the triple products $`s_u^0\stackrel{}{p}_u(\stackrel{}{p}_{\overline{u}}\times \stackrel{}{s}_{\overline{u}})`$ and $`s_{\overline{u}}^0\stackrel{}{p}_u(\stackrel{}{p}_{\overline{u}}\times \stackrel{}{s}_u)`$. However, the corresponding T-violating asymmetries $`𝒜_T^1`$ and $`𝒜_T^3`$ turned out to be suppressed dynamically. Consider the decay of a $`B`$-meson to two vector mesons, $`BV_1V_2`$, where the $`V_2`$ then subsequently decays to two mesons $`\mathrm{\Phi }_1\mathrm{\Phi }_2`$. Roughly speaking, one can relate $`s_u^0\stackrel{}{p}_u(\stackrel{}{p}_{\overline{u}}\times \stackrel{}{s}_{\overline{u}})`$ to $`ϵ_{V_1}^0\stackrel{}{p}_{V_1}(\stackrel{}{ϵ}_{V_2}\times \stackrel{}{p}_{\mathrm{\Phi }_1})`$. Thus, the measurement of a nonzero value for this latter triple-product asymmetry would be a signal for new physics.
## 4 Conclusions
We have calculated the quark-level triple-product correlations in the decay $`bsu\overline{u}`$ within the standard model. Although several such triple products are present, we find that only two types lead to sizeable T-violating asymmetries.
The first type includes $`\stackrel{}{p}_u(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})`$ and $`\stackrel{}{p}_{\overline{u}}(\stackrel{}{s}_u\times \stackrel{}{s}_{\overline{u}})`$. We find that the corresponding T-violating asymmetries can be as large as about 5%. This triple product can be probed in $`BV_1V_2`$ decays, where $`V_1`$ and $`V_2`$ are vector mesons .
The second type is $`\stackrel{}{s}_b(\stackrel{}{p}_u\times \stackrel{}{p}_s)`$, where $`\stackrel{}{s}_b`$ is the polarization of the $`b`$-quark, and $`\stackrel{}{p}_u`$ and $`\stackrel{}{p}_s`$ are the momenta of the $`u`$\- and $`s`$-quark, respectively. We calculate that the T-violating asymmetry for this triple product is in the range 1%–3%, which may be measurable. There are several ways to try to search for this triple-product asymmetry. For example, one could study the decay $`\mathrm{\Lambda }_b\mathrm{\Lambda }\pi ^+\pi ^{}`$, looking for a nonzero triple product $`\stackrel{}{s}_{\mathrm{\Lambda }_b}(\stackrel{}{p}_{\pi ^+}\times \stackrel{}{p}_\mathrm{\Lambda })`$. Another possibility is to examine the decay $`B^{}X_sX`$, where $`X_s`$ and $`X`$ then decay respectively into mesons $`\mathrm{\Phi }_1\mathrm{\Phi }_2`$ and $`\mathrm{\Phi }_3\mathrm{\Phi }_4`$, and to search for the triple product $`\stackrel{}{s}_B^{}(\stackrel{}{p}_{\mathrm{\Phi }_1}\times \stackrel{}{p}_{\mathrm{\Phi }_3})`$.
The fact that we find only two large triple-product correlations has interesting consequences. If a triple product is tiny at the quark level, it is probably tiny at the hadron level as well. After all, the hadronization of quarks into hadrons is a strong-interaction process, and QCD is CP-conserving. It is therefore difficult to see how one can generate a large triple-product correlation at the hadron level, given that it is small at the quark level. (Of course, the converse is not necessarily true: it is quite possible that a quark-level CP-violating effect might be “washed out” during hadronization.)
From the point of view of looking for physics beyond the SM, it is therefore important to identify those triple-product asymmetries which are expected to be small in the SM. If such asymmetries are found to be large, this is probably a signal of new physics. For example, we find that triple products involving the spin of the $`s`$-quark are suppressed by powers of its mass. Thus, if, for instance, a sizeable T-violating asymmetry of the form $`\stackrel{}{s}_{\mathrm{\Lambda }_b}(\stackrel{}{s}_\mathrm{\Lambda }\times \stackrel{}{p}_\mathrm{\Lambda })`$ were found in the decay $`\mathrm{\Lambda }_b\mathrm{\Lambda }\pi ^+\pi ^{}`$, this would be compelling evidence for the presence of new physics, since the spin of the $`\mathrm{\Lambda }`$ is due largely to the $`s`$-quark spin.
Note that we have performed the calculation at the quark level, and the passage from quarks to hadrons is not well understood. Thus, in addition to being an interesting signal of T and CP violation, the study of T-violating triple-product asymmetries may help us understand aspects of hadronization.
## 5 Acknowledgments
W.B. would like to thank P. Depommier for helpful discussions, and G. Azuelos for help with RAMBO. D.L. thanks B. Kayser and A. Soni for useful conversations. This work was financially supported by NSERC of Canada.
|
warning/0005/cond-mat0005181.html
|
ar5iv
|
text
|
# Spectral Functions for the Tomonaga–Luttinger and Luther–Emery Liquids
## I Introduction
The interacting one-dimensional electron gas (1DEG) has intrigued physicists since the pioneering work of Tomonaga. Through the collective effort of many of them, and especially due to the advancement of the bosonization technique, a large number of its properties have been discovered.
The past two decades have witnessed an experimental effort to identify and study, using various probes, quasi-one-dimensional systems. Among the materials studied one finds organic compounds such as polyacetylene and the Bechgaard salts as well as several families of inorganic quasi-one-dimensional materials (for example, the transition metal bronzes). The edge states in the quantum Hall effect constitute yet another realization of one-dimensional interacting systems. Progress made in nanoscale fabrication techniques have made it possible to artificially define one-dimensional channels in two-dimensional electronic heterostructures. Most recently, experimental evidence has emerged for the existence of stripe phases in doped antiferromagnets, such as the copper-oxide high temperature superconductors. In these phases the doped holes segregate into quasi-one-dimensional metallic regions embedded in a predominantly antifferomagnetic background.
In recent years angle resolved photoemission spectroscopy (ARPES) has matured into a powerful experimental method for probing the single particle properties of strongly correlated systems. ARPES measurements on the Bechgaard salts, the transition metalchalcogenides , the blue bronzes and the cuprates have provided evidence in favor of correlated one-dimensional physics although full agreement with the predicted theoretical picture is still lacking.
The zero temperature correlation functions of the gapless Tomonaga-Luttinger liquid, and in particular the single hole spectral function which is measured by ARPES, have been calculated previously for the spinless as well as for the spinfull case. Extending these results to finite temperatures is desirable on the following grounds. First, real experiments are carried out at finite temperatures and should be contrasted with theoretical predictions relevant for such conditions. Secondly, the Tomonaga-Luttinger liquid is a quantum critical system which consequently exhibits scaling behavior. It is a rare example where not only the scaling exponents but indeed the entire scaling functions can be computed explicitly. In Section II we obtain the scaling form of the single hole spectral function and the spin dynamic structure factor of the spinfull Tomonaga-Luttinger liquid. Closed form analytical expressions are derived for a number of special cases. These results complement previous numerical evaluations of the finite temperature spectral functions and allow for easy investigation of their properties in few limits which are of physical interest. Integrals of these spectral functions are considered as well.
When backward scattering of electrons becomes relevant the 1DEG develops a spin gap and is described by the Luther-Emery liquid (Umklapp processes may create a charge gap). While the spectrum of the Luther-Emery liquid can be readily derived at the special “free fermion” point $`K_s=1/2`$, the evaluation of electronic correlation functions there is non-trivial due to their highly non-local form in terms of the pseudo- fermions. Several authors have made progress in this direction, see Refs. B,B, and in section III we extend and correct their study and obtain exact expressions for the single hole spectral function at the free fermion point for temperatures much smaller than the spin gap. Some technical details are relegated to appendices.
## II The Tomonaga–Luttinger Liquid
### A The model
The Tomonaga–Luttinger model embodies the low energy and long wavelength physics of the 1DEG. It is composed of two branches of left $`(\eta =1)`$ and right $`(\eta =+1)`$ moving massless Dirac fermions constructed around the left and right Fermi points of the 1DEG. We will consider the case where the fermions carry spin $`1/2`$ and denote by $`\sigma =\pm 1`$ their spin polarization. In the absence of backward scattering and Umklapp processes the Hamiltonian density is given by $`(\mathrm{}=1)`$
$`=`$ $``$ $`v_F{\displaystyle \underset{\eta ,\sigma =\pm 1}{}}\eta \psi _{\eta ,\sigma }^{}i_x\psi _{\eta ,\sigma }`$ (1)
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\eta =\pm 1}{}}g_{2,c}\rho _\eta (x)\rho _\eta (x)+g_{4,c}\rho _\eta (x)\rho _\eta (x)`$ (2)
$`+`$ $`2{\displaystyle \underset{\eta =\pm 1}{}}g_{2,s}\mathrm{S}_\eta ^z(x)\mathrm{S}_\eta ^z(x)+g_{4,s}\mathrm{S}_\eta ^z(x)\mathrm{S}_\eta ^z(x),`$ (3)
where $`v_F`$ is the noninteracting Fermi velocity and
$`\rho _\eta `$ $`=`$ $`{\displaystyle \underset{\sigma }{}}\psi _{\eta ,\sigma }^{}\psi _{\eta ,\sigma },`$ (4)
$`𝐒_\eta `$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\sigma ,\sigma ^{}}{}}\psi _{\eta ,\sigma }^{}𝝉_{\sigma \sigma ^{}}\psi _{\eta ,\sigma ^{}},`$ (5)
with $`𝝉`$ being the Pauli matrices. Despite its appearance, the last term in (1) does not break the $`SU(2)`$ spin symmetry, since $`\mathrm{S}_{\eta }^{z}{}_{}{}^{2}(x)=lim_{x^{}x}𝐒_\eta (x)𝐒_\eta (x^{})/2+\rho _\eta ^2(x)/8`$; the $`g_{2,s}`$ term is the only term which breaks this symmetry.
In order to compute the single particle properties of the model one uses the bosonization identity
$$\psi _{\eta ,\sigma }=\frac{1}{\sqrt{2\pi a}}F_{\eta ,\sigma }\mathrm{exp}[i\eta k_Fxi\mathrm{\Phi }_{\eta ,\sigma }(x)],$$
(6)
which expresses the fermionic fields in terms of self-dual fields $`\mathrm{\Phi }_{\eta ,\sigma }(x)`$ obeying $`[\mathrm{\Phi }_{\eta ,\sigma }(x),\mathrm{\Phi }_{\eta ^{},\sigma ^{}}(x^{})]=i\pi \delta _{\eta ,\eta ^{}}\delta _{\sigma ,\sigma ^{}}\mathrm{sign}(xx^{})`$. The Klein factors $`F_{\eta ,\sigma }`$ are responsible for reproducing the correct anti-commutation relations between different fermionic species and $`a`$ is a short distance cutoff that is taken to zero at the end of the calculation. Here and throughout the paper we consider the limit where the size of the system $`L`$ is taken to infinity and correspondingly ignore terms of the order $`1/L`$ (for a discussion of finite size effects in the finite temperature case see Refs. B-B).
The demonstration of the celebrated charge-spin separation in the model is facilitated by expressing $`\mathrm{\Phi }_{\eta ,\sigma }`$ in terms of the bosonic fields $`\varphi _c,\varphi _s`$ and their conjugated momenta $`_x\theta _c,_x\theta _s`$
$$\mathrm{\Phi }_{\eta ,\sigma }=\sqrt{\pi /2}[(\theta _c\eta \varphi _c)+\sigma (\theta _s\eta \varphi _s)],$$
(7)
in terms of which the charge and spin densities are given by
$`\rho (x)`$ $`=`$ $`{\displaystyle \underset{\eta }{}}\rho _\eta (x)=\sqrt{2/\pi }_x\varphi _c,`$ (8)
$`\mathrm{S}^z(x)`$ $`=`$ $`{\displaystyle \underset{\eta }{}}\mathrm{S}_\eta ^z(x)=\sqrt{1/2\pi }_x\varphi _s.`$ (9)
The decomposition (7) also facilitates the diagonalization of the Hamiltonian. It becomes a sum of two independent pieces describing noninteracting charge and spin density waves which are the elementary excitations of the system
$$=\underset{\alpha =c,s}{}\frac{v_\alpha }{2}\left[K_\alpha (_x\theta _\alpha )^2+\frac{(_x\varphi _\alpha )^2}{K_\alpha }\right].$$
(10)
The velocities of the collective modes are
$$v_\alpha =\sqrt{\left(v_F+\frac{g_{4,\alpha }}{\pi }\right)^2\left(\frac{g_{2,\alpha }}{\pi }\right)^2},$$
(11)
and the parameters $`K_\alpha `$, which determine the power-law behavior of the correlation functions, read
$$K_\alpha =\sqrt{\frac{\pi v_F+g_{4,\alpha }g_{2,\alpha }}{\pi v_F+g_{4,\alpha }+g_{2,\alpha }}}.$$
(12)
The effective parameters that enter the Hamiltonian depend on the specifics of the model for which (1) is a low energy fixed point. We note, however, that spin-rotation invariance dictates $`g_{2,s}=0`$ and consequently $`K_s=1`$.
### B The space-time correlation functions
The bosonized expression for the fermion field operators (6) and the fact that the theory reduces to that of free bosons make it straightforward to derive expressions for the space-time response functions (which we denote with a tilde). We will focus on the finite temperature single hole Green function
$$\stackrel{~}{G}_\eta ^<(x,t;T)=\psi _{\eta ,\sigma }^{}(x,t)\psi _{\eta ,\sigma }(0,0).$$
(13)
We compute this Green function, rather than the more usual time ordered or retarded Green functions, because it includes only the one-hole states. This fact makes it relevant to ARPES whose cross section is directly proportional to the Fourier transform of $`\stackrel{~}{G}^<`$. Other Green functions can be easily obtained from it, since for the model (1) $`\stackrel{~}{G}_\eta ^>(x,t;T)=\psi _{\eta ,\sigma }(0,0)\psi _{\eta ,\sigma }^{}(x,t)=\stackrel{~}{G}_\eta ^<(x,t;T)`$.
We will also consider the $`2k_F`$ component of the transverse spin dynamic structure factor, which is measured by polarized neutron scattering
$`\stackrel{~}{𝒮}(x,t;T)`$ $`=`$ $`\mathrm{S}_{2k_F}^{x}{}_{}{}^{}(x,t)\mathrm{S}_{2k_F}^x(0,0)`$ (14)
$`+`$ $`\mathrm{S}_{2k_F}^{y}{}_{}{}^{}(x,t)\mathrm{S}_{2k_F}^y(0,0).`$ (15)
where
$$𝐒_{2k_F}=\frac{1}{2}\underset{\sigma ,\sigma ^{}}{}\psi _{1,\sigma }^{}𝝉_{\sigma \sigma ^{}}\psi _{1,\sigma ^{}},$$
(16)
and the $`𝝉`$ are the Pauli matrices.
The Tomonaga-Luttinger liquid (1) is a quantum critical system. It also exhibits spin-charge separation. This implies a scaling form for the response functions with separate spin and charge pieces. Specifically one finds
$`\stackrel{~}{G}_\eta ^<(x,t;T)=`$ $`{\displaystyle \frac{1}{2\pi a}}e^{i\eta k_Fx}\left({\displaystyle \frac{a}{\lambda _{T,c}}}\right)^{2\gamma _c+\frac{1}{2}}\left({\displaystyle \frac{a}{\lambda _{T,s}}}\right)^{2\gamma _s+\frac{1}{2}}`$ (18)
$`\times \stackrel{~}{g}_c({\displaystyle \frac{x}{\lambda _{T,c}}},{\displaystyle \frac{v_ct}{\lambda _{T,c}}})\stackrel{~}{g}_s({\displaystyle \frac{x}{\lambda _{T,s}}},{\displaystyle \frac{v_st}{\lambda _{T,s}}}),`$
where we introduced the thermal lengths ($`k_B=1`$)
$$\lambda _{T,\alpha }=\frac{v_\alpha }{\pi T},$$
(19)
and the exponents
$$\gamma _\alpha =\frac{1}{8}(K_\alpha +K_\alpha ^12),$$
(20)
defined so that $`\gamma _\alpha =0`$ for noninteracting fermions. Since the spin and charge sectors are formally invariant under separate Lorentz transformations the functions $`\stackrel{~}{g}_\alpha `$ also spilt into right and left moving parts
$$\stackrel{~}{g}_\alpha (x,t)=\stackrel{~}{h}_{\gamma _\alpha +\frac{1}{2}}(\eta xt)\stackrel{~}{h}_{\gamma _\alpha }^{}(\eta x+t),$$
(21)
where
$$\stackrel{~}{h}_\gamma (x)=\left[i\mathrm{sinh}(x+ia)\right]^\gamma .$$
(22)
Similarly, for the spin correlation function one finds
$`\stackrel{~}{𝒮}(x,t;T)=`$ $`{\displaystyle \frac{1}{(2\pi a)^2}}e^{2ik_Fx}\left({\displaystyle \frac{a}{\lambda _{T,c}}}\right)^{2\beta _c}\left({\displaystyle \frac{a}{\lambda _{T,s}}}\right)^{2\beta _s}`$ (24)
$`\times \stackrel{~}{C}_c({\displaystyle \frac{x}{\lambda _{T,c}}},{\displaystyle \frac{v_ct}{\lambda _{T,c}}})\stackrel{~}{C}_s({\displaystyle \frac{x}{\lambda _{T,s}}},{\displaystyle \frac{v_st}{\lambda _{T,s}}}),`$
where
$$\stackrel{~}{C}_\alpha (x,t)=\stackrel{~}{h}_{\beta _\alpha }(xt)\stackrel{~}{h}_{\beta _\alpha }^{}(x+t).$$
(26)
Here we introduced the exponents
$$\beta _c=\frac{K_c}{2},\beta _s=\frac{1}{2K_s}.$$
(27)
We note that the perpendicular component of the spin dynamic structure factor
$$\stackrel{~}{𝒮}^z(x,t;T)=\mathrm{S}_{2k_F}^{z}{}_{}{}^{}(x,t)\mathrm{S}_{2k_F}^z(0,0),$$
(28)
is obtained from the result for the transverse part (24) after multiplying it by an overall factor of $`1/2`$ and using the exponents
$$\beta _c=\frac{K_c}{2},\beta _s=\frac{K_s}{2},$$
(29)
instead of (27). This holds true also for the spectral functions calculated in the following subsections. Of coarse, in the spin rotation invariant case $`\stackrel{~}{𝒮}^z=\stackrel{~}{𝒮}/2`$.
### C The spectral functions
It is a general feature of the field-theoretic approach to this problem that relatively simple expressions are obtained for the space-time dependent correlation functions. For comparison with experiments, however, we are typically interested in the Fourier transform of these correlation functions. Conceptually, this is simple, and indeed, to evaluate the Fourier transform, we need only to perform a two dimensional integral. However, it is not generally simple to carry out this calculation analytically. Below we consider the cases where this can be done.
As we noted before, the critical nature of the model leads to a scaling form for the spectral functions. A simplifying feature introduced by the spin-charge separation is the ability to express these scaling functions as a convolution of spin and charge parts
$`G^<(\stackrel{~}{k},\stackrel{~}{\omega };T)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑x𝑑te^{i[\eta (k+k_F)x\omega t]}\stackrel{~}{G}_\eta ^<(x,t;T)`$ (30)
$`=`$ $`{\displaystyle \frac{a}{(2\pi )^3v_c}}\left({\displaystyle \frac{a}{\lambda _{T,c}}}\right)^{2\gamma _c\frac{1}{2}}\left({\displaystyle \frac{a}{\lambda _{T,s}}}\right)^{2\gamma _s\frac{1}{2}}`$ (31)
$`\times `$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑q𝑑\nu g_c(q,\nu )g_s(\stackrel{~}{k}rq,\stackrel{~}{\omega }\nu )`$ (32)
where we introduce the velocity ratio $`r=v_s/v_c`$, define the scaling variables
$$\stackrel{~}{k}=\frac{v_sk}{\pi T},\stackrel{~}{\omega }=\frac{\omega }{\pi T},$$
(34)
and
$$g_\alpha (k,\omega )=\frac{1}{2}h_{\gamma _\alpha +\frac{1}{2}}\left(\frac{\omega +k}{2}\right)h_{\gamma _\alpha }\left(\frac{\omega k}{2}\right).$$
(35)
$`h_\gamma (k)`$, the Fourier transform of $`\stackrel{~}{h}_\gamma (x)`$, is evaluated in Appendix A, where some of its properties are listed as well. For non-integer values of $`\gamma `$ it is given by
$$h_\gamma (k)=\mathrm{Re}\left[(2i)^\gamma B(\frac{\gamma ik}{2},1\gamma )\right],$$
(36)
where $`B(x,y)`$ is the Beta function.
In a similar fashion we obtain for the spin susceptibility
$`𝒮(\stackrel{~}{k},\stackrel{~}{\omega };T)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑x𝑑te^{i[(k+2k_F)x\omega t]}\stackrel{~}{𝒮}(x,t;T)`$ (37)
$`=`$ $`{\displaystyle \frac{1}{(2\pi )^4v_c}}\left({\displaystyle \frac{a}{\lambda _{T,c}}}\right)^{2\beta _c1}\left({\displaystyle \frac{a}{\lambda _{T,s}}}\right)^{2\beta _s1}`$ (38)
$`\times `$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑q𝑑\nu C_c(q,\nu )C_s(\stackrel{~}{k}rq,\stackrel{~}{\omega }\nu )`$ (39)
where
$$C_\alpha (k,\omega )=\frac{1}{2}h_{\beta _\alpha }\left(\frac{\omega +k}{2}\right)h_{\beta \alpha }\left(\frac{\omega k}{2}\right).$$
(41)
Henceforth $`k`$ is measured relative to $`k_F`$ and $`2k_F`$ when computing $`G^<`$ and $`𝒮`$, respectively. We also define the Fourier transform with respect to $`\eta x`$. This has the effect of making $`k`$ positive outside the Fermi surface and negative inside; thus giving the same expression for the spectral function of left and right moving holes.
Here we would like to note the non-commutativity of the limits $`k,\omega 0`$ and $`T0`$ in calculating the zero temperature dc response of the system. Due to the scaling form of the spectral functions taking the zero temperature limit first gives a result which is determined by the $`\stackrel{~}{k},\stackrel{~}{\omega }\mathrm{}`$ behavior of the scaling functions, while it is the $`\stackrel{~}{k},\stackrel{~}{\omega }0`$ behavior which is relevant in case $`k`$ or $`\omega `$ is set to zero from the outset.
Further analytic progress in calculating the spectral functions can be achieved in the following cases:
#### 1 The case $`v_c=v_s`$
Since there is no symmetry between the charge and spin sectors, we generally expect that the spin and charge velocities are different. This greatly complicates the explicit calculation of the Fourier transforms, as the model is not “Lorentz invariant” under transformations that involve both the spin and charge sectors. Such an invariance is restored if $`v_c=v_s=v`$. In this case the correlation functions have the same form as those of a related model of spinless electrons, for which the zero temperature spectral functions have been calculated by Luther and Peschel .
The major simplification that follows directly from the fact that the charge and spin velocities are equal is that the two-dimensional Fourier transform reduces to a product of two one-dimensional transforms - one for the right moving piece and one for the left moving piece. Specifically one finds, with $`\gamma _G=\gamma _c+\gamma _s`$, that
$`G^<(\stackrel{~}{k},\stackrel{~}{\omega };T)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2T}}\left({\displaystyle \frac{a}{\lambda _T}}\right)^{2\gamma _G}`$ (42)
$`\times `$ $`h_{\gamma _G+1}\left({\displaystyle \frac{\stackrel{~}{\omega }+\stackrel{~}{k}}{2}}\right)h_{\gamma _G}\left({\displaystyle \frac{\stackrel{~}{\omega }\stackrel{~}{k}}{2}}\right).`$ (43)
A similar analytic form can be obtained for the spin correlation function
$`𝒮(\stackrel{~}{k},\stackrel{~}{\omega };T)`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2v}}\left({\displaystyle \frac{a}{\lambda _T}}\right)^{2(\beta _𝒮1)}`$ (45)
$`\times `$ $`h_{\beta _𝒮}\left({\displaystyle \frac{\stackrel{~}{\omega }+\stackrel{~}{k}}{2}}\right)h_{\beta _𝒮}\left({\displaystyle \frac{\stackrel{~}{\omega }\stackrel{~}{k}}{2}}\right),`$ (46)
where $`\beta _𝒮=\beta _c+\beta _s`$.
#### 2 The spin-rotationally invariant case $`(K_s=1)`$
We already noted that when the system is invariant under spin rotations $`K_s=1`$ ($`\gamma _s=0`$). At this important special point there is no mixing between left and right moving spin excitations. As a result the expression for the hole spectral function is simplified, due to the appearance of the factor $`h_0`$ in the spin part of (35), and can be expressed as a single integral
$`G^<(\stackrel{~}{k},\stackrel{~}{\omega };T)={\displaystyle \frac{r^{1/2}}{4\pi ^3T}}\left({\displaystyle \frac{a}{\lambda _{T,c}}}\right)^{2\gamma _c}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑qh_{\frac{1}{2}}\left(\stackrel{~}{k}2rq\right)`$ (47)
$`\times h_{\gamma _c+\frac{1}{2}}\left[{\displaystyle \frac{\stackrel{~}{\omega }\stackrel{~}{k}}{2}}+(1+r)q\right]h_{\gamma _c}\left[{\displaystyle \frac{\stackrel{~}{\omega }\stackrel{~}{k}}{2}}(1r)q\right].`$ (48)
When $`K_c=1`$ as well, the integral in (47) is trivial and we obtain
$$G^<(\stackrel{~}{k},\stackrel{~}{\omega };T)=\frac{1}{2\pi ^2T}\frac{r^{1/2}}{|1r|}h_{\frac{1}{2}}\left(\frac{\stackrel{~}{\omega }\stackrel{~}{k}}{1r}\right)h_{\frac{1}{2}}\left(\frac{r\stackrel{~}{\omega }\stackrel{~}{k}}{r1}\right).$$
(50)
The case $`\gamma _s=\gamma _c=0`$ is unique in the sense that there is no mixing between left and right moving excitations. As a result there are severe kinematic constraints on $`G^<`$ that make it non-vanishing, in the zero temperature limit, only in a wedge in $`k\omega `$ plane defined by the lines $`\omega =v_sk`$ and $`\omega =v_ck`$ for $`k<0`$. While $`K_s=1`$ reflects a symmetry of the problem it is unlikely that $`K_c=1`$ ($`\gamma _c=0`$) is realized in any interacting system. However, if the effective interactions are not too strong then $`\gamma _c`$ may be small. For instance, for the Hubbard model, even in the $`U/t\mathrm{}`$ limit, $`\gamma _c=1/16`$. For such systems we expect (50) to be qualitatively correct with the exception of the behavior of $`G^<`$ outside the above mentioned wedge. In contrast to the result for $`\gamma _c=0`$ the $`T=0`$ support of $`G^<`$, for non-zero $`\gamma _c`$, extends up to the line $`\omega =v_ck`$ with $`k>0`$. Nevertheless, if the mixing between left and right moving charge excitations (i.e $`\gamma _c`$) is small the amplitude of $`G^<`$ outside the wedge is small and its gross features resemble the $`\gamma _c=0`$ result.
The customary way to present ARPES data is by plotting momentum distribution curves (MDCs) and energy distribution curves (EDCs). These curves are cuts in $`G^<(k,\omega )`$ of constant $`\omega `$ and constant $`k`$ respectively. In Fig. 1 we present MDCs at the Fermi energy ($`\omega =0`$) and EDCs at the Fermi wave-vector $`(k=0)`$ for a spin rotationally invariant Tomonaga-Luttinger liquid for various values of the parameter $`\gamma _c`$.
#### 3 The case $`v_s/v_c0`$
In the limit where one of the velocities is much smaller than the other the calculation becomes again more tractable. In many physical systems $`v_c>v_s`$ and we will consider this case (for example for the $`tJ`$ model away from half filling we have $`v_s/v_cJ/t<1`$). In the limit $`r=v_s/v_c0`$ the spin piece in Eq. (30) becomes $`q`$-independent and any singularities in $`G^<`$ disperse with the slow velocity $`v_s`$. The $`q`$ integral is then readily evaluated by expressing the factors in $`g_c`$ as Fourier transforms of their real-space counterparts. The result is
$`G^<(\stackrel{~}{k},\stackrel{~}{\omega };T)={\displaystyle \frac{a}{(2\pi )^2v_c}}\left({\displaystyle \frac{a}{\lambda _{T,c}}}\right)^{2\gamma _c\frac{1}{2}}\left({\displaystyle \frac{a}{\lambda _{T,s}}}\right)^{2\gamma _s\frac{1}{2}}`$ (51)
$`\times {\displaystyle _{\mathrm{}}^{\mathrm{}}}d\nu h_{2\gamma _c+\frac{1}{2}}(\stackrel{~}{\omega }\stackrel{~}{k}2\nu )h_{\gamma _s+\frac{1}{2}}(\stackrel{~}{k}+\nu )h_{\gamma _s}(\nu ).`$ (52)
A similar calculation gives for the spin spectral function
$`𝒮(\stackrel{~}{k},\stackrel{~}{\omega };T)={\displaystyle \frac{1}{(2\pi )^3v_c}}\left({\displaystyle \frac{a}{\lambda _{T,c}}}\right)^{2\beta _c1}\left({\displaystyle \frac{a}{\lambda _{T,s}}}\right)^{2\beta _s1}`$ (54)
$`\times {\displaystyle _{\mathrm{}}^{\mathrm{}}}d\nu h_{2\beta _c}(\stackrel{~}{\omega }\stackrel{~}{k}2\nu )h_{\beta _s}(\stackrel{~}{k}+\nu )h_{\beta _s}(\nu ).`$ (55)
At the spin-rotationally invariant point $`(K_s=1)`$ the integral in (51) can be performed to give a result for $`G^<`$ in the limit $`v_s/v_c0`$ but arbitrary $`\gamma _c`$
$$G^<(\stackrel{~}{k},\stackrel{~}{\omega };T)=\frac{r^{1/2}}{2\pi ^2T}\left(\frac{a}{\lambda _{T,c}}\right)^{2\gamma _c}h_{2\gamma _c+\frac{1}{2}}(\stackrel{~}{\omega }\stackrel{~}{k})h_{\frac{1}{2}}(\stackrel{~}{k}).$$
(56)
### D Integrals of the spectral functions
Integrals of the spectral functions, such as the density of states and the momentum occupation number, are of interest too. These simpler quantities give us a qualitative view of the spectrum, without too many complicated details.
#### 1 $`\rho ^<(\omega )`$ and $`𝒮_0(\omega )`$
The calculation of the density of states $`\rho ^<(\omega )`$ and its analogous quantity for the spin susceptibilities $`𝒮_0(\omega )`$ is straightforward. The results are
$`\rho ^<(\stackrel{~}{\omega };T)={\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{dt}{2\pi }}e^{i\omega t}\stackrel{~}{G}_\eta ^<(0,t)={\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑kG^<(k,\omega )`$ (57)
$`={\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \frac{1}{\sqrt{v_cv_s}}}\left({\displaystyle \frac{a}{\lambda _{T,c}}}\right)^{2\gamma _c}\left({\displaystyle \frac{a}{\lambda _{T,s}}}\right)^{2\gamma _s}h_{2\gamma _G+1}(\stackrel{~}{\omega }),`$ (58)
$`𝒮_0(\stackrel{~}{\omega };T)={\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{dt}{2\pi }}e^{i\omega t}\stackrel{~}{𝒮}(0,t)={\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑k𝒮(k,\omega )`$ (60)
$`={\displaystyle \frac{1}{8\pi ^3a}}{\displaystyle \frac{1}{\sqrt{v_cv_s}}}\left({\displaystyle \frac{a}{\lambda _{T,c}}}\right)^{2\beta _c\frac{1}{2}}\left({\displaystyle \frac{a}{\lambda _{T,s}}}\right)^{2\beta _s\frac{1}{2}}h_{2\beta _𝒮}(\stackrel{~}{\omega }).`$ (61)
#### 2 $`n(k)`$ and $`S(k)`$ for the case $`v_c=v_s`$
The evaluation of the momentum occupation number $`n(k)`$ and the spin structure factor $`S(k)`$ is complicated by the fact that they can not be expressed in terms of the functions $`h_\gamma (k)`$. Contrary to the scaling functions considered above the scaling functions of these quantities depend on the cut-off $`a^1`$. It is, however, possible to compute them in the case where the charge and spin velocities are equal.
$`n(\stackrel{~}{k};T)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑xe^{i\eta (k+k_F)x}\stackrel{~}{G}_\eta ^<(x,0)={\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\omega }{2\pi }}G^<(k,\omega )`$ (63)
$`=`$ $`{\displaystyle \frac{2^{\gamma _G+1}}{\pi }}\left({\displaystyle \frac{a}{\lambda _T}}\right)^{2\gamma _G}`$ (64)
$`\times `$ $`\mathrm{Im}\left\{{\displaystyle _0^{\mathrm{}}}𝑑x{\displaystyle \frac{e^{i\stackrel{~}{k}x}\mathrm{sinh}(xia/\lambda _T)}{[\mathrm{cosh}(2x)\mathrm{cosh}(2ia/\lambda _T)]^{\gamma _G+1}}}\right\},`$ (65)
where again $`\gamma _G=\gamma _c+\gamma _s`$. Since the integrand in (63) is imaginary in the interval $`[ia/\lambda _T,0]`$ of the imaginary axis, we can add the integral along this interval to the one already present in (63) without affecting $`n(k)`$.
$`n`$ $`(\stackrel{~}{k};T)={\displaystyle \frac{2^{\gamma _G1}}{\pi }}\left({\displaystyle \frac{a}{\lambda _T}}\right)^{2\gamma _G}`$ (68)
$`\times \mathrm{Im}\left\{{\displaystyle _{2ia/\lambda _T}^{\mathrm{}}}𝑑x{\displaystyle \frac{e^{\left(i\frac{\stackrel{~}{k}}{2}\frac{1}{2}\right)x+ia/\lambda _T}e^{\left(i\frac{\stackrel{~}{k}}{2}+\frac{1}{2}\right)xia/\lambda _T}}{[\mathrm{cosh}(x)\mathrm{cosh}(2ia/\lambda _T)]^{\gamma _G+1}}}\right\}.`$
Formally, for $`\gamma _G<0`$, Eq. (68) is related to the integral representation of the associated Legendre function (defined with a branch cut from $`\mathrm{}`$ to 1) $`Q_\nu ^\mu (z)`$, see Ref. B. Since $`n(k)`$ is regular in $`\gamma _G`$ we can use analytic continuation to obtain for all non-integer $`\gamma _G`$
$`n(\stackrel{~}{k};T)`$ $`=`$ $`{\displaystyle \frac{2^{\gamma _G1/2}}{\pi ^{3/2}}}\mathrm{\Gamma }(\gamma _G)\left({\displaystyle \frac{a}{\lambda _T}}\right)^{2\gamma _G}`$ (70)
$`\times `$ $`\mathrm{Re}\{e^{i\pi \gamma _G}[\mathrm{sinh}(2ia/\lambda _T)]^{(\gamma _G+1/2)}`$ (73)
$`\times [e^{ia/\lambda _T}Q_{i\stackrel{~}{k}/21}^{\gamma _G+1/2}\left[\mathrm{cos}(2a/\lambda _T)\right]`$
$`e^{ia/\lambda _T}Q_{i\stackrel{~}{k}/2}^{\gamma _G+1/2}\left[\mathrm{cos}(2a/\lambda _T)\right]]\}.`$
It can be shown that for $`\gamma _G1/2`$ and $`k(\lambda _Ta)^{1/2}`$ this expression reduces, in the limit $`a0`$, to
$`n(\stackrel{~}{k};T)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Gamma }(\gamma _G)}{2\pi ^{3/2}}}\mathrm{Re}\{\mathrm{\Gamma }(\gamma _G+1/2)e^{i\pi (\gamma _G+1/2)}`$ (74)
$`+`$ $`i\left({\displaystyle \frac{a}{\lambda _T}}\right)^{2\gamma _G}{\displaystyle \frac{\mathrm{\Gamma }(1/2\gamma _G)\mathrm{\Gamma }[(1+2\gamma _Gi\stackrel{~}{k})/2]}{\mathrm{\Gamma }[(12\gamma _Gi\stackrel{~}{k})/2]}}\}.`$ (75)
Using similar manipulations we obtain, for non-integer $`\beta _𝒮=\beta _c+\beta _s`$, the spin structure factor
$`S(\stackrel{~}{k};T)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑xe^{i(k+2k_F)x}\stackrel{~}{S}(x,0)={\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\omega }{2\pi }}S(k,\omega )`$ (77)
$`=`$ $`{\displaystyle \frac{2^{\beta _𝒮3/2}}{\pi ^{5/2}a}}\mathrm{\Gamma }(1\beta _𝒮)\left({\displaystyle \frac{a}{\lambda _T}}\right)^{2\beta _𝒮1}`$ (78)
$`\times `$ $`\mathrm{Re}\{e^{i\pi (1/2\beta _𝒮)}[\mathrm{sinh}(2ia/\lambda _T)]^{1/2\beta _𝒮}`$ (80)
$`\times Q_{i\stackrel{~}{k}/21/2}^{\beta _𝒮1/2}\left[\mathrm{cos}(2a/\lambda _T)\right]\},`$
which for $`a0`$, $`\beta _𝒮3/2`$ and $`k(\lambda _Ta)^{1/2}`$ tends to
$`S(\stackrel{~}{k};T)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Gamma }(1\beta _𝒮)}{4\pi ^{3/2}a}}\mathrm{Re}\{\mathrm{\Gamma }(\beta _𝒮1/2)e^{i\pi (1/2\beta _𝒮)}`$ (81)
$`+`$ $`\left({\displaystyle \frac{a}{\lambda _T}}\right)^{2\beta _𝒮1}{\displaystyle \frac{\mathrm{\Gamma }(1/2\beta _𝒮)\mathrm{\Gamma }(\beta _𝒮i\stackrel{~}{k}/2)}{\mathrm{\Gamma }(1\beta _𝒮i\stackrel{~}{k}/2)}}\}.`$ (82)
## III The Luther–Emery Liquid
### A The model
In the Tomonaga-Luttinger liquid both the charge and spin excitations are gapless. Spin and charge gaps may open up as a result of adding to the Tomonaga-Luttinger Hamiltonian (1) terms describing backward and Umklapp scattering respectively. In the following we will assume that the 1DEG is sufficiently incommensurate so that Umklapp scattering may be neglected. Consequently the charge sector remains gapless and continues to be described by a free bosonic theory \[the charge part of Eq. (10)\]. Including a backward scattering term $`g_1_\eta \psi _{\eta ,1}^{}\psi _{\eta ,1}^{}\psi _{\eta ,1}\psi _{\eta ,1}`$ results, after bosonization, in a spin Hamiltonian density of the sine-Gordon type
$$_s=\frac{v_s}{2}\left[K_s(_x\theta _s)^2+\frac{(_x\varphi _s)^2}{K_s}\right]+\frac{2g_1}{(2\pi a)^2}\mathrm{cos}(\sqrt{8\pi }\varphi _s).$$
(84)
The $`g_1`$ perturbation is relevant for $`K_s<1`$, in which case a spin gap is dynamically generated according to the scaling relation $`\mathrm{\Delta }_s(v_s/a)[g_1/2\pi ^2v_s]^{1/(22K_s)}`$ and the excitations are massive spin solitons. In the spin gapped phase the problem is most simply treated in terms of spin fermion fields and their mode decomposition $`C_\pm (\theta )`$ (obeying the usual fermionic anti-commutation relations) in momentum space
$`\mathrm{\Psi }_\eta `$ $`=`$ $`F_\eta \mathrm{exp}[i\sqrt{\pi /2}(\theta _s2\eta \varphi _s)]`$ (85)
$`=`$ $`\sqrt{{\displaystyle \frac{\mathrm{\Delta }_s}{4\pi v_s}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\theta e^{\eta \theta /2}\left[C_+(\theta )e^{i(x/\xi _s)\mathrm{sinh}\theta }\eta C_{}^{}(\theta )e^{i(x/\xi _s)\mathrm{sinh}\theta }\right],`$ (86)
where we introduce the spin correlation length $`\xi _s=v_s/\mathrm{\Delta }_s`$ and use the rapidity representation $`k=\mathrm{sinh}\theta /\xi _s`$.
For $`K_s=1/2`$, which is known as the free fermion or Luther-Emery point, the refermionized Hamiltonian is non-interacting and massive with a gap $`\mathrm{\Delta }_s=g_1/2\pi a`$
$`H_s`$ $`=`$ $`{\displaystyle 𝑑x\underset{\eta =\pm 1}{}\left[iv_s\eta \mathrm{\Psi }_\eta ^{}_x\mathrm{\Psi }_\eta +\mathrm{\Delta }_s\mathrm{\Psi }_\eta ^{}\mathrm{\Psi }_\eta \right]}`$ (87)
$`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\theta E_s(\theta )\left[C_+^{}(\theta )C_+(\theta )+C_{}^{}(\theta )C_{}(\theta )1\right],`$ (88)
where the spin excitation spectrum is
$$E_s(\theta )\mathrm{\Delta }_s\mathrm{cosh}\theta =\sqrt{\mathrm{\Delta }_s^2+(v_sk)^2}E_s(k).$$
(90)
In the following we will concentrate on computing the correlation functions at the Luther-Emery point. We will comment briefly on the effects of deviations from this point.
### B The spin part of the spectral functions
In contrast to the Tomonaga-Luttinger model the calculation of the spin part of most spectral functions for the Luther-Emery liquid is no longer trivial. The difficulty lies in the fact that generically the refermionized form for the spectral functions involves highly non-local operators. We start by evaluating the spin contribution to the transverse spin form factor, which fortunately has a simple representation in terms of the pseudo-fermions. We then consider the single hole spectral function that belongs to the wider class of functions which do not admit such a simple form. Interestingly, it is still possible to obtain an exact expression for this function too at the Luther-Emery point. This exact result extends and corrects earlier work of Voit and Wiegmann.
In computing the various spectral properties of the system we can distinguish between two temperature regions. At temperatures large compared to $`\mathrm{\Delta }_s`$ the spin gap can be ignored, and the results for the Tomonaga-Luttinger liquid apply. If the temperature is small compared to the spin gap we can evaluate the spin contribution to the correlation functions in the zero temperature limit while introducing errors of order $`\mathrm{exp}(\mathrm{\Delta }_s/T)`$. Henceforth we will be concerned with temperatures in the second region only.
#### 1 The spin part of $`𝒮(k,\omega )`$
The spin piece of the transverse spin correlation function has the following simple form in terms of the spin fermion fields
$$\stackrel{~}{𝒮}_s(x,t)=\mathrm{\Psi }_1^{}(x,t)\mathrm{\Psi }_1^{}(x,t)\mathrm{\Psi }_1(0,0)\mathrm{\Psi }_1(0,0).$$
(91)
Since the theory reduces, at the Luther-Emery point, to a theory of free massive fermions, the corresponding spectral function can be readily computed with the result, for $`T=0`$,
$$𝒮_s(k,\omega )=\frac{\omega ^24E_s^2(k/2)}{4v_s^2|q_1E_s(q_2)q_2E_s(q_1)|}\mathrm{\Theta }[\omega 2E_s(k/2)],$$
(92)
where the spin excitation spectrum $`E_s(k)`$ is given by Eq.(90) and $`q_{1,2}`$ are the solutions to the quadratic equation $`\omega +E_s(q)+E_s(kq)=0`$, that is
$$q_{1,2}=\frac{k}{2}\pm \frac{\omega }{2v_s}\sqrt{1+\frac{4\mathrm{\Delta }_s^2}{v_s^2k^2\omega ^2}}.$$
(93)
#### 2 The spin part of $`G^<(k,\omega )`$
The refermionized form of the single hole Green function (below we consider the case $`\eta =1`$ and $`\sigma =1`$)
$$\stackrel{~}{G}_s(x,t)=U_{\frac{1}{4}}^{}(x,t)\mathrm{\Psi }_1^{}(x,t)\mathrm{\Psi }_1(0,0)U_{\frac{1}{4}}(0,0),$$
(94)
involves the non-local vertex operators
$$U_\alpha (x)=\mathrm{exp}[i\sqrt{8\pi }\alpha \varphi _s(x)],$$
(95)
with
$$\varphi _s(x)=\sqrt{\frac{\pi }{2}}\underset{\eta =\pm 1}{}^x𝑑y\mathrm{\Psi }_\eta ^{}(y)\mathrm{\Psi }_\eta (y).$$
(96)
From Eqs. (85), (95) and (96) it is evident that $`\mathrm{\Psi }_\eta `$ and $`U_\alpha `$ create and destroy a single spin soliton or an integer number of soliton pairs, respectively. Therefor, the Green function consists of a coherent one spin soliton piece and an incoherent multi-soliton piece
$$G_s(k,\omega )=Z_s(k)\delta [\omega +E_s(k)]+G_s^{(multi)}(k,\omega ),$$
(97)
where the multi-soliton piece is proportional (at $`T=0`$) to $`\mathrm{\Theta }[\omega 3E_s(k/3)]`$. Deviations from the Luther-Emery point in the case $`K_s<1/2`$ will result in the formation of a spin soliton-antisoliton bound state, a “breather”, which can shift the threshold energy for multi-soliton excitations somewhat.
At the Luther-Emery point the form factors of the vertex operators, i.e. their matrix elements between the vacuum and various multi-soliton states, are known exactly. This fact enables us to obtain exact results for the different parts of the spectral function. The details of the calculation are presented in Appendix B. For the spectral weight of the coherent piece we find
$$Z_s(k)=\frac{8c^2}{\pi }\left(\frac{2\xi _s}{a}\right)^{\frac{3}{8}}\left[1\frac{v_sk}{E_s(k)}\right],$$
(98)
where $`c=0.101`$. We would like to note here that a simple scaling argument allows us to obtain the dependence of $`Z_s`$ on $`\xi _s`$ for arbitrary $`K_s<1`$. It follows from the observation that the sine-Gordon theory is asymptotically free and hence the dependence of $`G_s`$ on the short distance cutoff $`a`$ is unaffected by the opening of a spin gap. Since in the absence of a gap $`G_s`$ is proportional to $`a^{2\gamma _s1/2}`$ it is a matter of dimensional analysis to see that
$$Z_s(k)=(\xi _s/a)^{\frac{1}{2}2\gamma _s}f_s(k\xi _s),$$
(99)
where $`f_s`$ is an undetermined scaling function.
The incoherent piece of $`G_s^<(k,\omega )`$ consists of contributions from processes involving intermediate states containing $`N=2n+1`$ spin solitons, $`n=1,2,3,\mathrm{}`$. Each such contribution starts at an energy threshold $`NE_s(k/N)`$
$$G_s^{(multi)}(k,\omega )=\underset{N=3,5,\mathrm{}}{}G_s^{(N\mathrm{sol})}(k,\omega )\mathrm{\Theta }\left[\omega NE_s(\frac{k}{N})\right].$$
(100)
At the vicinity of the threshold i.e. for $`k\xi _s1`$ and $`|\omega +NE_s(k/N)|/\mathrm{\Delta }_s1`$ the behavior of $`G_s^{(N\mathrm{sol})}`$ is given by $`[\omega +NE_s(k/N)]^{n^2+n1}`$. In particular the 3 soliton part reads then
$$G_s^{(3\mathrm{sol})}(k,\omega )=\frac{8c^2}{\sqrt{3}\pi ^2}\left(\frac{2\xi _s}{a}\right)^{\frac{3}{8}}\left[\frac{\omega 3E_s(k/3)}{\mathrm{\Delta }_s^2}\right].$$
(101)
As discussed above, for large $`|\omega |`$ the incoherent piece should asymptotically approach the Tomonaga-Luttinger result $`\omega ^{2\gamma _s3/2}`$.
### C The single hole spectral function
Once the spin part is calculated there still remains the task of convolving it with the corresponding charge part in order to obtain an expression for the full spectral function. Analytically this is difficult, and like in the case of the Tomonaga-Luttinger liquid, progress can be made only in special cases. Below we carry out the convolution for the single hole spectral function. We consider temperatures well below the spin gap scale and correspondingly use the above derived zero temperature results for the spin part. For the gapless charge degrees of freedom we continue to use the finite temperature Tomonaga-Luttinger expressions.
The spectral function can be written as a sum of two contributions $`G^<=G_1+G_2`$ coming from the convolution of the charge part with the coherent (single soliton) and incoherent (multi-soliton) pieces of $`G_s^<`$. The frequency integral in $`G_1`$ is readily evaluated with the result
$`G_1`$ $`(k,\omega ;T)={\displaystyle \frac{c^2}{\pi ^3}}{\displaystyle \frac{\lambda _T^2}{v_c}}\left({\displaystyle \frac{a}{\lambda _T}}\right)^{2\gamma _c+\frac{1}{2}}\left({\displaystyle \frac{2\xi _s}{a}}\right)^{\frac{3}{8}}`$ (102)
$`\times `$ $`{\displaystyle 𝑑q\left[1\frac{v_s(kq)}{E_s(kq)}\right]h_{\gamma _c+\frac{1}{2}}\left[\frac{\omega +E_s(kq)+v_cq}{2\pi T}\right]}`$ (103)
$`\times `$ $`h_{\gamma _c}\left[{\displaystyle \frac{\omega +E_s(kq)v_cq}{2\pi T}}\right],`$ (104)
where here $`\lambda _T=v_c/\pi T`$.
In Fig. 2 we present representative contributions of the single spin soliton piece, $`G_1`$, to the MDCs and EDCs of a Luther-Emery liquid with various values of the charge exponent $`\gamma _c`$. Using some of the results derived below, we also indicate the asymptotic behavior of the three spin soliton contribution, $`G_2^{(3\mathrm{sol})}`$, to the EDCs in the vicinity of its zero temperature threshold $`\omega =3\mathrm{\Delta }_s`$.
We now restrict the discussion to the following special circumstances:
#### 1 The case $`v_s/v_c0`$
In this case the remaining integral in (102) is straightforward since the $`q`$ dependence of the spin part and $`E_s(kq)`$ disappears and we are left with
$`G_1(k,\omega ;T)`$ $`=`$ $`{\displaystyle \frac{4c^2}{\pi ^2}}{\displaystyle \frac{a}{v_c}}\left({\displaystyle \frac{a}{\lambda _T}}\right)^{2\gamma _c\frac{1}{2}}\left({\displaystyle \frac{2\xi _s}{a}}\right)^{\frac{3}{8}}`$ (105)
$`\times `$ $`\left[1{\displaystyle \frac{v_sk}{E_s(k)}}\right]h_{2\gamma _c+\frac{1}{2}}\left[{\displaystyle \frac{\omega +E_s(k)}{\pi T}}\right],`$ (106)
which in the limit of zero temperature reduces to
$`G_1(k,\omega ;0)`$ $`=`$ $`{\displaystyle \frac{8c^2}{\pi }}{\displaystyle \frac{1}{\mathrm{\Gamma }(2\gamma _c+1/2)}}\left({\displaystyle \frac{a}{v_c}}\right)^{2\gamma _c+\frac{1}{2}}\left({\displaystyle \frac{2\xi _s}{a}}\right)^{\frac{3}{8}}`$ (107)
$`\times `$ $`\left[\omega E_s(k)\right]^{2\gamma _c\frac{1}{2}}\mathrm{\Theta }[\omega E_s(k)].`$ (108)
For the 3-soliton contribution to $`G_2`$ we obtain, assuming $`k\xi _s1`$ and $`|\omega +3E_s(k/3)|\mathrm{\Delta }_s`$,
$`G_2^{(3\mathrm{sol})}(k,\omega ;T)={\displaystyle \frac{4c^2}{\sqrt{3}\pi ^4}}{\displaystyle \frac{1}{\mathrm{\Delta }_sT}}\left({\displaystyle \frac{a}{\lambda _T}}\right)^{2\gamma _c+\frac{1}{2}}\left({\displaystyle \frac{2\xi _s}{a}}\right)^{\frac{3}{8}}`$ (109)
$`\times {\displaystyle _{\omega +3E_s(k/3)}^{\mathrm{}}}d\nu h_{2\gamma _c+\frac{1}{2}}\left({\displaystyle \frac{\nu }{\pi T}}\right)\left[{\displaystyle \frac{\nu \omega 3E_s(k/3)}{\mathrm{\Delta }_s}}\right],`$ (110)
which at $`T=0`$ equals
$`G_2^{(3\mathrm{sol})}`$ $`(k,\omega ;0)={\displaystyle \frac{8c^2}{\sqrt{3}\pi ^2}}{\displaystyle \frac{1}{\mathrm{\Gamma }(2\gamma _c+5/2)}}\left({\displaystyle \frac{a}{v_c}}\right)^{2\gamma _c+\frac{1}{2}}\left({\displaystyle \frac{2\xi _s}{a}}\right)^{\frac{3}{8}}`$ (113)
$`\times {\displaystyle \frac{1}{\mathrm{\Delta }_s^2}}\left[\omega 3E_s(k/3)\right]^{2\gamma _c+\frac{3}{2}}\mathrm{\Theta }[\omega 3E_s(k/3)].`$
For this case the $`N=2n+1`$ soliton contribution to $`G_2`$ is proportional, in the vicinity of its threshold, to $`[\omega NE_s(k/N)]^{2\gamma _c+n^2+n1/2}\mathrm{\Theta }[\omega NE_s(k/N)]`$.
#### 2 the case $`\gamma _c=0`$
It is also possible to derive closed expressions for the spectral function when $`\gamma _c=0`$ but $`r=v_s/v_c`$ is arbitrary.
$`G_1(k,\omega `$ ; $`T)={\displaystyle \frac{4c^2}{\pi ^2}}{\displaystyle \frac{1}{1r}}{\displaystyle \frac{\sqrt{a\lambda _T}}{v_c}}({\displaystyle \frac{2\xi _s}{a}})^{\frac{3}{8}}`$ (115)
$`\times `$ $`h_{\frac{1}{2}}\left[{\displaystyle \frac{\omega rv_sk+\sqrt{(r\omega v_sk)^2+(1r^2)\mathrm{\Delta }_s^2}}{\pi (1r^2)T}}\right]`$ (116)
$`\times `$ $`{\displaystyle \frac{r\omega v_sk+\sqrt{(r\omega v_sk)^2+(1r^2)\mathrm{\Delta }_s^2}}{\sqrt{(r\omega v_sk)^2+(1r^2)\mathrm{\Delta }_s^2}}}.`$ (117)
Using the asymptotic form (A11) of $`h_\gamma (k)`$ one can obtain the zero temperature limit of $`G_1`$. When $`r<1`$ we find
$`G_1(k,\omega `$ ; $`0)={\displaystyle \frac{8c^2}{\pi ^{3/2}}}({\displaystyle \frac{1+r}{1r}}\left)^{\frac{1}{2}}\right({\displaystyle \frac{a}{v_c}}\left)^{\frac{1}{2}}\right({\displaystyle \frac{2\xi _s}{a}})^{\frac{3}{8}}`$ (118)
$`\times `$ $`{\displaystyle \frac{r\omega v_sk+\sqrt{(r\omega v_sk)^2+(1r^2)\mathrm{\Delta }_s^2}}{\sqrt{(r\omega v_sk)^2+(1r^2)\mathrm{\Delta }_s^2}}}`$ (119)
$`\times `$ $`{\displaystyle \frac{\mathrm{\Theta }[\omega E_s(k)]}{\sqrt{\omega +rv_sk\sqrt{(r\omega v_sk)^2+(1r^2)\mathrm{\Delta }_s^2}}}}.`$ (120)
The kinematics in the case $`r>1`$ is more involved. In particular we find that for sufficiently large $`r`$ it is possible to distribute the total momentum between the spin soliton and the gapless charge modes in a way that gives a contribution at frequencies smaller than $`E_s(k)`$. (In this case the effect exists only for $`k<0`$. This is a special property of the point $`\gamma _c=0`$ which precludes mixing of left and right moving charge excitations.)
$`G_1(k,\omega ;0)`$ $`=`$ $`{\displaystyle \frac{8c^2}{\pi ^{3/2}}}\left({\displaystyle \frac{r+1}{r1}}\right)^{\frac{1}{2}}\left({\displaystyle \frac{a}{v_c}}\right)^{\frac{1}{2}}\left({\displaystyle \frac{2\xi _s}{a}}\right)^{\frac{3}{8}}`$ (122)
$`\times `$ $`{\displaystyle \underset{\sigma }{}}{\displaystyle \frac{v_skr\omega \sigma \sqrt{(r\omega v_sk)^2+(1r^2)\mathrm{\Delta }_s^2}}{\sqrt{(r\omega v_sk)^2+(1r^2)\mathrm{\Delta }_s^2}\sqrt{\omega rv_sk+\sigma \sqrt{(r\omega v_sk)^2+(1r^2)\mathrm{\Delta }_s^2}}}}`$ (126)
$`\{\begin{array}{cc}\sigma =1\hfill & \mathrm{if}\omega <E_s(k)\hfill \\ \sigma =1,1\hfill & \mathrm{if}k\xi _s<\frac{1}{\sqrt{r^21}}\mathrm{and}E_s(k)<\omega <\frac{v_sk\sqrt{r^21}\mathrm{\Delta }_s}{r}\hfill \end{array}.`$
For $`k\xi _s1`$ and $`|\omega +3E_s(k/3)|\mathrm{\Delta }_s`$ the 3-soliton contribution to $`G_2`$ reads
$`G_2^{(3\mathrm{sol})}(k,\omega ;T)`$ $`=`$ $`{\displaystyle \frac{4c^2}{\sqrt{3}\pi ^3}}{\displaystyle \frac{\sqrt{a\lambda _T}}{\mathrm{\Delta }_s}}\left({\displaystyle \frac{2\xi _s}{a}}\right)^{\frac{3}{8}}{\displaystyle 𝑑qh_{\frac{1}{2}}(\lambda _Tq)}`$ (127)
$`\times `$ $`\left[{\displaystyle \frac{v_cq\omega 3E_s[(kq)/3]}{\mathrm{\Delta }_s}}\right]`$ (128)
$`\times `$ $`\mathrm{\Theta }\{v_cq\omega 3E_s[(kq)/3]\}.`$ (129)
When $`r<1`$ the zero temperature limit of the above is readily evaluated giving
$`G_2^{(3\mathrm{sol})}`$ $`(k,\omega ;0)={\displaystyle \frac{32c^2}{3^{3/2}\pi ^{5/2}}}\left({\displaystyle \frac{a}{\mathrm{\Delta }_sv_c}}\right)^{\frac{1}{2}}\left({\displaystyle \frac{2\xi _s}{a}}\right)^{\frac{3}{8}}`$ (131)
$`\times \left[{\displaystyle \frac{\omega +3E_s(k/3)}{\mathrm{\Delta }_s}}\right]^{\frac{3}{2}}\mathrm{\Theta }[\omega 3E_s(k/3)].`$
The zero temperature limit when $`r>1`$ is, once again, more complicated. However, it is possible to show that in the range of validity of Eq. (127), i.e. for $`k\xi _s1`$, and for $`1<r3`$ it coincides with the expression for the case $`r<1`$ Eq. (131). Deviations from this behavior may occur only if the velocity ratio is large and satisfies $`r3`$.
###### Acknowledgements.
It is a pleasure to thank S. Kivelson for many useful discussions and comments.
## A The function $`h_\gamma (k)`$
The function $`h_\gamma (k)`$ is real
$`h_\gamma (k)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑xe^{ikx}h_\gamma (x)`$ (A1)
$`=`$ $`\underset{a0}{lim}2\mathrm{R}\mathrm{e}\left\{{\displaystyle _0^{\mathrm{}}}𝑑x{\displaystyle \frac{e^{ikx}}{[i\mathrm{sinh}(x+ia)]^\gamma }}\right\}.`$ (A2)
Although the imaginary part of the integral diverges as $`a^{1\gamma }`$ for $`\gamma >1`$, its real part, and hence also $`h_\gamma (k)`$, are analytic for all values of $`\gamma `$. Substituting $`y=e^{2x}`$ we find
$$h_\gamma (k)=\underset{a0}{lim}\mathrm{Re}\left\{(2i)^\gamma _0^1𝑑yy^{\frac{\gamma ik}{2}1}\left(1e^{2ia}y\right)^\gamma \right\}.$$
(A3)
The integral is analytic in the limit $`a0`$ for $`\gamma <1`$. In this range the exponential factor in (A3) can be dropped and it reduces to the integral representation of the beta function . We can then use analytical continuation to obtain for all non-integer values of $`\gamma `$
$$h_\gamma (k)=\mathrm{Re}\left[(2i)^\gamma B(\frac{\gamma ik}{2},1\gamma )\right].$$
(A4)
For the integers, $`h_n(k)`$ can be calculated as follows. First, we obtain using standard residue technique
$`h_1(k)`$ $`=`$ $`2\pi f_+(\pi k),`$ (A5)
$`h_2(k)`$ $`=`$ $`2\pi kf_{}(\pi k),`$ (A6)
where $`f_\pm (k)=\left(e^k\pm 1\right)^1`$ are the fermionic and bosonic occupation functions. We then integrate (A1) by parts twice to find the recursion relation
$$h_{n+2}(k)=\frac{k^2+n^2}{n(n+1)}h_n(k),$$
(A7)
which implies
$`h_{2n+1}(k)`$ $`=`$ $`{\displaystyle \frac{2\pi }{\mathrm{\Gamma }(2n+1)}}{\displaystyle \underset{m=0}{\overset{n1}{}}}\left[(1+2m)^2+k^2\right]f_+(\pi k),`$ (A8)
$`h_{2n}(k)`$ $`=`$ $`{\displaystyle \frac{2\pi }{\mathrm{\Gamma }(2n)}}k{\displaystyle \underset{m=1}{\overset{n1}{}}}\left[(2m)^2+k^2\right]f_{}(\pi k).`$ (A9)
We also note that
$$h_0(k)=2\pi \delta (k).$$
(A10)
Finally, the asymptotic behavior of $`h_\gamma (k)`$ for large $`|k|`$ is easily evaluated with the result
$$h_\gamma (|k|\mathrm{})=\frac{2\pi }{\mathrm{\Gamma }(\gamma )}\mathrm{\Theta }(k)(k)^{\gamma 1}.$$
(A11)
Since the spectral functions have a scaling form in the variables $`\omega /T`$ and $`v_\alpha k/T`$ Eq. (A11) also determines the zero temperature limit of these functions.
## B Calculating $`G_s^<(k,\omega )`$ at the Luther-Emery point
Inserting the resolution of the identity into Eq. (94) one obtains
$`\stackrel{~}{G}_s^<(x,t)={\displaystyle \underset{N,N^{}}{}}0|U_{\frac{1}{4}}^{}(x,t)|N`$ $`N|\mathrm{\Psi }_1^{}(x,t)\mathrm{\Psi }_1(0,0)|N^{}`$ (B2)
$`\times N^{}|U_{\frac{1}{4}}(0,0)|0.`$
The matrix elements of the vertex operators appearing in (B2) are known as the form factors of these operators. They have been derived, at the Luther-Emery point (see also Ref. B), by a variety of ways. They were first obtained by Schroer and Truong who normal ordered the vertex operator with respect to the spin fermions. Smirnov derived them for $`\alpha =1/2`$ using bootstrap axioms. Most recently they were calculated using monodromy relations by Bernard and LeClair . Since the fields $`U_\alpha (0,0)`$ are neutral with respect to the topological $`U(1)`$ charge of the solitons $`(\pm )`$ the form factors are non-vanishing only for $`U(1)`$ neutral states
$`0|U_\alpha `$ $`(x,t)C_+^{}(\theta _{2n})\mathrm{}C_+^{}(\theta _{n+1})C_{}^{}(\theta _n)\mathrm{}C_{}^{}(\theta _1)|0=V_\alpha (1)^{n(n1)/2}\left({\displaystyle \frac{\mathrm{sin}\pi \alpha }{2\pi i}}\right)^n`$ (B3)
$`\times `$ $`\mathrm{exp}\left[i{\displaystyle \underset{k=1}{\overset{2n}{}}}\left({\displaystyle \frac{x}{\xi _s}}\mathrm{sinh}\theta _k\mathrm{\Delta }_st\mathrm{cosh}\theta _k\right)+\alpha {\displaystyle \underset{k=1}{\overset{n}{}}}(\theta _{n+k}\theta _k)\right]{\displaystyle \frac{_{1k<jn}\mathrm{sinh}\left(\frac{\theta _k\theta _j}{2}\right)\mathrm{sinh}\left(\frac{\theta _{n+k}\theta _{n+j}}{2}\right)}{_{1k,jn}\mathrm{cosh}\left(\frac{\theta _{n+k}\theta _j}{2}\right)}},`$ (B4)
where $`V_\alpha `$, the vacuum expectation value of the vertex operators, is given by
$$V_\alpha 0|U_\alpha (0,0)|0=c(\alpha )\left(\frac{2\xi _s}{a}\right)^{\alpha ^2},$$
(B5)
with
$$c(\alpha )=\mathrm{exp}\left\{_0^{\mathrm{}}\frac{dt}{t}\left[\frac{\mathrm{sinh}^2(\alpha t)}{\mathrm{sinh}^2t}\alpha ^2e^{2t}\right]\right\}.$$
(B6)
Furthermore, since $`\mathrm{\Psi }_\eta `$ creates and destroys a single soliton the state $`|N^{}`$ can differ from $`|N`$ by $`0,\pm 2`$ solitons only. It is also easy to check that $`G_s^<(k,\omega )`$ is real. Using Eqs. (85) and (B3) one obtains that each term in the Fourier transform of Eq. (B2) is proportional to $`i^{(N+N^{})/2}`$ times a real expression. Thus only terms with $`(N+N^{})/2`$ an even integer should be considered. Combining these two observations we conclude that the contribution to $`G_s^<(k,\omega )`$ comes solely from the terms $`N=N^{}`$ (the amplitudes for the cases $`N=N^{}\pm 2`$ are finite but cancel each other).
The coherent piece of the spectral function is due to the terms $`N=0`$ and $`N=2`$. The evaluation of the first is straightforward with the result
$`\pi V^2\left[1{\displaystyle \frac{v_sk}{E_s(k)}}\right]\delta [\omega +E_s(k)],`$
where $`V=V_{\frac{1}{4}}=c(\frac{1}{4})(2\xi _s/a)^{1/16}`$. The cutoff dependence of the above result is in conflict with the general scaling argument that was given following Eq. (98). It represents corrections to scaling coming from irrelevant operators. The leading scaling behavior is recovered by considering the contribution of the $`N=2`$ term to the coherent piece
$`{\displaystyle \frac{V^2}{32\pi ^3\xi _s}}\left[{\displaystyle 𝑑\beta \frac{e^{\frac{3}{4}\beta }}{\mathrm{cosh}\frac{\beta }{2}}}\right]^2{\displaystyle 𝑑\theta e^{\theta +i[(x/\xi _s)\mathrm{sinh}\theta \mathrm{\Delta }_st\mathrm{cosh}\theta ]}}.`$
The $`\beta `$ integral is divergent and a lower cutoff $`\mathrm{ln}(2\xi _s/a)`$ (corresponding to a cutoff $`k=1/a`$) should be introduced. We then find for the leading contribution to the coherent piece
$`\stackrel{~}{G}_s^{(coher)}(x,t)=`$ $`{\displaystyle \frac{4c^2}{i\pi ^3\xi _s}}\left({\displaystyle \frac{2\xi _s}{a}}\right)^{\frac{3}{8}}{\displaystyle \frac{xv_st}{\sqrt{x^2(v_st)^2+iϵt}}}`$ (B8)
$`\times K_1\left[{\displaystyle \frac{\sqrt{x^2(v_st)^2+iϵt}}{\xi _s}}\right],`$
where $`K_1(x)`$ is a modified Bessel function. Fourier transforming it one obtains $`Z_s(k)\delta [\omega +E_s(k)]`$ with the spectral weight $`Z_s(k)`$ given by Eq. (98).
The $`2n+1`$ soliton contribution to the incoherent piece of the spectral function comes from the terms $`N=2n`$ and $`N=2n+2`$. As is the case with the coherent piece the former contains corrections to scaling while the latter is responsible for the leading behavior which is proportional to
$`V^2{\displaystyle 𝑑\theta _1\mathrm{}𝑑\theta _{2n+1}}`$ $`\left[{\displaystyle 𝑑\beta e^{\frac{3}{4}\beta }\frac{_{n+2j2n+1}\mathrm{sinh}\left(\frac{\theta _j\beta }{2}\right)}{_{1jn+1}\mathrm{cosh}\left(\frac{\theta _j\beta }{2}\right)}}\right]^2\delta \left(k\xi _s{\displaystyle \underset{i=1}{\overset{2n+1}{}}}\mathrm{sinh}\theta _i\right)\delta \left({\displaystyle \frac{\omega }{\mathrm{\Delta }_s}}+{\displaystyle \underset{i=1}{\overset{2n+1}{}}}\mathrm{cosh}\theta _i\right)`$ (B9)
$`\times `$ $`\mathrm{exp}\left({\displaystyle \frac{1}{2}}{\displaystyle \underset{j=1}{\overset{n+1}{}}}\theta _j{\displaystyle \frac{1}{2}}{\displaystyle \underset{j=n+2}{\overset{2n+1}{}}}\theta _j\right){\displaystyle \frac{_{1k<jn+1}\mathrm{sinh}^2\left(\frac{\theta _k\theta _j}{2}\right)_{n+2k<j2n+1}\mathrm{sinh}^2\left(\frac{\theta _k\theta _j}{2}\right)}{_{n+2k2n+1,\mathrm{\hspace{0.17em}1}jn+1}\mathrm{cosh}^2\left(\frac{\theta _k\theta _j}{2}\right)}}.`$ (B10)
The exact evaluation of the above integrals is difficult but their behavior near the threshold $`(2n+1)E_s[k/(2n+1)]`$ may be extracted in the following way. Writing $`\omega =(2n+1)E_s[k/(2n+1)]\mathrm{\Delta }\omega `$ one finds that for $`k\xi _s1`$ and $`\mathrm{\Delta }\omega \mathrm{\Delta }_s`$ the second $`\delta `$-function restricts the integration region over $`\theta _i`$ to a small ball near the origin. It is then legitimate to expand the integrand to lowest order in these variables. The integral over $`\beta `$ gives (together with the factor $`V^2`$) the overall cutoff dependence $`(2\xi _s/a)^{3/8}`$ in accord with the general scaling argument. By changing variables to spherical coordinates it is easy to check that the remaining integral over $`\theta _i`$ is proportional to $`\mathrm{\Delta }\omega ^{n^2+n1}\mathrm{\Theta }(\mathrm{\Delta }\omega )`$. A detailed evaluation of the 3 solitons case results in Eq. (101).
|
warning/0005/quant-ph0005034.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
In contrast with relativistic quantum mechanics (RQM) non-relativistic quantum mechanics (NRQM), in the conventional formulation, lacks manifest covariance with respect to general, space-time coordinate transformations. The main reason for this is that basic quantities such as fields or Schr$`\ddot{\mathrm{o}}`$dinger probability amplitudes are taken there not to be vector, but to be projective representations of the 4-dimensional Galilei group $`G_4`$ or of the group of more general transformations. The reason can further be traced to the fact that Lagrangians of relevant systems recover invariance under those transformations, only when some time-derivative terms are excepted.
In our previous paper , we have shown, however, that NRQM in inertial systems can be formulated in a manifestly covariant manner, provided a 5-dimensional form $`G_5`$ is adopted for Galilei transformations . In the present report we generalize $`G_5`$ to those transformations $`G_5^{}`$ which connect inertial with non-inertial systems, and thereby finds a manifestly covariant formulation of NRQM, being valid for general coordinate systems, inertial or non-inertial. The formulation, in fact, runs quite parallel with that of RQM.
Interaction terms can be introduced in a way similar to RQM. In particular, those with external electromagnetic or gravitational fields are generated by certain replacements. The resulting equations for non-inertial cases contain terms corresponding to inertial forces, and this enables us to discuss, e.g., the equivalence principle solely within the framework of NRQM.
## 2 $`G_5`$ and Manifest Covariance
The basic ideas of are as follows. The Lagrangian $`=(1/2)m\dot{\stackrel{}{x}}^2`$ of a free particle of mass m is not invariant under $`G_4`$: $`\stackrel{}{x}\stackrel{}{x}^{}=R\stackrel{}{x}\stackrel{}{v}t,t^{}=t`$ with $`R^{tr}R=\mathrm{I}`$. However, $`\overline{}m\dot{s}`$ remains invariant, provided the new variable $`s`$ transforms under $`G_4`$ as $`ss^{}=s+f`$ with $`f(\stackrel{}{x},t)(R\stackrel{}{x})\stackrel{}{x}+(1/2)\stackrel{}{v}^2t.`$ Thus, instead of $`G_4`$ we employ $`G_5`$, a central extension of $`G_4`$, such that
$$\begin{array}{c}x^i=R^i{}_{j}{}^{}x_{}^{j}\frac{v^i}{u}x^4,x^4=x^4,\hfill \\ x^5=x^5\frac{v_i}{u}(R^i{}_{j}{}^{}x_{}^{j})+\frac{1}{2}\frac{\stackrel{}{v}^2}{u^2}x^4,\hfill \end{array}$$
(1)
where $`(x^1,x^2,x^3)\stackrel{}{x},x^4ut,x^5s/u`$ with $`[u]=[v]`$, or $`x^\mu \mathrm{\Lambda }^\mu {}_{\nu }{}^{}x_{}^{\nu }`$. Under (1) $`\eta _{\mu \nu }x^\mu x^\nu =\stackrel{}{x}^22ts`$ is invariant, with $`\eta _{\mu \nu }=\eta ^{\mu \nu }`$ such that $`\eta _{ij}=\delta _{ij}(i,j=1,2,3),g_{45}=g_{54}=1`$ and others $`=0`$. Note that $`\eta _{\mu \nu }x^\mu x^\nu `$ with $`x_\pm (x^4\pm x^5)/\sqrt{2}`$ leads to $`\stackrel{}{x}^2+\stackrel{}{x}_{}^2x_+^2`$, and $`\eta _{\mu \nu }p^\mu p^\nu =0`$ with $`p^\mu (\stackrel{}{p},mu,E/u)`$ to $`E=\stackrel{}{p}^2/2m`$.
Our basic assumption then is that NRQM for free systems be invariant under $`G_5`$. Thus for a scalar field $`\varphi (x)`$ a Klein-Gordon type equation
$$\eta ^{\mu \nu }_\mu _\nu \varphi (x)=0$$
(2)
with a subsidiary condition
$`(i\mathrm{}_5mu)\varphi (x)=0(2^{})`$
results, for $`\varphi (x)=\mathrm{exp}(ims/\mathrm{})\psi (\stackrel{}{x},x^4)`$, in
$$i\mathrm{}\frac{}{t}\psi (\stackrel{}{x},t)=\frac{\mathrm{}^2}{2m}\stackrel{}{}^2\psi (\stackrel{}{x},t).$$
(3)
Notice that $`\varphi ^{}(x^{})=\varphi (x)`$ leads immediately to $`\psi ^{}(\stackrel{}{x}^{},t^{})=\mathrm{exp}(imf/\mathrm{})\psi (\stackrel{}{x},t)`$ with $`\psi ^{}(\stackrel{}{x}^{},t^{})`$ similarly defined. Likewise, for a spinor field $`\chi (x)`$ we assume a Dirac-type equation
$$\gamma ^\mu _\mu \chi (x)=0$$
(4)
together with a subsidiary condition of the same form as (2). Here, the $`\gamma `$-matrices satisfying $`\gamma ^\mu \gamma ^\nu +\gamma ^\nu \gamma ^\mu =2\eta ^{\mu \nu }`$ have the 4$`\times `$ 4 irreducible representation. Under $`G_5`$ $`\chi (x)`$ transforms as $`\chi ^{}(x^{})=T\chi (x)`$, where $`T^1\gamma ^\mu T=\mathrm{\Lambda }^\mu {}_{\nu }{}^{}\gamma _{}^{\nu }`$. Substituting
$$\chi (x)=\mathrm{exp}\left(\frac{ims}{\mathrm{}}\right)\left(\begin{array}{c}\psi _1(\stackrel{}{x},x^4)\\ \psi _2(\stackrel{}{x},x^4)\end{array}\right)$$
(5)
into (4), we find that only the 2-component spinor $`\psi _1`$ is independent, and satisfies the equation of the same form as (3). The extension of the above arguments to the case of Bargmann-Wigner fields of higher spins is straightforward.
By using vector representations of $`G_5`$ such as exemplified in the above we can construct a manifestly covariant field theory. It should be noted here that when a field $`\psi (\stackrel{}{x},t)`$ satisfies a linear equation of the Schr$`\ddot{\mathrm{o}}`$dinger type such as (3), the probability amplitude $`\phi (\stackrel{}{x},t)`$ for a particle, resulting from the quantized $`\psi `$, does also satisfy the same equation. Thus, the same covariance arguments apply as well to $`\phi `$.
## 3 General Coordinate Transformations
Generalizing $`G_4`$ we consider space-time transformations $`G_4^{}`$ such as $`\stackrel{}{x}^{}=R(t)\stackrel{}{x}+\stackrel{}{A}(t),t^{}=t`$, where $`R`$ and a vector $`\stackrel{}{A}`$ are taken to be $`t`$-dependent. The corresponding transformation rule of $`s`$ can again be found by observing how $``$ transforms under $`G_4^{}`$. Thus, instead of (1) we now consider $`G_5^{}`$ such that
$$\begin{array}{c}x^i=R^i{}_{j}{}^{}x_{}^{j}+A^i,x^4=x^4,\hfill \\ x^5=x^5+\stackrel{ˇ}{\stackrel{~}{A}}_jx^j+\frac{1}{u}\stackrel{~}{A}_j\dot{\stackrel{~}{A}}{}_{}{}^{j}\frac{1}{2u^2}_0^{x^4}\dot{\stackrel{~}{A}}_j(\tau )\dot{\stackrel{~}{A}}{}_{}{}^{j}(\tau )d\tau ,\hfill \end{array}$$
(6)
where $`\stackrel{~}{A}_iR^j{}_{i}{}^{}A_{j}^{}`$. Obviously, the transformation converts, in general, an inertial system S<sub>0</sub> (coordinates : $`x^\mu `$) to a non-inertial system S (coordinates : $`x^\mu `$).
In S the metric tensor $`g^{\mu \nu }`$ is given by
$$\begin{array}{ccc}g^{ij}=\delta ^{ij},\hfill & g^{i4}=0,\hfill & g^{i5}=\frac{1}{u}\dot{R}^i{}_{j}{}^{}R_{k}^{}{}_{}{}^{j}x_{}^{k},\hfill \\ g^{44}=0,\hfill & g^{45}=1,\hfill & g^{55}=\frac{2}{u^2}R_{ji}\ddot{\stackrel{~}{A}}{}_{}{}^{i}x_{}^{j};\hfill \end{array}$$
(7)
and the affine connection $`\mathrm{\Gamma }_{\mu \nu }^\lambda `$ by
$$\begin{array}{cc}\mathrm{\Gamma }_{4j}^i=\frac{1}{u}R^i{}_{k}{}^{}\dot{R}_{j}^{}{}_{}{}^{k},\hfill & \mathrm{\Gamma }_{44}^i=\frac{1}{u^2}R^i{}_{j}{}^{}\ddot{R}_{k}^{}{}_{}{}^{j}x_{}^{k}\frac{1}{u^2}R^i{}_{j}{}^{}\ddot{\stackrel{~}{A}}{}_{}{}^{j},\hfill \\ \mathrm{\Gamma }_{4i}^5=\frac{1}{u^2}R_{ij}\ddot{\stackrel{~}{A}}{}_{}{}^{j},\hfill & \mathrm{\Gamma }_{44}^5=\frac{2}{u^3}\dot{R}_{ij}\ddot{\stackrel{~}{A}}{}_{}{}^{j}x_{}^{i}\frac{1}{u^3}R_{ij}\stackrel{\mathrm{}}{\stackrel{~}{A}}{}_{}{}^{j}x_{}^{i},\hfill \end{array}$$
(8)
$$\mathrm{others}=0.$$
## 4 Generally Covariant Field Equations
In order to generalize the field equation in S<sub>0</sub> to those in S we have only to follow the procedure employed in going from special to general relativity.
Thus, for the scalar field $`\varphi ^{}(x^{})`$ in S the equation (2) is changed to $`g^{\mu \nu }𝒟_\mu ^{}𝒟_\nu ^{}\varphi ^{}(x^{})=0`$, where $`𝒟_\mu ^{}`$ is the covariant derivative. Since, however, $`g^{\mu \nu }\mathrm{\Gamma }_{\mu \nu }^\lambda =0`$ from (7) and (8), the above equation reduces simply to
$$g^{\mu \nu }_\mu ^{}_\nu ^{}\varphi ^{}(x^{})=0.$$
(9)
On the other hand, the subsidiary condition takes the same form as (2) because of $`_5^{}=_5`$. Defining $`\psi ^{}(\stackrel{}{x}^{},t^{})`$ from $`\varphi ^{}(x^{})`$ in the same way as before we find from (9) and (7) that
$$i\mathrm{}\frac{}{t^{}}\psi ^{}(\stackrel{}{x}^{},t^{})=\left(\frac{\mathrm{}^2}{2m}\stackrel{}{}^2+_{inert}^{}\right)\psi ^{}(\stackrel{}{x}^{},t^{}),$$
(10)
$$_{inert}^{}mR_{jk}\ddot{\stackrel{~}{A}}{}_{}{}^{k}x_{}^{j}i\mathrm{}\dot{R}^{\mathrm{}}{}_{j}{}^{}R_{k}^{}{}_{}{}^{j}x_{}^{k}_{\mathrm{}}^{}.$$
(11)
For the spinor field $`\chi ^{}(x^{})`$, to be regarded as a scalar under (6), the equation (4) is generalized to
$$\gamma ^\mu (x^{})\left(_\mu ^{}+\mathrm{\Gamma }_\mu ^{}(x^{})\right)\chi ^{}(x^{})=0,$$
(12)
whereas the form of the subsidiary condition remains unchanged. The $`\gamma ^{}`$-matrices and the spin connection $`\mathrm{\Gamma }_\mu ^{}(x^{})`$ are given, in terms of the f$`\ddot{\mathrm{u}}`$nfbein $`h_a^\mu (x^{})`$, as
$`\gamma ^\mu (x^{})`$ $`=`$ $`h_a^\mu (x^{})\gamma ^a,`$
$`\mathrm{\Gamma }_\lambda ^{}(x^{})`$ $`=`$ $`{\displaystyle \frac{1}{8}}[\gamma ^a,\gamma ^b]g_{\mu \nu }^{}(x^{})h_a^\mu (x^{})𝒟_\lambda ^{}h_b^\nu (x^{}).`$ (13)
As usual, we have also $`g^{\mu \nu }(x^{})=h_a^\mu (x^{})h_b^\nu (x^{})\eta ^{ab}`$ and $`\gamma ^\mu \gamma ^\nu +\gamma ^\nu \gamma ^\mu =2g^{\mu \nu }(x^{})`$.
By construction (12) is invariant not only under (1), but also under local Galilei transformations $`\chi ^{}(x^{})\stackrel{~}{T}(x^{})\chi ^{}(x^{})`$ , where $`\stackrel{~}{T}(x^{})`$ corresponds to $`x^\mu `$-dependent $`\mathrm{\Lambda }^a_b`$ with $`\mu 5`$ . Incidentally, the latter is a kind of non-Abelian gauge transformations, and $`\mathrm{\Gamma }_\mu ^{}`$ a gauge dependent quantity.
When explicitly rewritten in S, (12) takes the following forms. We consider two choices of $`h_a^\mu `$ (or of the gauge). 1) We take $`h_a^\mu (x^{})=x^\mu /x^a`$. In this case $`𝒟_\lambda ^{}h_b^\nu (x^{})=0`$, hence $`\mathrm{\Gamma }_\mu ^{}(x^{})=0`$. Then, for the function $`\psi _1^{}(\stackrel{}{x}^{},t^{})`$, defined in a way similar to (5), there holds the equation, having the same form as (10) with (11). 2) Taking a $`t^{}`$-dependent orthogonal matrix $`\stackrel{~}{R}`$, we adopt $`h_j^i=\stackrel{~}{R}_j{}_{}{}^{k}h_{k}^{i},h_i^5=\stackrel{~}{R}_i{}_{}{}^{j}h_{j}^{5}`$, and $`h_a^\mu =h_a^\mu `$ for others. In this case, on the right hand side of the equation for $`\psi _1^{}(\stackrel{}{x}^{},t^{})`$ an additional term $`_{\mathrm{spin}}^{}`$ appears , such as
$$_{\mathrm{spin}}^{}\frac{\mathrm{}}{4}\dot{\stackrel{~}{R}}{}_{}{}^{\mathrm{}}{}_{j}{}^{}\stackrel{~}{R}_{}^{kj}ϵ_{k\mathrm{}m}\sigma _m$$
(14)
with $`\sigma _m`$’s being Pauli matrices.
For the case of Bargmann-Wigner fields the results are basically the same as above. Needless to say, (11) and (14) provide the general expressions for inertial force potentials.
## 5 Further Remarks
a) Interactions can be introduced to (2) or (4) by adding $`G_5`$-symmetry conserving or violating terms. In particular, the usual form of interactions with an external electromagnetic field $`A^\mu (x)`$ is reproduced by making the $`G_5`$\- and gauge invariant replacement:
$$_\mu D_\mu _\mu (ie/c\mathrm{})A_\mu (x),$$
(15)
where $`A^\mu (x)=(\stackrel{}{A},0,\frac{c}{u}A)`$ with $`\stackrel{}{A}`$ and $`A`$ being vector and scalar potentials, respectively, and $`_5A^\mu (x)=0`$ is assumed. In this way we find, e.g., the gyromagnetic ratio $`g=2`$ for the spin 1/2 case .
Similarly, the interaction with a given (gravitational) Newton potential $`\mathrm{\Phi }(\stackrel{}{x})`$ is introduced into (2) or (4) by making a series of rewriting and replacements : $`\eta ^{\mu \nu }=h_a^\mu h_b^\nu \eta ^{ab},\gamma ^\mu =h_a^\mu \gamma ^a`$;
$$h_4^5h_4^5\frac{1}{u^2}\mathrm{\Phi }(\stackrel{}{x});$$
(16)
and then $`h_a^\mu =\delta _a^\mu `$. The potential term thus obtained is $`m\mathrm{\Phi }(\stackrel{}{x})`$.
b) In an inertial system S<sub>0</sub> we introduce by (16) the interaction with $`\mathrm{\Phi }(\stackrel{}{x})=\stackrel{}{g}\stackrel{}{x}`$ ($`\stackrel{}{g}`$: gravitational acceleration), and move afterwards to the system S by (6) with $`R(t)=\mathrm{I}`$, and $`\stackrel{}{A}(t)=\frac{1}{2}\stackrel{}{a}t^2`$ ($`\stackrel{}{a}`$: constant vector). The resulting equation for $`\psi ^{}(\stackrel{}{x}^{},t^{})`$ or $`\psi _1^{}(\stackrel{}{x}^{},t^{})`$ then obtains the potential term $`m(\stackrel{}{g}\stackrel{}{a})\stackrel{}{x}^{}`$. Thus, in S with $`\stackrel{}{a}=\stackrel{}{g}`$ the gravitational effect completely disappears. The result is basically the same for the case of general $`\mathrm{\Phi }`$’s. This implies that NRQM is compatible with Einstein’s equivalence principle.
c) Results from our formalism agree with the non-relativistic, Pauli-, Newton-,$`\mathrm{}`$ approximations to the corresponding cases of RQM. In fact, our calculations run almost parallel with those of RQM; approximate equations in the latter hold true exactly in the former.
d) Our work shows that the method proposed by Marmo et al. can be extended to those cases in which Lagrangians do not remain invariant after transformations.
e) In our formalism some of the problems which have so far been discussed by invoking classical mechanics become discussible within the framework of NRQM.
f) NRQM is usually regarded as a theory subordinate to RQM, for the former is a special case of, and hence obtainable by approximation from, the latter. In view of the present results, however, we should say that NRQM is an independent theory, being comparable to RQM.
|
warning/0005/astro-ph0005370.html
|
ar5iv
|
text
|
# H𝐼 Observations of the starburst galaxy NGC 2146
## 1 Introduction
NGC 2146 is a peculiar spiral galaxy as seen from both the optical image and the H I line profile. Although it is classified as Sab by de Vaucouleurs et al. (dev76 (1976)), it shows a broad range of peculiarities. Measured by its far infrared flux, it is one of the 12 brightest galaxies in the sky and lies at a distance of 12.2 Mpc ($`H_0`$=75 km s<sup>-1</sup> Mpc<sup>-1</sup>, and 1 corresponds to 3.5 kpc). In optical images there are two well-defined arms which mark the principal plane of rotation. Superimposed across part of the nucleus is an absorption band having “the form of a hand, with four talon-like fingers” (Pease pease (1920)), which de Vaucouleurs (dev50 (1950)) interpreted as being a third arm inclined to the plane of rotation. Further optical studies by Benvenuti et al. (benvenuti (1975)), however, suggest that a simple spiral model is not adequate.
In 1976, Fisher & Tully mapped the region around this galaxy in the 21cm using the NRAO 91-m telescope with a resolution of $`11^{}.3\times 10^{}.2`$ in the N-S and E-W directions respectively and discovered an extensive “envelope” of neutral hydrogen around it which extends up to six Holmberg radii (100 Kpc). They suggest that the abnormalities seen both optically and in the neutral hydrogen profile of the main disk are probably related to the large H I extensions observed, and believe that they might be the result of
a) tidal interaction,
b) explosion/ejection, or
c) galaxy formation.
They rule out any form of interaction between NGC 2146 and NGC 2146A, an Sc typed galaxy with no evident optical abnormalities, which lies about 20 arcseconds away and is redshifted 595 km s<sup>-1</sup> with respect to NGC 2146. However, their observations mainly aimed to trace the total extent of the H I cloud, and their resolution was not good enough to allow them to draw more definite conclusions on the causes of these abnormalities. Also, the appearance of the H I envelope as a large gaseous halo around the main galaxy left room for speculations as to how large galactic halos really are, and what the impact of this might be on the QSO absorption line system statistics.
NGC 2146 contains a strong radio source, $``$ 3 kpc in size, within its nuclear region, identified with 4C 78.06 (Caswell & Wills caswell (1967)). Kronberg & Biermann (kronberg (1981)) used the NRAO interferometer and the VLA to map the radio structure of the nuclear region. They found that the radio center lies in the optically obscured dust lane, but it shows no evidence of a double nucleus. The radio continuum map agrees very well with the CO intensity map, and unlike the optical image, it shows a remarkable degree of symmetry (Kronberg & Biermann kronberg (1981); Jackson & Ho jackson (1988)). The velocity curves measured in various lines in the optical and infrared are quite regular after allowance for extinction effects due to the dust lane (Prada et al. prada (1994)). Kronberg & Biermann (kronberg (1981)) suggested that a strong burst of star formation is responsible for the strong radio and infrared emission. A <sup>12</sup>CO, <sup>13</sup>CO and CS study undertaken by Xie et al. (xie (1994)) places the average temperature in the nuclear region about 55 K, and the average density at 2 $`\times `$ 10<sup>4</sup> cm<sup>-3</sup>. Further evidence of a burst of start formation comes from X-ray observations carried out by Armus et al. (armus (1995)) and Della Ceca et al. (dellaceca (1999)) which reveal a galactic-scale outflow of gas driven by an intensive star bursting activity, referred to as a starburst-driven superwind. Furthermore, this activity can produce long-lived bending instabilities as suggested by N-body simulations carried out by Griv & Chiueh (griv (1998)) to explain the snake-shaped radio structure observed by Zhao et al. (zhao (1996)) at an angular resolution of 2<sup>′′</sup> using the VLA.
Observations probing the molecular content in CO and H<sub>2</sub> as well as the ionized gas content (H II regions) were made by Young et al. (young (1988)). They found a very large concentration of gas in the nucleus, confirming Condon et al. (condon (1982)) earlier conclusion that this galaxy has a high star-formation rate and then, derived a mass-to-light ratio characteristic of very young stellar systems. All the above led these authors to suggest that NGC 2146 has recently undergone a collision with some other galaxy. The existance of an extended arc of H II regions encircling the central bright region, which exhibit velocities which are 130$``$200 km s<sup>-1</sup> higher than those expected if they are rotating in the plane of the galaxy (Young et al. young (1988)), might also be interpreted as evidence of a collision.
After undertaking an optical and infrared study of the galaxy, Hutchings et al. (hutchings (1990)) found no sign of an active nucleus but did find many signs of a significant population of hot young stars in the central regions of the system. They concluded that NGC 2146 is a merging system, now in its final stages. The dominant galaxy is seen close to edge on, and the small companion has been stripped, leaving no sign of its nucleus. They also note that such a scenario can also be supported by numerical simulations (Barnes barnes (1990)).
In order to better understand the nature of this system and decide amongst the various scenaria which have been suggested, we obtained higher resolution 21cm maps of the H I distribution around NGC 2146 using the VLA, and combined these with 21cm observations of the NRAO 91-m telescope to recover the emission on large angular scales, which the interferometric observations alone are incapable of sensing.The nature of large gaseous halos is important in the interpretation of QSO absorption line spectra (c.f Rao & Briggs rao (1993)), where the large cross sections implied by the Fisher & Tully observations (ft76 (1976)) would cause this kind of galaxies to intervene frequently by chance if they are common in the galaxy population.
## 2 Observations and Reduction
The observations presented here consist of two parts. Interferometry observations, carried out with the Very Large Array, and single dish observations, carried out with the 300ft NRAO antenna before it collapsed.
On 25,26 and 29 October 1984, the VLA was used in its most compact configuration to observe two fields, centered at $`\alpha _{1950}=6^h12^m00^s,\delta _{1950}=+78^o10^{}00^{\prime \prime }`$ and at $`\alpha _{1950}=6^h10^m00^s,\delta _{1950}=+78^o35^{}00^{\prime \prime }`$ respectively. The center fields were chosen in such a way that their combination produces a nearly uniform response over the area between and including the two pointing coordinates; this covers the main body of the galaxy and substantial fraction of the extended H I emission. A total of 5.5 hours per field with 25 antennas with baselines ranging from 36 m to a little more than 1 km were used. We observed only the right polarization in a band centered at a heliocentric velocity of 915 km s<sup>-1</sup>. After on-line Hanning smoothing, we recorded 31 independent channels ranging from 583 to 1226 km s<sup>-1</sup> with a FWHM velocity resolution of 20.75 km s<sup>-1</sup>. A broader bandwidth “continuum channel 0” to accompany the line data was calculated on-line from the average of 48 such channels centered at 915 km s<sup>-1</sup>. The VLA observational parameters are summarized in Table 1. For the flux calibration we used 3C286, adopting a flux density of 14.88 Jy according to the latest VLA measurments. The phase calibration source was 0836+710 for which a flux density of 4.12 Jy at 1.416 MHz was determined. Both calibrators were used in extracting a measure of the shape of the passband which we applied as a correction to the observations of NGC 2146.
To optimize the sensitivity to weak extended H I emission, the Fourier transformation was carried out using a natural weighting for the data, i.e. each visibility cell in the $`uv`$ plane has been weighted according to the integration time spent on it. This gives the best signal-to-noise ratio for detecting weak sources. Since $`uv`$ tracks tend to spend more time per unit area near the $`uv`$ origin, natural weighting emphasizes the data from short spacings, and produces a broad synthesized beam with an extended, low-level sideloabe. The natural-weighted maps provide sensitive information on the larger scale and are suitable for analyzing the structure of the extended H I cloud of NGC 2146. However, the poor synthesized beam shape means that care must be taken in the decovolution phase of the data processing. Incorporation of the single dish spectral is required for reliable mapping of the extended structure.
The construction of the continuum subtracted channel maps for each of the two fields is a crucial step in the reduction. Due to restrictions in the number of spectral channels that could be recorded in 1986 by the VLA, the correlator configuration was chosen so that it had H I emission in 29 of the 31 velocity channels. The on-line “channel 0”, being a mean of 48 channels, 29 of which are corrupted by the strong H I emission does not represent a true continuum channel. A true continuum map was obtained according to the following scheme, which is a weighted difference between spectral channel maps with gas and the “channel 0” map:
$$Cont.Map=\frac{1}{(4831)}(48(Ch.0)\underset{i=1}{\overset{31}{}}Ch.i)$$
In this way, the continuum map was formed from the frequency band outside the channels containing H I emission. The continuum was then subtracted from the dirty line channel maps to obtain the narrow band H I emission. This yielded a continuum field of total flux 1.83 Jy with the brightest source being the galaxy NGC 2146 (0.75 Jy).
During September 1987 the 300ft NRAO telescope was used to observe the area of NGC 2146. On each day a single drift scan was performed at a fixed declination, and spectra were recorded in short integrations. At the end of the observing session, a critically sampled map of the region surrounding NGC 2146 for merging with the interferometer data was constructed by gridding the data onto the same spatial and velocity grids as used for the VLA data cubes. For the joint deconvolution performed simultaneously on the VLA and 300ft data, a gaussian beam of HPBW 11.3 arcmin in the direction N-S and 10.2 arcmin in the direction E-W was adopted for the 300ft (Fisher & Tully ft75 (1975) ).
The continuum-free VLA maps were then simultaneously cleaned, mosaiced and combined together with the single dish maps with the AIPS maximum entropy based task UTESS to form single maps for each channel. At this stage, tests were performed using other alternative approaches of deconvolution like CLEAN algorithms alone, or combinations of CLEAN and maximum entropy based algorithms (implemented by VTESS in AIPS), yielding similar results in terms of map quality, but making apparent the fact that the single dish data were adding to the final maps substantial flux that the interferometry maps alone were missing, due to the lack of short spacing $`uv`$ coverage.
The final natural-weighted channel maps have an rms noise per channel of 1.0 to 1.25 mJy/beam, which is approximately the theoretical value (1.0 mJy/beam). The final synthesized beam is 49$`{}_{}{}^{\prime \prime }\times `$77<sup>′′</sup> in the North-South and East-West directions respectively.
## 3 The H I distribution
The channel maps of the H I flux of the area around the optical galaxy are shown in Figs. 1 and 2. The channels containing emission at large angular distance from the galaxy, together with a map of the radio continuum at 1420MHz, appear in Figs. 3 and 4. These maps reveal elongated streams of neutral hydrogen towards the north and the south of the main galaxy, extending out up to 6 Holmberg radii. This extensive distribution of the H I has a “tail” morphology suggestive of a tidal interaction of NGC 2146 and a companion, which has not been identified in optical images.
The southern stream is detected at higher signal to noise ratio than the northern stream. This occurs because the southern gas appears in only a few velocity channels, where the spatial structure is dominated by a single long arc that is nearly coincident in the three main contributing channels. There are extensions to this arcing stream, which, instead of following the general outflow of the H I, lie in smaller arcs of gas that appear to turn eastward and separate from the stream, falling back toward the central potential of the system. Further out from the tip of the southern stream appear some small H I clouds, which do not seem to be gravitationally bound to the whole system and may be escaping towards the south east.
The stream at the north is less prominent, due in part to being more heavily resolved, both spatially and in velocity, than the southern stream. Unlike the nearly constant velocity measured in the south, there is a prominent velocity gradient across the nothern extended emission from $``$975 km s<sup>-1</sup> at position angle $``$15 to $``$850 km s<sup>-1</sup> at position angle $``$30. Thus, the northern stream is symmetrically placed with respect to the southern one, but is not its mirror image. Instead of being concentrated in only three of our channels ($``$60 km s<sup>-1</sup>), the northern stream appears to be spread in a fan shaped outflow with line–of–sight velocity components ranging from $``$60 km/s redshifted to $``$65 km/s blueshifted as a function of position angle on the sky.
The observed H I distribution is consistent with that found by Fisher & Tully (ft76 (1976)), given the 10 beam (FWHM) of the 91m NRAO telescope. One significant difference is that the VLA detects absorption against the position of the nuclear continuum emission of the galaxy. This absorption feature persisted throughout all our reduction, and was consistently present in both the VLA maps of the North and the South part of NGC 2146, although the position of the galaxy was at a different point in the primary beam for each case. Also, the fact that no deep bowls were seen around the H I distribution suggests that the absorption is real and not an artifact of the interferometer data reduction. The spectrum of the absorption is shown in Fig. 5 with 5-$`\sigma `$ errorbars plotted for each spectral channel. This absorption feature would certainly be missed by a single dish telescope because of its large beam, in which the absorption is diluted and overwhelmed by the integrated emission. Fig. 5 also shows the spectrum of NGC 2146 obtained by Fisher & Tully (ft76 (1976)) by pointing the 91m telescope directly at the center of NGC 2146 (dot-dashed line), the spectrum of the whole area containing the H I cloud made by adding up all the flux seen by the NRAO 91-m telescope (solid line), and the spectrum of the total H I obtained after combining the VLA data with those of the NRAO 91-m telescope (solid line with Xs). Apparently, we are still missing about 20% of the total H I flux; maps produced from the VLA data alone were missing about 50% before being combined with the single dish data. This deconvolution procedure was successful in bringing up some fine details in the faint extended emission, especially in the northern stream of gas, where the signal to noise ratio is low due to being heavily resolved, and also in the low level emission to the east of the southern arm.
The absorption is seen against the radio center of the galaxy, which lies in the region of the optically obscuring dust, and is unresolved by our observations. It has a velocity width of $``$ 350 km s<sup>-1</sup> and an average optical depth of 0.03. Assuming a spin temperature of 50 K, the measured optical depth requires an H I column density of $``$10<sup>21</sup> atoms cm<sup>-2</sup>. The absorption is centered at the galaxy’s systemic velocity and seems to be due to H I clouds rotating together with the rest of the H I seen in emission, which happen to lie in front of the strong radio continuum source of NGC 2146, 4C 78.06. However, the large velocity gradient seen accross its 1 span, compared to the total width of the H I profile of NGC 2146 ($``$ 500 km s<sup>-1</sup>) implies that the H I absorbing layer is not very far from the nucleus of NGC 2146. If $`V_0`$ is the total H I profile velocity width of NGC 2146 ($``$ 500 km s<sup>-1</sup>), and $`\mathrm{\Delta }v`$ is the velocity width of the absorbing layer seen against the NGC 2146 continuum emission ($``$ 350 km s<sup>-1</sup>) then, assuming a circularly rotating system
$$\frac{\mathrm{\Delta }v}{2}=\frac{V_0}{2}sin\varphi _{abs}$$
where $`\varphi _{abs}`$ is shown in Fig. 6. From Fig. 6 we also see that
$$sin\varphi _{abs}=\frac{L/2}{R}$$
where $`L`$ is the length of the absorbing layer projected on the plane of the sky ($``$ 1 arcsec, or 3.5 kpc; this is an upper limit imposed by the spatial resolution of our observations), and $`R`$ is its distance from the nucleus. Combining these two equations we estimate an upper limit for the distance of the absorbing layer from the continuum background to be
$$R=\frac{LV_0}{2\mathrm{\Delta }v}=2.5\mathrm{kpc}$$
The integrated H I flux density map is shown in Figs. 7 (large scale) and 8 (small scale). We estimate the total mass for the cloud to be about $`6.2\times 10^9M_{}`$ ($`H_0`$=75 km s<sup>-1</sup> Mpc<sup>-1</sup>) of which $`1.6\times 10^9M_{}`$ come from the bright region around the galaxy itself, and $`4.6\times 10^9M_{}`$ from the extended component. Furthermore, the extended distribution is not symmetric with respect to the amount of gas in the south ($`3.1\times 10^9M_{}`$) and the north ($`1.5\times 10^9M_{}`$). For the above calculations a velocity component of 223 km s<sup>-1</sup> of our Sun towards NGC 2146’s direction has been assumed, in order to refer the redshift velocity to the center of mass of the Local Group.
Figs. 9 and 10 show the intensity weighted mean radial velocity field on both large and small scale. Points with negative flux due to absorption have been excluded from this calculation, and only emission brighter than 1.3 mJy has been taken into account. It is clear from the velocity field that the main galaxy has a differentially rotating disk with characteristic rotational speed of about 250 km s<sup>-1</sup>. The extended H I disk, however, appears severly lopsided as can immediately be seen from Fig. 10 since the galaxy is seen almost edge on.
The H I disk of the main galaxy has a well defined plane of rotation, with an inclination nearly perpendicular to the sky plane. The extended H I stream to the south, however, has very little velocity gradient along its entire length, with an average velocity that is coincident with the systemic velocity of the central galaxy. This situation occurs when the outlying stream and the galaxy centroid lie in the plane of the sky and implies that we observe the southern stream from a preferred viewing angle, in which it appears laid out in the probable plane of the interaction, which must have been very nearly perpendicular to the disk plane of NGC 2146. The northern stream, on the other hand, does show a significant velocity gradient across it, and the orientation of the stream cannot be inferred in a model–independent way.
In Fig. 11 the total integrated H I flux density of NGC 2146 is plotted superimposed on an optical image of the galaxy from the Digitized Sky Survey (DSS1). The morphology of the H I cloud does not show any preferential alignment with respect to the direction of NGC 2146A, and therefore shows no evidence for interaction between NGC 2146 and NGC 2146A. The faint arm extending southwards from the top NW corner of the optical image is in the general direction of the H I southern stream. However, the velocity gradient of H II regions of $``$100 km sec<sup>-1</sup> observed by Young et al. (young (1988)) across this feature is incompatible with the absence of a velocity gradient in the southern H I stream and its alignment with the plane of the sky. It seems thus that this arm is part of the main galaxy which was disturbed from an interaction but still remained under the strong gravitational influence of NGC 2146. The loop structure at the SE of the optical image of the galaxy shows higher velocities than those expected if it was rotating in the plane of the galaxy (Young et al., young (1988)) presenting further evidence for a tidal interaction. However, there is no apparent correspondence with any H I feature.
## 4 Discussion
The complex appearance of this system may represent the aftermath of an encounter between NGC 2146 and a slightly less massive but gas–rich galaxy, probably a late–type LSB spiral, with a slowly rising rotation curve indicating little or no bulge (de Blok et al. deblok (1996)). Numerical simulations by Wallin & Stuart (wallin (1992)) indicate that the outcome of an interaction depends crucially on the orientation of each galaxy’s rotational angular momentum vector relative to the plane of the interaction. If the rotational angular momentum vector lies in the plane of the interaction, the galaxy comes through with very little loss of mass. On the other hand, if the rotational angular momentum aligns with the orbital angular momentum, a large fraction of the galaxy is stripped away.
NGC 2146 may have interacted with a gas–rich “intruder” whose rotational angular momentum vector was oriented perpendicular to that of NGC 2146 and perpendicular to the plane of the interaction, which in this case is close to being coincident with the plane of the sky. Fig. 12 is a schematic diagram of the encounter. In this configuration, NGC 2146 has a rotational angular momentum vector lying in the interaction plane and is able to preserve its identity as a rotating disk galaxy, while the same encounter essentially destroys the intruder.
The numerical simulations of Wallin & Stuart (wallin (1992)) predict in such cases that the mass fraction transfered from the intruder to the main galaxy can be as large as 0.5, and the mass fraction lost by the intruder that eventually escapes from the whole system can reach up to 0.2. Wallin & Stuart used a model with restricted 3–body interactions (two large point masses binding massless test particles), and more realistic models with distributed mass are likely to suffer even greater destruction for the case of aligned orbital and rotational anglular momentum, as is indicated by some test cases run for us by J. Gerritsen (private communication) using his tree-code implementation (Gerritsen gerritsen (1997)).
The outcome of the tidal interactions is the development of gas streams on opposite sides of both galaxies. The velocity gradient of the gas in the northern stream is consistent with the sense of rotation of NGC 2146, which may indicate that this gas originated from the H I disk of NGC 2146. The gas at the southern stream, however, appears only around the systemic velocity of the system, and is likely to be the tidally dispersed remnant of the intruder. As the galaxies swirled around each other, H I clouds released from the intruder’s side opposite to NGC 2146 were boosted to escape velocity by the combination of the intruder’s translational and rotational velocities. In this scenario, the outermost clouds of the intruder experience the gravitational slingshot effect familiar in interplanetary spacecraft trajectories to reach the outer solar system. We next test the plausibility of such an encounter by exploring the timing arguments for the formation and lifetime of the southern plume near NGC 2146. For our purposes, we approximate the relative speeds of the masses using the vis viva equation to describe the relative speed $`V`$ of the centers of mass for the two masses, $`M_{NGC2146}`$ and $`M_{intr}`$, as a function of their separation $`R`$ and $`a`$, the semimajor axis of a possible elliptical orbit of the intruder about NGC 2146. Then,
$`V^2`$ $`=`$ $`G(M_{NGC2146}+M_{intr})\left[{\displaystyle \frac{2}{R}}{\displaystyle \frac{1}{a}}\right]`$
$`=`$ $`GM_{NGC2146}\left(1+{\displaystyle \frac{M_{intr}}{M_{N2146}}}\right)\left[{\displaystyle \frac{2}{R}}\right]`$
$``$ $`(350\mathrm{km}\mathrm{s}^1)^2\left(1+{\displaystyle \frac{M_{intr}}{M_{NGC2146}}}\right)\left[{\displaystyle \frac{15\mathrm{k}\mathrm{p}\mathrm{c}}{R}}\right]`$
where we have made the assumption that the intruder falls in from large distance so that $`a\mathrm{}`$ and that the mass of NGC 2146 can be approximated by adopting the rotational velocities of $`V_o`$250 km s<sup>-1</sup> that are measured in the outskirts of NGC 2146 at $`R_o=`$15 kpc to obtain $`GM_{NGC2146}R_oV_o^2`$.
For purposes of this illustration, we consider an HI-rich, “LSB” intruder, with $`M_{intr}=M_{NGC2146}/4`$ and $`R_{intr}R_o`$, having a gradually rising rotation curve that reaches $`V_{rot}125`$ km s<sup>-1</sup> at the outer edge of the galaxy’s disk. Under these assumptions, the relative speeds of the centers of mass for the two galaxies will be $`V277`$ km s<sup>-1</sup> for a grazing encounter where $`R=30`$ kpc. Four-fifths of the encounter speed ($``$222 km s<sup>-1</sup>) is carried by the center of mass of this intruder at a distance of 24 kpc from the system barycenter. By the time of closest approach, the intruder will have already become tidally distorted, and as it passes pericenter, simple tidal force arguments show that it will be unbound throughout its disk, provided it has a slowly rising rotation curve. However, in order to estimate the “launch speed” of debris in the outskirts of the intruder (trajectory A in Figure 12), we consider the effect of adding the $`V_{rot}`$ of the intruder to the speed of its center of mass.
Under these assumptions, gas clouds at the far side of the intruder from NGC 2146 are traveling at $``$346 km s<sup>-1</sup> at a distance of 39 kpc from the barycenter. We estimate the escape velocity from this distance by lumping $`M_{NGC2146}+M_{intr}`$ at the barycenter to find $`V_{esc}240`$ km s<sup>-1</sup>, which is roughly 100 km s<sup>-1</sup> less than the speed of the outer gas clouds. Once the process of dismantling the intruder begins, the mass lost lowers the binding energy, allowing the intruder to be shredded throughout.
Gas clouds on the inside of the intruder closest to the system barycenter, of course, find that whatever rotational component of velocity they still carry is counter to the orbit around NGC 2146. The dynamics in this region are clearly complicated, so that subtracting the 125 km s<sup>-1</sup> rotational speed from the 222 km s<sup>-1</sup> intruder center of mass speed to leave $``$100 km s<sup>-1</sup> forms only a rough guess. However, the implication is still that the inner gas is progressing in the same direction as the bulk of the material from the dismemberment of the intruder.
As a result, we picture the southern HI plume to be the result of tidal shredding of a high angular momentum LSB galaxy. The test simulation by Gerritsen (private communication) confirms that this could be accomplished without enhancing star formation in the intruder.
In order to make some statement about the time span since closest approach, we make the following simplifying assumptions: First, the high speed clouds from the outskirts of the intruder escape (trajectory A in Figure 12) from the system by overcoming the gravitation potential created by the the sum of the two original masses, concentrated at the system barycenter. Second, the remnants such as the HI spur running nearly east-west at declination $`78^{}12^{}`$ find their acceleration dominated by the mass of NGC 2146 alone. If we assume that the gas in this spur is located at the turn-around point of its trajectory, we can estimate the major axis of its orbit and deduce the time span required for it to return to closest approach to NGC 2146. The spur falls $`14^{}`$ or 49 kpc from the center of NGC 2146. Assuming these clouds were launched from a distance of $`25`$ kpc on the opposite side of NGC 2146, this trajectory (B in Figure 12) can be approximated by an ellipse of major axis $`2a=74`$ kpc. Using the rotational velocity $`V_o=`$250 km s<sup>-1</sup> for NGC 2146 at $`R_o=`$15 kpc, we deduce that the orbital period $`T(a)`$ would be $`(T/0.4\mathrm{Gyr})^2=(a/15\mathrm{k}\mathrm{p}\mathrm{c})^3`$ as a function of $`a`$. Thus, half of the period for $`a=37`$ kpc is $``$0.8 Gyr. This forms an estimate of the time elapsed since the closest approach.
The gas clouds ejected along trajectory A have reached $`50^{}`$ or 180 kpc during 0.8 Gyr, for an average speed of 225 km s<sup>-1</sup>. If the clouds were launched with the 346 km s<sup>-1</sup> estimated above, then they will have decelerated along their path, consistent with this estimate for average speed.
In this scenario, the central regions of the destroyed intruder are likely to lie somewhere along the south stream. Indeed, there is a concentration of gas, located at declination 7804, where the south stream has a kink (most clearly seen in channels with velocities of 853 and 832 km s<sup>-1</sup> in Fig. 4). The concentration has galaxy sized dimensions and a velocity gradient of 60 km s<sup>-1</sup> across its 30 kpc extent, contains $``$ 1.5$`\times `$10$`{}_{}{}^{8}M_{}^{}`$ in neutral hydrogen, and has an H I column density of $``$ 2$`\times `$10<sup>20</sup> atoms cm<sup>-2</sup>, which is high enough to trigger star formation that might be detectable in deep CCD images. All these make this object a very good candidate of being a remnant of the long sought companion of NGC 2146. Clearly more observations need to be undertaken to clarify the nature of this object, and test the validity of this scenario.
Another plausible scenario for this system is that it is at the final stage of merging and the small companion has been completely stripped off of its gas leaving no sign of its nucleus, as suggested by Hutchings et al. (hutchings (1990)) and Lisenfeld et al. (lisenfeld (1996)). In view of the support that numerical simulations can provide to such mergers (Barnes barnes (1990)) and if the above mentioned H I concentration at the south does not provide evidence for a companion nucleus around NGC 2146 this seems also a very attractive possibility.
The NGC 2146 system provides another example to add to those of Hibbard & van Gorkom (hibbard (1996)) where galaxy-galaxy interactions inject galactic gas into galactic halo regions, as well as ejecting gas with sufficient velocity that it can escape to the intergalactic medium. Depending on the mass ratio of the interacting galaxies, the relative inclinations of their disks and the impact parameter of the encounter the mass lost in the interaction can be up to 60% of the mass of the companion galaxy (Wallin & Stuart wallin (1992)). This gas contributes to the enrichment of the intergalactic medium in metals and if such events had been very common at some earlier epochs they may help towards finding the objects responsible for the metal absorption-line systems seen in abundance in QSO spectra.
Which of the above scenaria is correct remains yet undetermined. Numerical simulations based on the morphological and kinematical information presented may give us a clearer picture of how the system looked a billion years ago. Nevertheless it is striking that the final result of this collision is a morphologically classified spiral galaxy, although both the galaxy NGC 2146 and its surrounding H I cloud seem to have contained approximately equal amounts of neutral hydrogen. This system does provide evidence that mergers of two significant systems can be important events in the history of spiral as well as elliptical galaxies.
The encounter will affect the host galaxy for a long time into the future, since the outlying gas residue is likely to fall into the galaxy both from north and south meeting the rotational plane of the galaxy at a a significant angle of inclination to its internal orbital plane; this may provide a trickle of missaligned angular momentum for a long time as gas on different trajectories turns around and falls back. In this sense, the effect of this collisional debris will resemble Binney’s “Cosmic Drizzle” (binney (1990)) as a mechanism for creating long–lived warps in large, isolated galaxies.
## 5 Conclusions
By combining single dish observations and D - configuration VLA data we have produced sensitive, high quality images of the starburst galaxy NGC 2146. These images reveal elongated streams of neutral hydrogen extending out up to 6 Holmberg radii towards both the north and the south of the main galaxy.
The H I morphological and kinematical picture suggests that this galaxy underwent a strong interaction with a LSB companion, which was destroyed during the encounter. Although quite disturbed, NGC 2146 retained the morphological characteristics of a spiral galaxy. Analysis of the trajectory of the outlying gas suggests that the interaction was in its most violent phase about 0.8$`\times `$10<sup>9</sup> years ago. The infall of the outlying gas to NGC 2146 will continue for a similar timespan preserving a long–lived warp in the disk of this galaxy.
###### Acknowledgements.
We wish to acknowledge the National Astronomy and Ionosphere Center at the Arecibo Observatory for their hospitality during part of the data reduction. The National Radio Astronomy Observatory (NRAO) is operated by Associated Universities, Inc. under cooperative agreement with the National Science Foundation. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, Caltech, under contract with the National Aeronautics ans Space Administration. This work has been supported by National Science Foundation Grant AST 91-19930.
|
warning/0005/quant-ph0005125.html
|
ar5iv
|
text
|
# Optimal entanglement purifying via entanglement swapping
## Abstract
It is known that entanglement swapping can be used to realize entanglement purifying. By this way, two particles belong to different non-maximally entangled pairs can be projected probabilisticly to a maximally entangled state or to a less entangled state. In this report, we show, when the less entangled state is obtained, if a unitary transformation is introduced locally, then a maximally entangled state can be obtained probabilisticly from this less entangled state. The total successful probability of our scheme is equal to the entanglement of a single pair purification (if two original pairs are in the same non-maximally entangled states) or to the smaller entanglement of a single pair purification of these two pairs ( if two original pairs are not in the same non-maximally entangled states). The advantage of our scheme is no continuous indefinite iterative procedure is needed to achieve optimal purifying.
Entanglement is at the source of a number of pure quantum phenomena, such as the correlations violating Bell’s inequalities , quantum key distribution , quantum teleportation , Greanberger-Horne-Zeilinger \[GHZ\] correlations , and various other nonclassical interference phenomena . Polarization entangled photons have been used to demonstrate both dense coding and teleportation in the laboratory. Teleportation has also been realized using path-entangled photons and entangled electromagnetic field modes . In order to realize these schemes, the entanglement between distant particles should be set up. One of possible way is entanglement swapping , which has been demonstrated experimentally . Recently, Bose et al showed that entanglement swapping can be used to realize entanglement purifying. In their scheme, if an ensemble of two photon pair is given, in which all pairs are in the same non-maximally entangled states, then two photons belong to different photon pairs can be projected probabilisticly into a maximally entangled Bell state or a less entangled state by their scheme. If one continues this process indenfinitely, in the limit of an infinite sequence, the final ensemble generated will comprise of a certain fraction of Bell pairs and a certain fraction of completely disentangled pairs. This fraction of Bell pairs should be equal to twice the modulus square of the Schmidt coefficient of small magnitude of original pair. In this report, we show that if a unitary transformation is used followingly when a less entangled state is obtained by entanglement swapping, then, a maximally entangled Bell state can be obtained probabilisticly from this less entangled state. The maximum probability with which a Bell state can be obtained by our scheme is two the modulus square of the Schmidt coefficient of small magnitude, which means our scheme is optimal. One advantage of our scheme is no continuous indenfinite iterative procedure is needed. Furthermore, if two particle pairs are not the same type of entangled state, two particles belong to different pairs can also be projected into a maximally entangled state with certain probability by the same way. This probability is equal to the smaller entanglement of single pair purification of these two pairs, which also means our scheme is optimal.
Let pairs of particles (1, 2) and (3, 4) be in the following entangled states:
$$|\mathrm{\Phi }_{12}=\alpha |00_{12}+\beta |11_{12},$$
(1)
$$|\mathrm{\Phi }_{34}=\alpha |00_{34}+\beta |11_{34},$$
(2)
where, $`\left|\alpha \right|>\left|\beta \right|`$, and $`\left|\alpha \right|^2+\left|\beta \right|^2=1`$. Suppose that the particle pair (1, 2) and the particle 3 belong to Ailce and the particle 4 belongs to Bob. If a Bell state measurement on particles 2 and 3 is operated by Alice, then the particles 1 and 4 will be projected into one of the following states
$$\mathrm{\Phi }^\pm |_{23}\mathrm{\Phi }_{12}|\mathrm{\Phi }_{34}=\frac{\alpha ^2}{\sqrt{2}}|00_{14}\pm \frac{\beta ^2}{\sqrt{2}}|11_{14},$$
(3)
$$\mathrm{\Psi }^\pm |_{23}\mathrm{\Phi }_{12}|\mathrm{\Phi }_{34}=\alpha \beta [\frac{1}{\sqrt{2}}(|01_{14}\pm |10_{14})].$$
(4)
Where $`|\mathrm{\Phi }^\pm _{23}=\frac{1}{\sqrt{2}}(|00_{23}\pm |11_{23})`$ and $`|\mathrm{\Psi }^\pm _{23}=\frac{1}{\sqrt{2}}(|01_{23}\pm |10_{23})`$. The particles 1 and 4 will be projected into a less entangled state with probability $`\frac{\alpha ^4+\beta ^4}{2}`$. In order to get optimal entanglement purifying, a unitary transformation in Alice’ s side ( or in Bob’s side, for example, in Alice’s side) followes when a less entangled state is obtained. To carry out this evolution, an auxiliary qubit with the original state $`|0_a`$ is introduced by Alice. Under the basis {$`|0_1|0_a`$, $`|1_1|0_a`$, $`|0_1|1_a`$, $`|1_1|1_a\}`$, this unitary transformation can be written as
$$\left[\begin{array}{cccc}\frac{\beta ^2}{\alpha ^2}& 0& \sqrt{1\frac{\beta ^4}{\alpha ^4}}& 0\\ 0& 1& 0& 0\\ 0& 0& 0& 1\\ \sqrt{1\frac{\beta ^4}{\alpha ^4}}& 0& \frac{\beta ^2}{\alpha ^2}& 0\end{array}\right].$$
(5)
This transformation will transform Eq. (3) to the follow state
$$\beta ^2[\frac{1}{\sqrt{2}}(|00_{14}\pm |11_{14})]|0_a+\frac{\alpha ^2}{\sqrt{2}}\sqrt{1\frac{\beta ^4}{\alpha ^4}}|1_1|0_4|1_a.$$
(6)
A measurement to the auxiliary particle follows. If the result is $`|0_a`$, then the particles 1 and 4 will be projected a maximally Bell state with probability $`\beta ^4`$ . If the result of the measurement is $`|1_a`$, the particles 1 and 4 are completely disentangled.
The maximally probability with which a Bell state can be obtained by purifying a single entangled pair is $`2\beta ^2`$, which is equal to the maximum probability with which a Bell state can be obtained, so our scheme is optimal.
Next, we proceed to consider the case when particle pairs (1, 2) and (3, 4) are not in the same type of entangled state. Suppose the particles 1 and 2 are in the entangled state $`|\mathrm{\Phi }_{12}`$ and the particles 3 and 4 are in another entangled state $`|\mathrm{\Phi }_{34}`$ which are the following respectively
$$|\mathrm{\Phi }_{12}=\alpha |00_{12}+\beta |11_{12}$$
(7)
and
$$|\mathrm{\Phi }_{34}=a|00_{34}+b|11_{34}.$$
(8)
Where $`\left|a\right|>\left|b\right|,`$ $`\left|a\right|^2+\left|b\right|^2=1.`$ Suppose that the particles 1, 2 and 3 belong to Alice and the particle 4 belongs to Bob. If Alice make a Bell state measurement on particles 2 and 3, then the particles 1 and 4 will be projected into one of the follow states
$$\mathrm{\Phi }^\pm |_{23}\mathrm{\Phi }_{12}|\mathrm{\Phi }_{34}=\frac{\alpha a}{\sqrt{2}}|00_{14}\pm \frac{\beta b}{\sqrt{2}}|11_{14},$$
(9)
$$\mathrm{\Psi }^\pm |_{23}\mathrm{\Phi }_{12}|\mathrm{\Phi }_{34}=\frac{\alpha b}{\sqrt{2}}|01_{14}\pm \frac{\beta a}{\sqrt{2}}|10_{14}.$$
(10)
If Eq.(9) is obtained, a unitary transformation which is made on the particle 1 and an auxiliary qubit with the original state $`|0_a`$ is introduced by Alice (or on the particles 4 and auxiliary qubit $`|0_a`$ by Bob). Under the basis {$`|0_1|0_a`$, $`|1_1|0_a`$, $`|0_1|1_a`$, $`|1_1|1_a\}`$, this unitary transformation is
$$\left[\begin{array}{cccc}\frac{\beta b}{\alpha a}& 0& \sqrt{1\frac{\beta ^2b^2}{\alpha ^2a^2}}& 0\\ 0& 1& 0& 0\\ 0& 0& 0& 1\\ \sqrt{1\frac{\beta ^2b^2}{\alpha ^2a^2}}& 0& \frac{\beta b}{\alpha a}& 0\end{array}\right].$$
(11)
Under this transformation, Eq. (9) will be transformed into the state
$$\beta b[\frac{1}{\sqrt{2}}(|00_{14}\pm |11_{14})]|0_a+\frac{\alpha a}{\sqrt{2}}\sqrt{1\frac{\beta ^2b^2}{\alpha ^2a^2}}|1_1|0_4|1_a.$$
(12)
A measurement on the auxiliary particle follows. If the result is $`|0_a`$, the particles 1 and 4 will be projected into a maximally entangled state with probability $`\beta ^2b^2`$. If the result is $`|1_a`$, the particles 1 and 4 are completely disentangled.
If Eq. (10) is obtained, two different cases should be considered:
1: $`\left|\alpha b\right|>\left|a\beta \right|`$
In this case, the unitary transformation on the particles 1 and auxiliary qubit is
$$\left[\begin{array}{cccc}\frac{\beta a}{\alpha b}& 0& \sqrt{1\frac{\beta ^2a^2}{\alpha ^2b^2}}& 0\\ 0& 1& 0& 0\\ 0& 0& 0& 1\\ \sqrt{1\frac{\beta ^2a^2}{\alpha ^2b^2}}& 0& \frac{\beta a}{\alpha b}& 0\end{array}\right].$$
(13)
By the same procedure, the particles 1 and 4 will be projected into a maximally entangled state with probability $`a^2\beta ^2`$.
2: $`\left|\alpha b\right|<\left|a\beta \right|`$
In this case, the probability of obtaining a maximally entangled is $`\alpha ^2b^2`$. The unitary transformation is on the particles 1 and auxiliary qubit is
$$\left[\begin{array}{cccc}\frac{\alpha b}{a\beta }& 0& \sqrt{1\frac{\alpha ^2b^2}{a^2\beta ^2}}& 0\\ 0& 1& 0& 0\\ 0& 0& 0& 1\\ \sqrt{1\frac{\alpha ^2b^2}{a^2\beta ^2}}& 0& \frac{\alpha b}{a\beta }& 0\end{array}\right].$$
(14)
The maximally probability of obtained a maximally entangled state is $`2\beta ^2`$ or $`2b^2`$. The first means that entanglement of single pair purification of the $`|\mathrm{\Phi }_{12}`$ is less than that of the state $`|\mathrm{\Phi }_{34}`$. The second means that the entanglement of single pair purification of $`|\mathrm{\Phi }_{34}`$ is less than that of the state $`|\mathrm{\Phi }_{12}`$. This probability is equal to the smaller entanglement of single pair purification of these two pairs before entanglement swapping. Obviously, our scheme is optimal.
In conclusion . In the Ref. , an iterative procedure indefinitely is needed in order to achieve the optimal entanglement purifying. In our scheme, when a less entangled state is obtained during the entanglement purifying, if a unitary transformation is introduced locally, a maximally entangled state can be obtained with certain probability. The successful probability of our scheme is equal to the entanglement of a single pair purification if two original pairs are in the same non-maximally entangled states, or to the smaller entanglement of a single pair purification of these two pairs if they are not in the same non-maximally entangled states, which means our scheme is optimal. No continuous indefinite iterative procedure is needed, which makes our scheme implementable easily in practice.
This subject is supported by the National Natural Science Foundation of China (Grant No. 69907005).
|
warning/0005/cond-mat0005506.html
|
ar5iv
|
text
|
# Orbitally Degenerate Spin-1 Model for Insulating V2O3
\[
## Abstract
Motivated by recent neutron, X-ray absorption and resonant scattering experiments, we revisit the electronic structure of V<sub>2</sub>O<sub>3</sub>. We propose a model in which S=1 V<sup>3+</sup> ions are coupled in the vertical V-V pairs forming two-fold orbitally degenerate configurations with S=2. Ferro-orbital ordering of the V-V pairs gives a description which is consistent with all experiments in the antiferromagnetic insulating phase.
PACS Nos : 71.30.+h, 75.10.-b, 75.50.Ee
\]
Although the metal-insulator transition in V<sub>2</sub>O<sub>3</sub> has long been studied as a classic Mott-transition, the detailed electronic structure remains open. Recently new experimental techniques have been applied but these have not resolved the issue. Rather they have reopened the long standing controversy between an $`S=1`$ model without an orbital degeneracy and the $`S=1/2`$ orbitally degenerate model of Castellani et al.. In this Letter we propose a new model for the AF-ordered insulating (AFI) phase based on the molecular orbitals of the $`c`$-axis V-V pairs, which combines features of both existing models and which reconciles the apparently conflicting experiments supporting each.
The V-ions in the corundum structure of V<sub>2</sub>O<sub>3</sub> sit in a O-octahedron with a small trigonal distortion causing a small splitting in the non-bonding $`t_{2g}`$-shell between the $`a_{1g}`$-orbital oriented along the $`c`$-axis and doublet planar $`e_g`$-orbitals (see Fig. 1). In their early work, Castellani et al. proposed that one electron of the $`3d^2`$ V<sup>3+</sup>-ion entered a spin singlet covalent $`a_{1g}`$-bond in the V-V pair while the remaining electron was in the $`e_g`$-doublet. Orbital ordering of these $`e_g`$-doublets allowed them to explain the unusual magnetic structure of the AFI-phase with inequivalent $`n.n.`$ exchange constants in the $`ab`$ plane (2 antiferromagnetic, 1 ferromagnetic). Paolasini et al. interpreted their recent resonant x-ray experiments as a confirmation of this orbital ordering. On the other hand the polarized soft x-ray experiments by Park et al. showed a coexistence of both $`(e_ge_g)`$ and $`(e_ga_{1g})`$ configurations in roughly equal amounts and these led Ezhov et al. to argue for a $`S=1`$ model with a $`(e_ge_g)`$ configuration and no orbital degeneracy. This is favored by the atomic Hund’s Rule whose strength, as they point out, is not screened in the crystal. The differing planar exchange constants they attribute to the monoclinic distortion in the AFI-phase. Yet general considerations of the phase diagram and NMR investigations all point towards to the presence of an orbital degeneracy.
Here we take a different approach to the AFI-phase and start from an atomic limit but consider first the V-V pairs, since the intersite $`a_{1g}`$-hopping matrix elements are the largest. Keeping a strong Hund’s Rule coupling, as proposed by Ezhov et al., leads us to molecular orbitals for a V-V pair consisting of a superposition of $`(e_ge_g)`$ on one V-site and of $`(e_ga_{1g})`$ on the second site with a total spin $`S=2`$. This delocalized molecular orbital has also a two-fold degeneracy due to a choice in $`(e_ga_{1g})`$ among the $`e_g`$-doublet. Next we consider planar hopping processes and show that in a reasonable parameter range the real spin (RS) structure is the most stable. This state has a ferro-arrangement of the molecular orbitals which agrees with the monoclinic structure and, as we shall see, also with the x-ray experiments of Paolasini et al.
Let us start with a description of a vertical pair. Following Ref., the two $`e_g`$ orbitals are specified by a further index as $`|e_{g1}=1/\sqrt{2}(|d_{yz}|d_{zx})`$ and $`|e_{g2}=1/\sqrt{6}(2|d_{xy}|d_{yz}|d_{zx})`$, while the $`a_{1g}`$ orbital is given by $`|a_{1g}=1/\sqrt{3}(|d_{xy}+|d_{yz}+|d_{zx})`$. For each V ion, the d-orbitals are defined in a local coordinate system whose axis point towards the surrounding O ions, and thus refer to different, symmetry related orbitals for the different V ions in the unit cell. Consequently, the $`e_{g1}`$ and $`e_{g2}`$ orbitals on the two V ions of a vertical pair are not identical. This will be important when we compare our results to that of resonant scattering experiments. The intra-atomic interaction is described by three parameters: $`U`$, the Coulomb interaction in the same orbital, $`U^{}`$, the Coulomb interaction in different orbitals, and $`J`$, the Hund’s Rule coupling, which we assume satisfy the usual relation for $`t_{2g}`$ orbitals: $`U=U^{}+2J`$. The trigonal crystal field induces an energy splitting $`\mathrm{\Delta }`$ between the low-lying $`e_g`$ orbitals and the excited $`a_{1g}`$. Finally, the hopping integrals are denoted by $`t_{ij}^\delta `$ where $`\delta =a,b,c,d`$ stands for the direction of the bond ($`a,b,c`$: bonds inside the hexagonal planes, $`d`$: vertical bond) while $`(i,j)=1,2,3`$ denote the orbitals ($`e_{g1}`$,$`e_{g2}`$ and $`a_{1g}`$ respectively).
The main difference with Ref. comes from the values of the interaction parameters. The values used in Ref. ($`U2eV`$, $`J=0.2eV`$) are now believed to be much too small: Recent estimates based on LDA+U calculations are in the range $`U5`$ eV and $`J1`$ eV. It turns out that this makes a dramatic difference for the ground state of a V-V pair. To be specific, if we consider the same hopping and crystal field parameters as in Ref., and if we fix the ratio $`J/U=0.1`$ to the value they used, there is a level crossing as a function of $`U`$ between two very different situations. At small $`U`$, the ground state is 3-fold degenerate, with 3 levels nearby. This corresponds to the limit of Ref. where two electrons go into the bonding molecular orbital built out of $`a_{1g}`$ orbitals, the other two electrons being described by a spin 1/2 - pseudo spin 1/2 Kugel-Khomskii model. At large $`U`$ however, the ground state is 10-fold degenerate. It corresponds to a total spin 2 with a two-fold degenerate orbital state. Since by symmetry $`t_{ij}^d=0`$ if $`ij`$, this orbital wavefunction can actually be written down explicitly:
$$|\pm =\frac{|e_g,a_{1g}|e_{g1},e_{g2}+|e_{g1},e_{g2}|e_g,a_{1g}}{\sqrt{2}}$$
(1)
where $`e_g`$ stands for $`e_{g1}`$ ($`e_{g2}`$) in $`|`$ ($`|+`$). This situation is generic for a large range of $`J/U`$ including $`J/U=0.2`$, and with the parameters proposed in Ref., we found that the ground state is clearly of this second type.
It is interesting to compare this state with the spin 1 picture of Ezhov et al.. When the Hund’s Rule coupling is large, all low-lying states can indeed be described by considering only the states with total spin 1 at each site. However, the resulting effective Hamiltonian for a pair of sites is not simply a Heisenberg Hamiltonian, since this would correspond to only 9 low-lying states. In fact there are 81 low-lying states, suggesting that it one wants to describe this system with a spin 1 operator, $`\stackrel{}{S}`$, at each site, one should also include a pseudo-spin 1 operator, $`\stackrel{}{T}`$, to describe the quasi-degeneracy of the $`t_{2g}`$ orbitals. This orbital degree of freedom is crucial since it is responsible for a factor 2 in the 10-fold groundstate degeneracy.
These results suggest that, instead of starting from a spin-orbital model with a spin 1/2 and a pseudo-spin 1/2 at each V site, one should start from a spin-orbital model in which each vertical V-V pair is decribed by a spin 2 for the total spin, say $`\stackrel{}{\sigma }`$, and a pseudo-spin 1/2 for the orbital degeneracy, say $`\stackrel{}{\tau }`$, $`\tau ^z=1/2`$ ($`\tau ^z=1/2`$) corresponding to $`|+`$ ($`|`$) in Eq.1. The low energy properties are then determined by the way the degeneracy is lifted when these pairs are coupled by the in-plane hopping integrals. Since these hopping parameters are small, we can treat them within second-order perturbation theory. For simplicity, we include only the largest hopping integral $`t_{23}^at`$, and the corresponding hopping integrals for directions $`b`$ and $`c`$, in the present discussion. We have checked that the conclusions are unaffected by this simplification. The second order effective spin-orbital Hamiltonian for n.n. along the $`a`$-axis then reads:
$$\stackrel{~}{H}_{\mathrm{eff}}(a)=G\stackrel{}{\sigma }_l\stackrel{}{\sigma }_m+\frac{1}{4}G_3(\tau _l^z+\tau _m^z)\stackrel{}{\sigma }_l\stackrel{}{\sigma }_m,$$
(2)
with
$`G={\displaystyle \frac{1}{3}}G_1+{\displaystyle \frac{1}{3}}G_2+{\displaystyle \frac{3}{4}}G_3`$ , $`G_1={\displaystyle \frac{t^2}{4(U^{}J)}}`$ (3)
$`G_2={\displaystyle \frac{t^2}{4(U^{}+2J)}}`$ $`,`$ $`G_3={\displaystyle \frac{t^2}{4(U+J)}}`$ (4)
The effective Hamiltonians for n.n. along the $`b`$\- and $`c`$-axes are easily obtained by the trigonal rotation of $`\stackrel{~}{H}_{eff}(a)`$ equivalent to the following replacement of the orbital pseudo-spin $`\tau ^z1/2\tau ^z\pm \sqrt{3}/2\tau ^x`$. While there is a strong anisotropy in orbital space, the interaction preserves $`SU(2)`$ symmetry for the spin operator $`\stackrel{}{\sigma }`$.
Remarkably enough, the symmetry properties of this model are quite similar to the Kugel-Khomskii model for the cubic perovskite. In fact one can regard the corundum lattice as a distorted simple cubic ($`sc`$) lattice of the V-V pairs. This analogy is useful to give a systematic analysis for such a complicated system. Namely, it is promising that the stable magnetic phases of Kugel-Khomskii-type models are collinear two-sublattice orderings with associated orbital orderings. Within this criterion the possibility is naturally restricted into G, F, C, and A-type magnetic patterns. In this language, the realistic magnetic structure of V<sub>2</sub>O<sub>3</sub> corresponds to the $`C`$-type arrangement in the pair $`sc`$ lattice: One of 3 in-plane bonds is ferromagnetic and other two bonds are antiferromagnetic.
Keeping in mind these relations, we have examined the stable phase in the molecular model by comparing the energies of all magnetic phases. This has been done, as for the Kugel-Khomskii model, within a mean-field decoupling based on the order parameters $`\tau ^\alpha `$, $`\sigma ^\alpha `$, and $`\tau ^\alpha \sigma ^\beta `$. Details will be given in a forthcoming paper. The results are plotted in Fig.2 as a function of Hund’s Rule coupling $`J`$. It turns out that the stable phase changes successively from G to F phase as $`J`$ increases. In order to gain energy by the orbital-dependent $`G_3`$ term, the symmetry-broken C and A phases are stabilized in the intermediate-$`J`$ region. In particular the realistic $`C`$-type phase is found to be the lowest for $`J/U`$ around $`0.2`$, which agrees with the estimates of Ezhov et al., and which is consistent also with the stability region for the $`S=2`$ degenerate molecular orbitals of a V-pair. For this phase, the orbital order parameter is ferromagnetic with $`\tau ^z<0`$, i.e. the $`e_{g1}`$ orbital is favoured (state $`|`$ of Eq. 1). Such a ferro molecular orbital order will cause an effective uniaxial stress on the lattice degrees of freedom, leading to a uniform rotation of V-V pairs. This is consistent with the monoclinic distortion proposed by Dernier and Marezio.
The physical picture that emerges from this model is very encouraging. First of all, the observed magnetic arrangement is consistent with this model for reasonable values of the parameters. Second, the atomic configuration is a mixture of ($`e_ge_g`$) and ($`a_{1g}e_g`$), in agreement with X-ray absorption. Third, there is an orbital degree of freedom whose ordering is consistent with the monoclinic distortion of the low-temperature phase. It corresponds to choosing between $`e_{g1}`$ and $`e_{g2}`$ for the V-V pairs.
The results of our model are also consistent with the resonant x-ray scattering experiment of Paolasini et al.. In that experiment, resonant scattering was observed in the low temperature phase at wavevector $`q=(111)`$ and at energies corresponding to the transition from 1s to unoccupied 3d states on V-ions. Since this is forbidden by symmetry if all the orbital configurations on V-atoms are equal, this led Paolasini et al. to conclude to the existence of orbital ordering. They further suggested that their experiment can be taken to confirm one of the orbital ordered phases previously proposed by Castellani et al. We now show that the orbital ordering proposed in this paper for our model, while different from that interpreted by Paolasini et al., is also consistent with the experimental results.
The resonant scattering amplitude as a function of the energy $`\omega `$ and $`q`$ of x-rays for a crystal $`V_2O_3`$ is given by $`F=_{i=1}^8e^{i\stackrel{}{q}\stackrel{}{\rho _i}}f_i(\omega )`$, where $`f_i(\omega )`$ is the amplitude contributed from the V atom at position $`\stackrel{}{\rho _i}`$ in the monoclinic unit cell of $`V_2O_3`$. The low-temperature monoclinic lattice of $`V_2O_3`$ has eight atoms in a unit cell (Fig.3). Atoms 1-4 have spin-up magnetic moments and atoms 5-8 have spin-down magnetic moments. F at $`q=(111)`$ is given by
$$F_{111}=(f_1f_5+f_8f_4)e^{i\alpha }+(f_2f_6+f_7f_3)e^{i\alpha }$$
where $`\alpha `$ is a phase factor which depends on $`\stackrel{}{\rho }_1\stackrel{}{\rho }_2`$. $`f_i(\omega )`$ depends in general on the magnetic moment and orbital occupation. The nonvanishing intensity of (111) reflection for this energy implies that the combinations $`(f_1f_5+f_8f_4)`$ and $`(f_2f_6+f_7f_3)`$ are nonzero. The ferro-orbital phase in our model exhibits this feature for the following reasons. As discussed before, the $`e_g`$ orbitals are defined with respect to a local co-ordinate system on each V ion. In particular, for the two V ions on a vertical bond, they are related by a rotation around the $`y`$-axis: $`C_2(x,y,z)=(x,y,z)`$ (the trigonal coordinate system is used here with $`z`$-axis directed perpendicular to the hexagonal plane) while for the V ions in the same hexagonal plane local coordinate systems are identical. Thus, the ferro-orbital phase actually corresponds to having different orbitals on alternate hexagonal planes. Denoting these as 1 and 2 and denoting $`u`$ and $`d`$ for the spin up and down magnetic state, we then have e.g.
$$f_1f_5+f_8f_4=f(1,u)f(1,d)+f(2,d)f(2,u)$$
Since f depends on both the orbital occupation and the magnetic moment of the V-atom, $`f(u)f(d)`$. Thus, our model gives nonzero $`F_{111}`$, and is qualitatively consistent with the experimental observation of Paolasini et al. . More work is needed to compare our theory with the observed polarization and the azimuthal dependences of the resonance intensity.
According to this explanation, the intensity of the (111) reflection is not simply a direct consequence of the orbital order, but comes both from magnetic and orbital order. This should be contrasted to Paolasini et al.’s explanation based on Ref., where the form factor of Eq. (5) is non zero because the orbitals occupied on the two V of a vertical pair are different linear combinations of $`eg_1`$ and $`eg_2`$. While Castellani et al.’s picture is very specific to their S=1/2 model, there is room a priori within the S=1 model for a similar orbital ordering. Energetic considerations then show that the only serious candidate is an orbital ordering in which one V of a vertical pair would be in the ($`eg`$ $`a_{1g}`$) configuration, while the other one would be in the ($`eg`$ $`eg`$) configuration. To be more precise, the model considered thus far corresponds to having all interplane hopping amplitudes much larger than intraplane ones. If we maintain the limit of $`t_{33}^d`$ large, but allows $`t_{11}^d`$ and $`t_{22}^d`$ to be comparable to intraplane hoppings, it can be shown that the problem can then be mapped into a transverse field Ising model. Details will be given in a forthcoming publication . Here we briefly summarize the main results. In this mapping, there is an Ising spin associated with each V-V pair which corresponds to the orbital occupation of $`(e_ge_g)`$ and $`(e_ga_{1g})`$ on the two site. The transverse field strength $`h`$ is given by the energy difference of the bonding and antibonding states of a V-V pair, a measure of the orbital quantum fluctuation in a V-V pair. The molecular orbital model corresponds to the large $`h`$ limit. In the opposite limit $`h0`$, the spin and orbital ordering depends on the relative strengths of the intraplane hoppings. In a reasonable parameter range, we found a ground state with RS spin and an orbital ordering corresponding to the pattern reported by Paolasini et al. However, since in this state the V ions of a vertical pair are in $`(e_ge_g)`$ and $`(e_ga_{1g})`$ configurations, the electronic densities are very different. This should lead to different local distortions of the O octahedra, which is inconsistent with the monoclinic structure reported by Dernier and Marezio, where all V are equivalent. So we do not think that this kind of orbital ordering is realized in V<sub>2</sub>O<sub>3</sub>.
To summarize, we have proposed a model for the AF insulating phase of V<sub>2</sub>O<sub>3</sub>. This model seems to be the only way to combine basic facts about the electronic structure (S=1, orbital degeneracy, strong coupling along vertical pairs) into a coherent picture that agrees with all experiments. Further investigation of the resulting two-fold degenerate, S=2 model for the vertical pair is in progress.
We acknowledge useful discussions with W. Bao, B. Normand, P. M. Platzmann, G. Sawatzky, L. H. Tjeng and C. Vettier, and the hospitality of the Center for Theoretical Studies of ETH Zürich. This work was in part supported by the DOE grant DE/FG03-98ER45687, Russian Foundation for Basis Reasearch grants RFFI-98-02-17275 and RFFI-00-15-96575.
|
warning/0005/gr-qc0005037.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
One of the most non-trivial aspects of perturbation equations for the Kerr-de Sitter geometry is the separability of the radial and angular parts. Carter first showed that the perturbation equation for a scalar field is separable in the Kerr-Newman-de Sitter geometry . This observation was extended for spin $`1/2`$, electromagnetic fields, gravitational perturbations and gravitinos for the Kerr geometries and even for the Kerr-de Sitter class of geometries. These perturbation equations are called Teukolsky equations . Except for electromagnetic and gravitational perturbations, the separability persist even for the Kerr-Newman-de Sitter solutions.
An important application of the separability is the proof of the stability of the black hole. The proof of mode stability of the Kerr black hole was provided by Whiting in 1989. The proof is more complicated than one for the Schwarzschild black hole, because in the Kerr geometry there is no Killing vector which is timelike everywhere in the region exterior of the outer horizon. In his proof, he skillfully used differential and integral transformations of solutions of perturbation equations, and obtained a conserved quantity which is well-defined for unstable modes which are purely incoming on the outer horizon and purely outgoing at the infinity, and are characterized by positive imaginary part of their frequencies. Then he showed that the positivity of the quantity bounds the magnitudes of the time derivative of perturbations.
On the other hand, it was shown that the perturbation equations in the Kerr geometry are obtained from those in the Kerr-de Sitter geometry in the confluent limit. In other words, a irregular singularity at infinity of the equation in the Kerr case is separated into two regular singularities of the equation in the Kerr-de Sitter geometry by cosmological constant. In a series of paper , Suzuki, Takasugi and the author have constructed analytic solutions of the perturbation equations of massless fields in the Kerr-de Sitter geometries. We found transformations such that both the angular and the radial equations are reduced to Heun’s equation . The solution of Heun’s equation (Heun’s function) is expressed in the form of a series of hypergeometric functions, and its coefficients are determined by three-term recurrence relations . The solution of the radial equation which is valid in the entire physical region is obtained by matching two solutions which have different convergence regions in the region where both solutions are convergent. We examined properties of the solution in detail and, in particular, analytically showed that our solution satisfies the Teukolsky-Starobinsky identities . There are similarities between the procedures for solving the perturbation equations in the Kerr and Kerr-de Sitter geometries. Thus differential and integral transformations of Heun’s function may be useful for studying mode stability of the Kerr-de Sitter geometry.
In this paper, we investigate differential and integral transformations of solutions of massless perturbation equations in the Kerr-de Sitter geometry. These transformations map a solution of a Heun’s equation to a solution of another Heun’s equation which has different parameters from original ones. Differential transformations include the Teukolsky-Starobinsky identities as special cases, and also other transformations for angular functions. We will consider two specific integral transformations because we do not know systematic way to study integral transformations of Heun’s function. Then we will apply the integral transformation to a solution of the radial equation. It will turn out to be possible to make the transformations of radial functions only for unstable modes which are purely incoming on the outer horizon and purely outgoing on the de Sitter horizon, and have positive imaginary part of their frequencies. A conserved energy integral will be constructed from the transforms of the angular and radial functions for unstable modes, and we discuss properties of the quantity. In the Kerr limit, these transformations and the conserved quantity are coincide with those given in Ref. by Whiting in order to prove mode stability of the Kerr black hole.
## 2 The Teukolsky equations for the Kerr-de Sitter <br>geometry
We consider perturbation equations for massless fields in the Kerr-de Sitter geometries. In the Boyer-Lindquist coordinates, the Kerr-de Sitter metric has the form,
$`ds^2`$ $`=`$ $`\rho ^2\left({\displaystyle \frac{dr^2}{\mathrm{\Delta }_r}}+{\displaystyle \frac{d\theta ^2}{\mathrm{\Delta }_\theta }}\right){\displaystyle \frac{\mathrm{\Delta }_\theta \mathrm{sin}^2\theta }{(1+\alpha )^2\rho ^2}}[adt(r^2+a^2)d\phi ]^2`$ (2.1)
$`+{\displaystyle \frac{\mathrm{\Delta }_r}{(1+\alpha )^2\rho ^2}}(dta\mathrm{sin}^2\theta d\phi )^2,`$
where $`\alpha =\mathrm{\Lambda }a^2/3`$, $`\rho ^2=\overline{\rho }\overline{\rho }^{}`$, $`\overline{\rho }=r+ia\mathrm{cos}\theta `$ and
$`\mathrm{\Delta }_r`$ $`=`$ $`(r^2+a^2)\left(1{\displaystyle \frac{\alpha }{a^2}}r^2\right)2Mr`$
$`=`$ $`{\displaystyle \frac{\alpha }{a^2}}(rr_+)(rr_{})(rr_+^{})(rr_{}^{}),`$
$`\mathrm{\Delta }_\theta `$ $`=`$ $`1+\alpha \mathrm{cos}^2\theta .`$ (2.2)
Here $`\mathrm{\Lambda }`$ is the cosmological constant, $`M`$ is the mass of the black hole, and $`aM`$ is its angular momentum.
We assume that the coordinate dependences of the perturbation of the field have form $`\mathrm{\Phi }_s=e^{i(\omega tm\phi )}R_s(r)S_s(\theta )`$. Then angular Teukolsky equation for massless field with spin weight $`s`$ is given by
$`\{`$ $`{\displaystyle \frac{d}{dx}}(1+\alpha x^2)(1x^2){\displaystyle \frac{d}{dx}}+\lambda _s+{\displaystyle \frac{(1+\alpha )^2}{\alpha }}\xi ^22\alpha x^2`$ (2.3)
$`+`$ $`{\displaystyle \frac{1+\alpha }{1+\alpha x^2}}\left[2s(\alpha m(1+\alpha )\xi )x{\displaystyle \frac{(1+\alpha )^2}{\alpha }}\xi ^2+2m(1+\alpha )\xi +s^2\right]`$
$``$ $`{\displaystyle \frac{(1+\alpha )^2m^2}{(1+\alpha x^2)(1x^2)}}{\displaystyle \frac{(1+\alpha )(s^2+2smx)}{1x^2}}\}S_s(x)=0,`$
where $`x=\mathrm{cos}\theta `$ and $`\xi =a\omega `$. The separation constant $`\lambda _s`$ is an even function of $`s`$ as shown in Ref.. This equation has five regular singularities at $`\pm 1`$, $`\pm \frac{i}{\sqrt{\alpha }}`$ and $`\mathrm{}`$. We define the variable $`z`$ by
$$z=\frac{1\frac{i}{\sqrt{\alpha }}}{2}\frac{x+1}{x\frac{i}{\sqrt{\alpha }}}.$$
(2.4)
Then the singularities are transformed to $`z=0,1,z_s,z_{\mathrm{}}`$, and $`\mathrm{}`$ where $`z_s=\frac{i(1+i\sqrt{\alpha })^2}{4\sqrt{\alpha }}`$ and $`z_{\mathrm{}}=\frac{i(1+i\sqrt{\alpha })}{2\sqrt{\alpha }}`$. The singularity at $`z=z_{\mathrm{}}`$ which corresponds to $`x=\mathrm{}`$ can be factored out by the transformation
$$S_s(z)=z^{C_1}(z1)^{C_2}(zz_s)^{C_3}(zz_{\mathrm{}})f_S(z),$$
(2.5)
where $`C_1=\delta _1(ms)/2`$, $`C_2=\delta _2(m+s)/2`$ and $`C_3=\delta _3\frac{i}{2}\left(\frac{1+\alpha }{\sqrt{\alpha }}\xi \sqrt{\alpha }mis\right)`$, ($`\delta _1,\delta _2,\delta _3=\pm 1`$). Now $`f_S(z)`$ satisfies the equation
$$\left\{\frac{d^2}{dz^2}+\left[\frac{2C_1+1}{z}+\frac{2C_2+1}{z1}+\frac{2C_3+1}{zz_s}\right]\frac{d}{dz}+\frac{\rho _+\rho _{}z+u}{z(z1)(zz_s)}\right\}f_S(z)=0,$$
(2.6)
where
$`\rho _\pm `$ $`=`$ $`C_1+C_2+C_3\pm C_3^{}+1,`$ (2.7)
$`u`$ $`=`$ $`{\displaystyle \frac{i}{4\sqrt{\alpha }}}\{\lambda _s2i\sqrt{\alpha }+2(1+\alpha )(m+s)\xi (1+i\sqrt{\alpha })^2(2C_1C_2+C_1+C_2)`$ (2.8)
$`4i\sqrt{\alpha }(2C_1C_3+C_1+C_3){\displaystyle \frac{m^2}{2}}\left[4\alpha +(1+i\sqrt{\alpha })^2\right]`$
$`+{\displaystyle \frac{s^2}{2}}(1i\sqrt{\alpha })^2+2ims\sqrt{\alpha }(1+i\sqrt{\alpha })\}.`$
Equation (2.6) is called the Heun’s equation which has four regular singularities.
Next we consider the radial Teukolsky equation which is given by
$`\{\mathrm{\Delta }_r^s{\displaystyle \frac{d}{dr}}\mathrm{\Delta }_r^{s+1}{\displaystyle \frac{d}{dr}}+{\displaystyle \frac{1}{\mathrm{\Delta }_r}}[(1+\alpha )^2K^2is(1+\alpha )K{\displaystyle \frac{d\mathrm{\Delta }_r}{dr}}]`$
$`+4is(1+\alpha )\omega r{\displaystyle \frac{2\alpha }{a^2}}(s+1)(2s+1)r^2+s(1\alpha )\lambda _s\}R_s(r)=0,`$ (2.9)
with $`K=\omega (r^2+a^2)am`$. This equation has five regular singularities at $`r_\pm ,r_\pm ^{}`$, and $`\mathrm{}`$ which are assigned such that $`r_\pm M\pm \sqrt{M^2a^2Q^2}r_\pm ^0`$ and $`r_\pm ^{}\pm a/\sqrt{\alpha }`$ in the limit $`\alpha 0(\mathrm{\Lambda }0)`$. And the coefficients of the equation are complex for spin fields ($`s0`$). We assume that the cosmological constant is sufficiently small so that all $`r_\pm ,r_\pm ^{}`$ are real. By using the new variable
$$z=\left(\frac{r_+r_{}^{}}{r_+r_{}}\right)\left(\frac{rr_{}}{rr_{}^{}}\right),$$
(2.10)
Eq.(2.9) becomes an equation which has regular singularities at $`0,1,z_r,z_{\mathrm{}}`$ and $`\mathrm{}`$,
$$z_r=\left(\frac{r_+r_{}^{}}{r_+r_{}}\right)\left(\frac{r_+^{}r_{}}{r_+^{}r_{}^{}}\right),z_{\mathrm{}}=\frac{r_+r_{}^{}}{r_+r_{}}.$$
To proceed further, we define the parameters
$`D_{i\pm }`$ $`=`$ $`{\displaystyle \frac{1}{2}}\{s\pm (2a_i+s)\}.(i=1,2,3,4)`$ (2.11)
Here $`a`$’s are purely imaginary numbers defined by
$`a_1`$ $`=`$ $`i{\displaystyle \frac{a^2(1+\alpha )\left(\omega (r_+^2+a^2)am\right)}{\alpha (r_+^{}r_+)(r_{}^{}r_+)(r_{}r_+)}},`$
$`a_2`$ $`=`$ $`i{\displaystyle \frac{a^2(1+\alpha )\left(\omega (r_{}^2+a^2)am\right)}{\alpha (r_+^{}r_{})(r_{}^{}r_{})(r_+r_{})}},`$
$`a_3`$ $`=`$ $`i{\displaystyle \frac{a^2(1+\alpha )\left(\omega (r_{+}^{}{}_{}{}^{2}+a^2)am\right)}{\alpha (r_{}r_+^{})(r_{}^{}r_+^{})(r_+r_+^{})}},`$
$`a_4`$ $`=`$ $`i{\displaystyle \frac{a^2(1+\alpha )\left(\omega (r_{}^2+a^2)am\right)}{\alpha (r_{}r_{}^{})(r_+^{}r_{}^{})(r_+r_{}^{})}},`$ (2.12)
and the relation $`a_1+a_2+a_3+a_4=0`$ is satisfied. Again we can factor out the singularity at $`z=z_{\mathrm{}}`$ by the transformation as
$$R_s(z)=z^{D_2}(z1)^{D_1}(zz_r)^{D_3}(zz_{\mathrm{}})^{2s+1}f_R(z),$$
(2.13)
where $`D_i(i=1,2,3)`$ is either $`D_{i+}`$ or $`D_i`$. Then, $`f_R(z)`$ satisfies the Heun’s equation as
$$\left\{\frac{d^2}{dz^2}+\left[\frac{\gamma }{z}+\frac{\delta }{z1}+\frac{ϵ}{zz_r}\right]\frac{d}{dz}+\frac{\sigma _+\sigma _{}z+v}{z(z1)(zz_r)}\right\}f_R(z)=0,$$
(2.14)
where
$`\gamma =2D_2+s+1,`$ $`\delta =2D_1+s+1,ϵ=2D_3+s+1,`$
$`\sigma _\pm `$ $`=`$ $`D_1+D_2+D_3+D_4+2s+1.`$ (2.15)
The parameters $`\gamma `$, $`\delta `$, $`ϵ`$ and $`\sigma _\pm `$ satisfy the following relation,
$$\gamma +\delta 1=\sigma _++\sigma _{}ϵ,$$
(2.16)
which is required for Eq.(2.14) to be a Heun’s equation. The remaining parameter $`v`$ is given by
$`v`$ $`=`$ $`{\displaystyle \frac{2a^4(1+\alpha )^2}{\alpha ^2𝒟}}{\displaystyle \frac{(r_+r_+^{})^2(r_+r_{}^{})^2(r_{}r_{}^{})(r_+^{}r_{}^{})}{r_+r_{}}}`$ (2.17)
$`\times \{\omega ^2r_{}^3(r_+r_{}2r_+r_+^{}+r_{}r_+^{})+2a\omega (a\omega m)r_{}(r_+r_+^{}r_{}^2)`$
$`a^2(a\omega m)^2(2r_{}r_+r_+^{})`$
$`+{\displaystyle \frac{2isa^2(1+\alpha )}{\alpha }}{\displaystyle \frac{\left[\omega (r_{}r_{}^{}+a^2)am\right]}{(r_+r_{})(r_+^{}r_{}^{})(r_{}r_{}^{})}}`$
$`+(s+1)(2s+1)\left[{\displaystyle \frac{2r_{}^{}{}_{}{}^{2}}{(r_+r_{})(r_+^{}r_{}^{})}}z_{\mathrm{}}\right]`$
$`2D_2(z_rD_1+D_3)(s+1)\left[(1+z_r)D_2+z_rD_1+D_3\right]`$
$`{\displaystyle \frac{a^2}{\alpha (r_+r_{})(r_+^{}r_{}^{})}}[\lambda _s+s(1\alpha )]\}.`$
Here $`𝒟`$ is the discriminant of $`\mathrm{\Delta }_r=0`$,
$`𝒟`$ $`=`$ $`(r_+r_{})^2(r_+r_+^{})^2(r_+r_{}^{})^2(r_{}r_+^{})^2(r_{}r_{}^{})^2(r_+^{}r_{}^{})^2`$
$`=`$ $`{\displaystyle \frac{16a^{10}}{\alpha ^5}}\{(1\alpha )^3[M^2(1\alpha )a^2]`$
$`+{\displaystyle \frac{\alpha }{a^2}}[27M^4+36(1\alpha )M^2a^28(1\alpha )^2a^4]{\displaystyle \frac{16\alpha ^2}{a^4}}a^6\}.`$
It should be noted that the equation (2.9) for $`s=0`$ is the perturbation equation for a conformal scalar field which satisfies $`\mathrm{}\varphi =\frac{1}{6}R\varphi `$. In the case of an ordinary scalar field which satisfies $`\mathrm{}\varphi =0`$, the term $`\frac{2\alpha }{a^2}r^2`$ in Eq.(2.9) is absent.
## 3 Differential transformations
We consider a differential transformation of a Heun’s function $`f(z)`$ which satisfies
$`M_z(\gamma ,\delta ,ϵ;\alpha ,\beta ;q)f(z)\{z(z1)(za_H){\displaystyle \frac{^2}{z^2}}`$
$`+[\gamma (z1)(za_H)+\delta z(za_H)+ϵz(z1)]{\displaystyle \frac{}{z}}+\alpha \beta z+q\}f(z)=0.`$ (3.1)
Differentiating this equation $`N`$ times, we obtain
$`\left({\displaystyle \frac{d}{dz}}\right)^NM_z(\gamma ,\delta ,ϵ;\alpha ,\beta ;q)f(z)=\{z(z1)(za_H){\displaystyle \frac{^2}{z^2}}`$
$`+\left[(\gamma +N)(z1)(za_H)+(\delta +N)z(za_H)+(ϵ+N)z(z1)\right]{\displaystyle \frac{}{z}}`$
$`+\left[3N^2+(2\alpha +2\beta 1)N+\alpha \beta \right]z+qN(N1)(a_H+1)N((a_H+1)\gamma +a_H\delta +ϵ)`$
$`+N(N1+\alpha )(N1+\beta )\left({\displaystyle \frac{}{z}}\right)^1\left\}\right({\displaystyle \frac{d}{dz}})^Nf(z).`$ (3.2)
Therefore
$$\stackrel{~}{f}(z)=\left(\frac{d}{dz}\right)^Nf(z)$$
(3.3)
with $`N=1\alpha `$ or $`1\beta `$ formally satisfies another Heun’s equation in which the parameters $`(\gamma ,\delta ,\mathrm{})`$ are replaced by $`\stackrel{~}{\gamma }=\gamma +N,\stackrel{~}{\delta }=\delta +N,\stackrel{~}{ϵ}=ϵ+N,\stackrel{~}{q}=qN(N1)(a_H+1)N((a_H+1)\gamma +a_H\delta +ϵ)`$, and $`\stackrel{~}{\alpha },\stackrel{~}{\beta }`$ which are determined by relations $`\stackrel{~}{\alpha }+\stackrel{~}{\beta }=\alpha +\beta +3N`$ and $`\stackrel{~}{\alpha }\stackrel{~}{\beta }=3N^2+(2\alpha +2\beta 1)N+\alpha \beta `$. Here we choose $`N=1\alpha `$. Properly $`N`$ should be an positive integer. In both angular and radial cases, we can take parameters so that $`N=|2s|`$. These transformations are the Teukolsky-Starobinsky relations . In Ref. it was shown that the radial Teukolsky-Starobinsky relations are satisfied by using an analytic solution expressed by the form of a series of hypergeometric functions.
In the angular case, $`N=|sm|`$ and $`N=|s+m|`$ are also possible. We now explain the case with $`\delta _1=\delta _2=\delta _3`$ in the definitions of $`C_i(i=1,2,3)`$, explicitly. Then $`N=\delta _1(sm)`$ and thus we can take $`\delta _1(sm)=|sm|`$. The transform of angular function $`\stackrel{~}{f}_S(z)=\left(\frac{d}{dz}\right)^Nf_S(z)`$ satisfies the Heun’s equation with the following parameters,
$`\stackrel{~}{\gamma }=1,\stackrel{~}{\delta }=2\delta _1s+1,\stackrel{~}{ϵ}=\delta _1\left[i{\displaystyle \frac{1+\alpha }{\sqrt{\alpha }}}\xi +(1i\sqrt{\alpha })m\right]+1,`$
$`\stackrel{~}{\rho }_+=1+|sm|,\stackrel{~}{\rho }_{}=\delta _1i\left[{\displaystyle \frac{1+\alpha }{\sqrt{\alpha }}}\xi \sqrt{\alpha }m+is\right]+1,`$ (3.4)
$`\stackrel{~}{u}=u|sm|\left[z_s(\delta _1(m+s)+1)\delta _1i\left({\displaystyle \frac{1+\alpha }{\sqrt{\alpha }}}\xi \sqrt{\alpha }mis\right)+1\right].`$
We next define $`\stackrel{~}{S}_s(z)`$ from $`\stackrel{~}{f}_S(z)`$ through the inverse procedures to those used to derive $`f_S(z)`$ from $`S_s(z)`$. By setting $`\stackrel{~}{\gamma }=2\stackrel{~}{C}_1+1,\stackrel{~}{\delta }=2\stackrel{~}{C}_2+1`$ and $`\stackrel{~}{ϵ}=2\stackrel{~}{C}_3+1`$, we obtain $`\stackrel{~}{C}_1=0,\stackrel{~}{C}_2=\delta _1s`$ and $`\stackrel{~}{C}_3=\frac{\delta _1}{2}\left[i\frac{1+\alpha }{\sqrt{\alpha }}\xi +(1i\sqrt{\alpha })m\right]`$. Using these $`\stackrel{~}{C}_i`$ as the exponents at the singularities, $`\stackrel{~}{S}_s(z)`$ is given by
$$\stackrel{~}{S}_s(z)z^{\stackrel{~}{C}_1}(z1)^{\stackrel{~}{C}_2}(zz_s)^{\stackrel{~}{C}_3}(zz_{\mathrm{}})\stackrel{~}{f}_S(z).$$
(3.5)
This new angular function $`\stackrel{~}{S}_s`$ satisfies a similar equation to the angular Teukolsky equation (2.3) satisfied by $`S_s`$;
$`\{{\displaystyle \frac{d}{dx}}(1x^2)(1+\alpha x^2){\displaystyle \frac{d}{dx}}+\lambda _s2\alpha x^2(1+\alpha )^2{\displaystyle \frac{1x^2}{1+\alpha x^2}}\xi ^2+2(1+\alpha )^2{\displaystyle \frac{1x}{1+\alpha x^2}}m\xi `$
$`\alpha (1+\alpha ){\displaystyle \frac{(1x)^2}{1+\alpha x^2}}m^2(1+\alpha ){\displaystyle \frac{1+x}{1x}}s^2\}\stackrel{~}{S}_s(x)=0,`$ (3.6)
where we used (2.4). We note that the operator acting on $`\stackrel{~}{S}_s`$ in the above equation is invariant under $`ss`$ because $`\lambda _s`$ is an even function of $`s`$.
## 4 Integral transformations
We construct integral transformations of Heun’s function $`f(z)`$ which satisfies Eq.(3). The integral transformation which maps a solution of a Heun’s equation to a solution of another Heun’s equation is given by
$$\stackrel{~}{f}(z)=_𝒞dtt^{\gamma 1}(t1)^{\delta 1}(ta_H)^{ϵ1}K(z,t)f(t).(\mathrm{type}\mathrm{A})$$
(4.1)
The function $`\stackrel{~}{f}(z)`$ is a solution of $`M_z(\stackrel{~}{\gamma },\stackrel{~}{\delta },\stackrel{~}{ϵ};\stackrel{~}{\alpha },\stackrel{~}{\beta };\stackrel{~}{q})\stackrel{~}{f}(z)=0`$ if the kernel satisfy the condition
$$\left[M_z(\stackrel{~}{\gamma },\stackrel{~}{\delta },\stackrel{~}{ϵ};\stackrel{~}{\alpha },\stackrel{~}{\beta };\stackrel{~}{q})M_t(\gamma ,\delta ,ϵ;\alpha ,\beta ;q)\right]K(z,t)=0,$$
(4.2)
and the surface term of the integral
$$t^\gamma (t1)^\delta (ta_H)^ϵ\left\{K(z,t)\frac{f(t)}{t}\frac{K(z,t)}{t}f(t)\right\}W(z,t),$$
(4.3)
vanishes at the ends of $`𝒞`$.
In this paper we consider the kernels which depend on $`z`$ and $`t`$ only by the form of their product $`zt\zeta `$ thus $`K(z,t)=K(\zeta )`$. Then Eq.(4.2) can be rewritten as
$`\{t[\zeta (\zeta a_H){\displaystyle \frac{d^2}{d\zeta ^2}}+((\alpha +\beta +1)\zeta a_H\stackrel{~}{\gamma }){\displaystyle \frac{d}{d\zeta }}+\alpha \beta ]`$
$`z\left[\zeta (\zeta a_H){\displaystyle \frac{d^2}{d\zeta ^2}}+\left((\stackrel{~}{\alpha }+\stackrel{~}{\beta }+1)\zeta a_H\gamma \right){\displaystyle \frac{d}{d\zeta }}+\stackrel{~}{\alpha }\stackrel{~}{\beta }\right]`$
$`+[(1+a_H)(\stackrel{~}{\gamma }\gamma )+a_H(\stackrel{~}{\delta }\delta )+\stackrel{~}{ϵ}ϵ]\zeta {\displaystyle \frac{d}{d\zeta }}+q\stackrel{~}{q}\}K(\zeta )=0.`$ (4.4)
In particular, we require the kernel to satisfy the following equations,
$`\left[\zeta (\zeta a_H){\displaystyle \frac{d^2}{d\zeta ^2}}+\left((\alpha +\beta +1)\zeta a_H\stackrel{~}{\gamma }\right){\displaystyle \frac{d}{d\zeta }}+\alpha \beta \right]K(\zeta )=0,`$
$`\left[\zeta (\zeta a_H){\displaystyle \frac{d^2}{d\zeta ^2}}+\left((\stackrel{~}{\alpha }+\stackrel{~}{\beta }+1)\zeta a_H\gamma \right){\displaystyle \frac{d}{d\zeta }}+\stackrel{~}{\alpha }\stackrel{~}{\beta }\right]K(\zeta )=0,`$ (4.5)
$`(1+a_H)(\stackrel{~}{\gamma }\gamma )+a_H(\stackrel{~}{\delta }\delta )+\stackrel{~}{ϵ}ϵ=0,`$
$`q=\stackrel{~}{q}.`$
There are four sets of solutions of these equations. Using invariance of the equations under $`\alpha \beta `$ and $`\stackrel{~}{\alpha }\stackrel{~}{\beta }`$, we can take the solutions without loss of generality as
$`K(\zeta )=\left(\zeta a_H\right)^\beta ,`$ (4.6)
$`\stackrel{~}{\alpha }=\gamma ,\stackrel{~}{\beta }=\beta ,\stackrel{~}{\gamma }=\alpha ,\stackrel{~}{\delta }=\beta ϵ+1,\stackrel{~}{ϵ}=\beta \delta +1,\stackrel{~}{q}=q.`$ (4.7)
Next we define another type of integral transformation,
$$\stackrel{~}{f}(z)=_1^{\frac{a_H}{z}}𝑑tt^{\gamma 1}(t1)^{\delta 1}(ta_H)^{ϵ1}K(z,t)f(t),(\mathrm{type}\mathrm{B})$$
(4.8)
where it should be noted that the integral region depends on $`z`$. In order for $`\stackrel{~}{f}(z)`$ to satisfy Heun’s equation, the following condition must hold instead of the conditions (4.3);
$`W(z,t)|_1^{\frac{a_H}{z}}+z(z1)(za_H)\{{\displaystyle \frac{}{z}}\left[{\displaystyle \frac{a_H}{z^2}}K(z,t)t^{\gamma 1}(t1)^{\delta 1}(ta_H)^{ϵ1}f(t)|_{t=\frac{a_H}{z}}\right]`$
$`+{\displaystyle \frac{a_H}{z^2}}{\displaystyle \frac{K(z,t)}{z}}t^{\gamma 1}(t1)^{\delta 1}(ta_H)^{ϵ1}f(t)|_{t=\frac{a_H}{z}}\}`$ (4.9)
$`+[\stackrel{~}{\gamma }(z1)(za_H)+\stackrel{~}{\delta }z(za_H)+\stackrel{~}{ϵ}z(z1)]{\displaystyle \frac{a_H}{z^2}}K(z,t)t^{\gamma 1}(t1)^{\delta 1}(ta_H)^{ϵ1}f(t)|_{t=\frac{a_H}{z}}`$
$`=0.`$
We can show by using the kernel in Eq.(4.6) that $`\stackrel{~}{f}(z)`$ is a solution of the Heun’s equation with the parameters in Eq.(4.7), provided that this condition is satisfied.
## 5 Conserved energy integral for unstable modes
In this section we are going to construct a conserved quantity for unstable modes in the Kerr-de Sitter geometry by using the differential and integral transformations of Heun’s function given in section 3 and 4. We will assume that the cosmological constant is sufficiently small so that $`r_{}^{}r_{}<r_+r_+^{}`$.
First, it can be easily checked that the differential transformations of the angular function for the Kerr-de Sitter black hole which we provided in section 3 coincide with those given by Whiting in the Kerr limit $`\mathrm{\Lambda }0(\alpha 0)`$. Indeed since the Teukolsky equations for the Kerr black hole have the forms of confluent Heun’s equation , this differential transformation in the Kerr limit becomes one for the confluent Heun’s function.
Next we consider the integral transformations of the radial function $`f_R(z)`$ which satisfies Eq.(2.14). We choose $`D_1=a_1s,D_2=a_2s`$ and $`D_3=a_3s`$ as the parameters included in the radial equation (2.14). Then we have
$`\sigma _+=2s+1,\sigma _{}=2a_4s+1,`$
$`\gamma =2a_2s+1,\delta =2a_1s+1,ϵ=2a_3s+1.`$ (5.1)
The unstable modes are purely incoming on the outer horizon $`z=1`$ and purely outgoing on the de Sitter horizon $`z=z_r`$,
$`R_s(z)`$ $``$ $`(z1)^{sa_1},(z1)`$ (5.2)
$``$ $`\left(1{\displaystyle \frac{z}{z_r}}\right)^{a_3},(zz_r)`$
or equivalently
$`f_R(z)`$ $``$ $`(z1)^0,(z1)`$ (5.3)
$``$ $`\left(1{\displaystyle \frac{z}{z_r}}\right)^{2a_3+s},(zz_r)`$
where we used the parameters determined above. Furthermore the unstable modes are characterized by having positive imaginary part of frequency $`\omega `$. In our discussions below, we choose the integral region as $`𝒞=(1,z_r)`$ in the integral transformation of type A.
We first examine the Kerr limit. In the Kerr limit, we find
$$z_r\frac{a}{2\sqrt{\alpha }(r_+^0r_{}^0)},a_4\frac{ia\omega }{2\sqrt{\alpha }},$$
(5.4)
where $`r_\pm ^0=M\pm \sqrt{M^2a^2}`$. Then both integral regions of the transformations of type A and B become $`(1,\mathrm{})`$ and the kernel (4.6) becomes of the Laplace type,
$$K(\zeta )e^{2i\omega (r_+^0r_{}^0)\zeta }.$$
(5.5)
Thus both integral transformations coincide with those used in the proof of mode stability of the Kerr black hole in this limit.
We next consider the boundary term $`W(z,t)|_1^{z_r}`$. In type A case, $`W(z,t)`$ behaves as
$`W(z,t)`$ $``$ $`(t1)^{2a_1s+1},(t1)`$ (5.6)
$``$ $`(tz_r)^{2a_3s+1}{\displaystyle \frac{d}{dt}}(tz_r)^{2a_3+s}.(tz_r)`$
The real part of $`a_1`$ is negative if the imaginary part of $`\omega `$ is positive. Thus if $`s0`$, $`W(z,t)`$ vanishes at $`t=1`$ for unstable modes. On the other hand, $`W(z,t)`$ does not vanish at $`t=z_r`$. Therefore this transformation is not appropriate. Although there is also a possibility that our choice of the integral region $`𝒞`$ is wrong, we do not adopt the transformation of type A in our discussions below. In type B case, $`W(z,t)`$ behaves as
$`W(z,t)`$ $``$ $`(t1)^{2a_1s+1},(t1)`$ (5.7)
$``$ $`(t{\displaystyle \frac{z_r}{z}})^{2a_4+s1}.(t{\displaystyle \frac{z_r}{z}})`$
Since the real part of $`a_4`$ is negative for unstable modes, it is clear that the condition (4) holds for unstable modes from the form of the kernel $`K(z,t)=(ztz_r)^{2a_4+s1}`$. Hence it is possible to make the integral transformation of type B for radial functions of unstable modes.
The radial function $`\stackrel{~}{f}_R(z)`$ given by the integral transformation of type B of $`f_R(z)`$,
$$\stackrel{~}{f}_R(z)=_1^{\frac{z_r}{z}}𝑑tt^{2a_2s}(t1)^{2a_1s}(tz_r)^{2a_3s}(ztz_r)^{2a_4+s1}f_R(t),$$
(5.8)
satisfies the Heun’s equation with parameters (4.7),
$`\stackrel{~}{\sigma }_+=2a_2s+1,\stackrel{~}{\sigma }_{}=2a_4s+1,`$
$`\stackrel{~}{\gamma }=2s+12\stackrel{~}{D}_2+1,\stackrel{~}{\delta }=2a_12a_2+12\stackrel{~}{D}_1+1,`$
$`\stackrel{~}{ϵ}=2a_22a_3+12\stackrel{~}{D}_3+1,\stackrel{~}{v}=v.`$ (5.9)
We define a new radial function $`\stackrel{~}{R}_s(z)`$ from $`\stackrel{~}{f}_R`$ by
$$\stackrel{~}{R}_s(z)=z^{\stackrel{~}{D}_2}(z1)^{\stackrel{~}{D}_1}(zz_r)^{\stackrel{~}{D}_3}(zz_{\mathrm{}})\stackrel{~}{f}_R(z).$$
(5.10)
This function satisfies the following equation
$$\left\{\frac{d}{dr}\mathrm{\Delta }_r\frac{d}{dr}\frac{2\alpha }{a^2}s^2F_s(r)m^2F_m(r)+\omega ^2F_\omega (r)+m\omega F_{m\omega }(r)\lambda _s\right\}\stackrel{~}{R}_s(r)=0,$$
(5.11)
where
$`F_s(r)`$ $`=`$ $`{\displaystyle \frac{\alpha }{a^2}}(r_+^{}r_{})(r_{}^{}r_{}){\displaystyle \frac{rr_+}{rr_{}}}+{\displaystyle \frac{\alpha }{a^2}}(r_++r_{})^2,`$
$`F_m(r)`$ $`=`$ $`{\displaystyle \frac{4a^4(1+\alpha )^2}{\alpha ^2(r_+r_{})^2(r_+^{}r_{})^2(r_{}^{}r_{})^2}}{\displaystyle \frac{(rr_{})^2}{(rr_+)(rr_+^{})(rr_{}^{})}}`$
$`\times \left\{\alpha (r_+r_{})(r_++r_{})^2+\left[(1\alpha )a^22\alpha r_{}^2\right](rr_+)\right\},`$
$`F_\omega (r)`$ $`=`$ $`{\displaystyle \frac{a^2(1+\alpha )^2}{\alpha ^3(r_+r_{})^2(r_+^{}r_{})^2(r_{}^{}r_{})^2}}{\displaystyle \frac{(rr_{})}{(rr_+)(rr_+^{})(rr_{}^{})}}`$
$`\times \{(r_+^2r_{}^2)^2[a^2(1+\alpha )\alpha (r_++r_{})^2]^2`$
$`+2(r_+r_{})[2(1\alpha )a^6+a^4((14\alpha +5\alpha ^2)r_+^24\alpha (12\alpha )r_+r_{}+(18\alpha +5\alpha ^2)r_{}^2)`$
$`2\alpha a^2\left((1\alpha )r_+^4+2(1\alpha )r_+^3r_{}+(37\alpha )r_+^2r_{}^28\alpha r_+r_{}^3+(15\alpha )r_{}^4\right)`$
$`+\alpha ^2(r_+^6+4r_+^5r_{}+9r_+^4r_{}^2+8r_+^3r_{}^3+5r_+^2r_{}^4+r_{}^6)](rr_+)`$
$`+[4\alpha (1\alpha )a^6+a^4((1\alpha )^2r_+^22(1\alpha )^2r_+r_{}+(1+6\alpha 15\alpha ^2)r_{}^2)`$
$`2\alpha a^2\left((1\alpha )r_+^44(1\alpha )r_+r_{}^3+(1+7\alpha )r_{}^4\right)`$
$`+\alpha ^2(r_+^6+2r_+^5r_{}+3r_+^4r_{}^24r_+^3r_{}^35r_+^2r_{}^46r_+r_{}^5+r_{}^6)](rr_+)^2\},`$
$`F_{m\omega }(r)`$ $`=`$ $`{\displaystyle \frac{4a^3(1+\alpha )^2}{\alpha ^3(r_+r_{})^2(r_+^{}r_{})^2(r_{}^{}r_{})^2}}{\displaystyle \frac{(rr_{})}{(rr_+)(rr_+^{})(rr_{}^{})}}`$ (5.12)
$`\times \{\alpha (r_+^2r_{}^2)^2[(1+\alpha )a^2+\alpha (r_+r_{})^2]`$
$`+(r_+r_{})[(1\alpha )a^44\alpha ^2(r_++r_{})^2r_{}^2`$
$`+\alpha a^2((13\alpha )(r_++r_{})^2+2(1+\alpha )r_{}^2)](rr_+)`$
$`2\alpha (a^2+r_{}^2)[a^2(1\alpha )2\alpha r_{}^2](rr_+)^2\},`$
Here we used Eq.(2.10) for rewriting the equation in terms of $`r`$. All the coefficients of this equation are real for any spin weight $`s`$ in contrast with the radial Teukolsky equation (2.9). We note that $`\stackrel{~}{R}_s(r)`$ behaves near $`rr_+`$ and $`r_+^{}`$ as
$`\stackrel{~}{R}_s(r)`$ $``$ $`(rr_+)^{a_1a_2},(rr_+)`$ (5.13)
$``$ $`(rr_+^{})^{a_1a_4}.(rr_+^{})`$
We construct the function $`\stackrel{~}{\mathrm{\Phi }}_s`$ from $`\stackrel{~}{S}_s(\theta )`$ and $`\stackrel{~}{R}_s(r)`$ as
$$\stackrel{~}{\mathrm{\Phi }}_s=e^{i(\omega tm\phi )}\stackrel{~}{R}_s(r)\stackrel{~}{S}_s(\theta ),$$
(5.14)
then this function satisfies
$`\{{\displaystyle \frac{}{r}}\mathrm{\Delta }_r{\displaystyle \frac{}{r}}+{\displaystyle \frac{1}{\mathrm{sin}\theta }}{\displaystyle \frac{}{\theta }}\mathrm{sin}\theta (1+\alpha \mathrm{cos}^2\theta ){\displaystyle \frac{}{\theta }}[F_\omega (r)(1+\alpha )^2a^2{\displaystyle \frac{\mathrm{sin}^2\theta }{1+\alpha \mathrm{cos}^2\theta }}]{\displaystyle \frac{^2}{t^2}}`$
$`+\left[F_m(r)+\alpha (1+\alpha ){\displaystyle \frac{(1\mathrm{cos}\theta )^2}{1+\alpha \mathrm{cos}^2\theta }}\right]{\displaystyle \frac{^2}{\phi ^2}}+\left[F_{m\omega }+2(1+\alpha )^2a{\displaystyle \frac{1\mathrm{cos}\theta }{1+\alpha \mathrm{cos}^2\theta }}\right]{\displaystyle \frac{^2}{t\phi }}`$
$`s^2[F_s(r)+(1+\alpha ){\displaystyle \frac{1+\mathrm{cos}\theta }{1+\alpha \mathrm{cos}^2\theta }}]2\alpha ({\displaystyle \frac{r^2}{a^2}}+\mathrm{cos}^2\theta )\}\stackrel{~}{\mathrm{\Phi }}_s=0.`$ (5.15)
All the coefficient in this equation are real and invariant under $`ss`$.
Finally we obtain a conserved energy integral from this equation in the following form:
$`{\displaystyle }drd\theta d\phi \mathrm{sin}\theta \left\{[F_\omega (r)(1+\alpha )^2a^2{\displaystyle \frac{\mathrm{sin}^2\theta }{1+\alpha \mathrm{cos}^2\theta }}]\right|{\displaystyle \frac{\stackrel{~}{\mathrm{\Phi }}_s}{t}}|^2+\mathrm{\Delta }_r|{\displaystyle \frac{\stackrel{~}{\mathrm{\Phi }}_s}{r}}|^2`$
$`+(1+\alpha \mathrm{cos}^2\theta )\left|{\displaystyle \frac{\stackrel{~}{\mathrm{\Phi }}_s}{\theta }}\right|^2+\left[F_m(r)+\alpha (1+\alpha ){\displaystyle \frac{(1\mathrm{cos}\theta )^2}{1+\alpha \mathrm{cos}^2\theta }}\right]\left|{\displaystyle \frac{\stackrel{~}{\mathrm{\Phi }}_s}{\phi }}\right|^2`$
$`+s^2[F_s(r)+(1+\alpha ){\displaystyle \frac{1+\mathrm{cos}\theta }{1\mathrm{cos}\theta }}]|\stackrel{~}{\mathrm{\Phi }}_s|^2+2\alpha ({\displaystyle \frac{r^2}{a^2}}+\mathrm{cos}^2\theta )|\stackrel{~}{\mathrm{\Phi }}_s|^2\},`$ (5.16)
where $`r`$ integration is performed over $`(r_+,r_+^{})`$. From Eq.(5.13), it can be understood that this integration is finite for unstable modes, provided that the cosmological constant is sufficiently small.
## 6 Summary and discussions
Solutions of the perturbation equations of massless fields in the Kerr-de Sitter geometries can be obtained by using Heun’s functions. In this paper, we constructed the differential and integral transformations of Heun’s function. The differential transformations of the angular and radial functions include the Teukolsky-Starobinsky identities as special cases. Although we don’t know generic way to study integral transformation of Heun’s function, we provided two types of integral transformations. In the Kerr limit, these differential and integral transformations coincide with those considered in the case of perturbations for the Kerr geometries . From the equation satisfied by the transform of the perturbation, we have succeeded in obtaining a conserved energy integral for unstable modes.
In the proof of mode stability of the Kerr black hole, it is crucial that the conserved quantity which is obtained by similar procedures to those used in this paper is positive definite. From the positivity, it is concluded that the value of the conserved quantity bounds the integral of $`\left|\frac{\mathrm{\Phi }_s}{t}\right|^2`$ and thus unstable modes cannot exist. However the conserved quantity obtained for the Kerr-de Sitter black hole is not positive definite for the sufficiently small ( but non-vanishing ) cosmological constant because the coefficient of $`\left|\frac{\mathrm{\Phi }_s}{\phi }\right|^2`$ in the quantity, which vanishes in the Kerr limit $`\mathrm{\Lambda }0`$ and the Schwarzschild-de Sitter limit $`a0`$, can become negative in the situation considered here. Therefore we cannot rule out possibility that there are unstable modes. We also point out that although $`\frac{}{t}`$ is globally null in the metric derived from the equation in the Kerr geometry which can be obtained from Eq.(5) in $`\alpha 0`$ limit, it does not hold in the metric derived from Eq.(5) with non-vanishing cosmological constant. However we think that the procedures performed here are natural extensions of those used in the Kerr case. It may be possible to improve the analysis which we provided here. To the end, we think that the systematic study of integral transformations of Heun’s function will be required.
The analyses given in our previous papers and here are applicable to the case of the Kerr-anti-de Sitter geometry similarly. We hope that those analyses may give deeper insight for the correspondence between quantum gravity in anti-de Sitter space and conformal field theory defined on the boundary .
Acknowledgments
I would like to thank E. Takasugi for discussions. I would like to thank H. Suzuki for discussions and careful reading of this manuscript. I would also like to thank the members of high energy theory group in Osaka University.
|
warning/0005/gr-qc0005104.html
|
ar5iv
|
text
|
# Quintessential Inflation with Dissipative Fluid
## I Introduction
A number of recent observations suggest that the $`\mathrm{\Omega }_m`$, the ratio of the matter density(baryonic+dark) to the critical density, is significantly less than unity suggesting that either the universe is open or that there is some other sources of this missing energy which makes $`\mathrm{\Omega }_{total}1`$. The recent findings of BOOMERANG experiments strongly suggests the second possibility of a flat universe. At the same time, the measurements of the luminosity-redshift relations observed for the 50 newly discovered type Ia supernova with redshift $`z>0.35`$ indicate that at present the universe is expanding in an accelerated fashion suggesting a net negative pressure for the universe.
Initial suggestions were to identify this missing energy density to a cosmological constant $`\mathrm{\Lambda }`$. For a flat matter dominated universe with $`\mathrm{\Lambda }`$ in Einstein gravity observations strongly suggest $`\mathrm{\Omega }_\mathrm{\Lambda }0.72`$. However, this possibility that $`\mathrm{\Lambda }`$ could be the dominant energy density has the drawback that the energy scale involved is lower than normal energy scale predicted by the most particle physics models by a factor of $`10^{123}`$. An alternative source of energy density that may be admissible for this acceleration could be a dynamical $`\mathrm{\Lambda }`$ in the form of a scalar field with some self interacting potential . If the energy density of this kind of source varies slowly with time, it mimics an effective cosmological constant. The idea of this candidate, called quintessence , is borrowed from the inflationary paradigm of the early universe. The difference, however, is that this new field evolves at a much lower energy scale. The energy density of this field, though dominant at present epoch, must remain sub-dominant at very early stages and should have evolved in such a way that it becomes comparable to the matter density $`\rho _m`$ now. When quintessence is modeled using a minimally coupled scalar field, in general, parameters need to be fine-tuned so as to ensure that $`\rho _m`$ and $`\rho _\varphi `$ are of the same order today. This fine tuning problem has been termed as the cosmic coincidence problem. A new form of quintessence field called “tracker field has been proposed to solve the cosmic coincidence problem. It has an equation of motion with an attractor like solution in the sense that for a wide range of initial conditions the equation of motion converges to the same solution.
There are a number of quintessence models which have been suggested and most of these involve scalar fields with minimal coupling with potentials dominating over the kinetic energy of the field. A purely exponential potential is one of the widely studied cases . Inspite of the several advantages the energy density is not enough to make up for the missing part. Inverse power law is another form of the potential (-) that has been considered extensively for quintessence models, in particular, for solving the cosmic coincidence problem. Though many of the problems are resolved successfully with this potential, the predicted value for the equation of state for the quintessence field, $`\gamma _Q`$, is not in good agreement with the observed results. In search of suitable models that would eliminate the problems, new types of potentials, like $`V_0[\mathrm{cos}h\lambda \varphi 1]^p`$ and $`V_0\mathrm{sin}h(\alpha \sqrt{k}_0\mathrm{\Delta }\varphi )^\beta `$ have been considered, which have asymptotic forms like the inverse power law or exponential ones. Different physical considerations have lead to the study of other types of the potentials also. Recently Saini et al have reconstructed the potential in context of general relativity and minimally coupled quintessence field from the expression of the luminosity distance $`d_L(z)`$ as function of redshift obtained from the observational data. However, none of these potentials are entirely free of problems. Hence, there is still a need to identify appropriate potentials to explain current observations . Also it has been recently shown by Pietro and Demaret that for constant scalar field equation of state, which is a good approximation for a tracker field solutions, the field equations and the conservation equations strongly constrain the scalar field potential. Most of the widely used potential for quintessence, such as inverse power law one, exponential or the cosine form, are incompatible with these constraints.
The CDM is in general considered to be a perfect fluid. However, in some scenarios, certain physical processes can make the CDM fluid effectively a dissipative one. In such a situation the fluid has an effective pressure that is negetive. Recently it has been proposed that the CDM must self interact in order to explain the detailed structure of the galactic halos . This self interaction will create a viscous pressure whose magnitude will depend on the mean free path of the CDM particles. In a recent work Chimento et.al have shown that a mixture of minimally coupled self interacting scalar field and a perfect fluid is unable to drive the accelerated expansion as well as solve the cosmic coincidence problem at the same time . However, a mixture of a dissipative CDM with bulk viscosity and a minimally coupled self interacting scalar field can successfully achieve both features simultaneously. Also, as demonstrated in a recent paper by Zimdahl et. al. one can also have a negative $`\pi `$ if there exists an interaction which does not conserve particle numbers. This may be due to the particle production out of gravitational field. In this case, the CDM is not a conventional dissipative fluid, but a perfect fluid with varying particle number. Substantial particle production is an event that occurs in the early universe. But Zimdahl et. al. have shown that even extremely small particle production rate can also cause the sufficiently negative $`\pi `$ to violate the strong energy condition.
In this paper we have used a minimally coupled scalar field with a self interacting potential together with a matter fluid having a dissipative pressure over and above its positive equilibrium pressure. We have not assumed any particular model for this negative pressure. Instead, we have investigated what kind effects it has in the expansion of the universe. Unlike other works in scalar field cosmology with a dissipative pressure we have neither assumed the behaviour of the scale factor nor have we assumed any specific form of the potential. Rather we have expressed all the variables in terms of what we call the ‘generating function’. For this we have followed the method described by Chimento et.al. with some additional assumptions. We have proceeded with a particular choice of the generating function for which the potential is constructed using a combination of different power-law functions of $`\varphi `$. From the behaviour of decelerating parameter it has been shown that one can indeed generate both inflationary era in the early time and also an late time accelerating phase with a decelerating period in between. We have also investigated the stability and attractor structure of the general solutions of the field equations with this kind of potential and have found that for certain choices of the constants the solutions indeed exhibit attractor behaviour in the late times.
## II Field Equations
Let us consider a homogeneous, isotropic, spatially flat FRW universe with a line element
$$ds^2=dt^2+a(t)^2\underset{i=1}{\overset{3}{}}dx_i^2,$$
(1)
where $`a(t)`$ is the scale factor. The energy density consists of a massive scalar field with a self-interacting potential $`V(\varphi )`$
$$S_{\mu \nu }=\varphi _{,\mu }\varphi _{,\nu }g_{\mu \nu }\left(\frac{1}{2}\varphi _{,\alpha }\varphi ^{,\alpha }+V(\varphi )\right),$$
(2)
together with a dissipative fluid having bulk viscosity as the only dissipative term:
$$T_{\mu \nu }=\rho u_\mu u_\nu +h_{\mu \nu }\left(p+\pi \right),$$
(3)
where $`\rho `$ is the energy density, $`p`$ is the equilibrium pressure, $`\pi `$ is the bulk viscous pressure, and $`h_{\mu \nu }=g_{\mu \nu }+u_\mu u_\nu `$ is the projection tensor.
The independent field equations describing the system are
$`\ddot{\varphi }+3H\dot{\varphi }+{\displaystyle \frac{dV}{d\varphi }}=0,`$ (4)
$`3H^2={\displaystyle \frac{1}{2}}\dot{\varphi }^2+V+\rho ,`$ (5)
$`\dot{\rho }+3H\left(\rho +p+\pi \right)=0`$ (6)
After some straightforward calculations one can construct the equation
$$2\dot{H}=\dot{\varphi }^2\left(\rho +p+\pi \right).$$
(7)
For $`\dot{H}0`$, the system consists of three independent equations, viz., (6), (5) and (7) and we have six unknowns, $`H,\varphi ,V(\varphi ),\rho ,p`$ and $`\pi `$. Hence, three constraint equations are required in order to have a closed system of equations. We first assume that $`p`$,$`\rho `$ and $`\pi `$ are related to $`\dot{\varphi }`$ by,
$$\rho +p+\pi =m\dot{\varphi }^2,$$
(8)
where $`m`$ is an arbitrary constant. We emphasize that this choice is considered purely because it makes the system of equations simple to solve. We use this to see whether such a choice leads to any physically acceptable solutions.
The nature of the potential determines the evolution of $`\varphi `$. However, instead of choosing a particular form for $`V(\varphi )`$, we can alternatively describe the evolution of the scalar field by expressing $`\dot{\varphi }`$ as a function of $`\varphi `$ as,
$$\dot{\varphi }=F(\varphi )$$
(9)
where $`F(\varphi )`$ is called the generating function. In this approach, we start with a specific trajectory, $`\dot{\varphi }=F(\varphi )`$, in phase space for $`\varphi `$ and determine the potential that evolves the scalar field in that manner. However, choosing a potential $`V(\varphi )`$ or choosing a generating function, $`F(\varphi )`$ is not completely equivalent. Choosing a form for $`F(\varphi )`$ is more restrictive than choosing a potential due to the following reason. A specific $`F(\varphi )`$ refers to a specific class of initial conditions. Hence the set of solutions represented by a specific form for $`F(\varphi )`$ is a subset of the full set of solutions for the corresponding potential. Having constructed the potential function, we study the evolution of the system in phase space for a more general initial condition, i.e. ones which are not restricted by any specific form for $`F(\varphi )`$.
Using equation (8), equation (7) and equation (9) one can write,
$$2\frac{dH}{d\varphi }=(m+1)F(\varphi ).$$
(10)
In this paper, we first choose $`F(\varphi )`$ and then calculate the $`V(\varphi )`$ that is consistent with our choice of $`F(\varphi )`$ as well as with the Einstein’s equations. The solutions for which the phase space trajectory is given by $`\dot{\varphi }=F(\varphi )`$ is a subset of the of the general class of solutions for this potential. (Later in subsection III C we will draw the phase space trajectory for a set of general solutions (i.e. with different initial conditions) including the one for which $`\dot{\varphi }=F(\varphi )`$) and check the conditions for this to be a stable solution.)
For any given $`F(\varphi )`$ which we term as the “generating function”, one can, in principle, solve the system as follows:
$`H(\varphi )`$ $`=`$ $`{\displaystyle \frac{(m+1)}{2}}{\displaystyle F(\varphi )𝑑\varphi }+H_0,`$ (11)
$`\rho (\varphi )`$ $`=`$ $`3m{\displaystyle H(\varphi )F(\varphi )𝑑\varphi }+\rho _0,`$ (12)
$`a(\varphi )`$ $`=`$ $`a_0\mathrm{exp}\left[{\displaystyle \frac{H(\varphi )}{F(\varphi )}𝑑\varphi }\right]`$ (13)
$`V(\varphi )`$ $`=`$ $`3H^2(\varphi ){\displaystyle \frac{1}{2}}F(\varphi )^2\rho (\varphi ),`$ (14)
$`p+\pi `$ $`=`$ $`mF(\varphi )^2\rho (\varphi ).`$ (15)
where $`H_0,\rho _0`$ and $`a_0`$ are integration constant.With the assumption of an equation of state $`p=p(\rho )`$, one can also calculate $`\pi `$. Hence, our main aim is to choose $`F(\varphi )`$ properly to have some physically acceptable behaviour for different variables.
## III Solutions for $`F(\varphi )=\omega \varphi `$
### A Exact solutions
With this choice of the generating function, the differential equation governing the time evolution of $`\varphi `$ is
$$\dot{\varphi }=\omega \varphi .$$
(16)
In this case, the exact solutions turn out to be,
$`\varphi (t)`$ $`=`$ $`\varphi _0\mathrm{exp}(\omega t),`$ (17)
$`H(t)`$ $`=`$ $`{\displaystyle \frac{(m+1)}{4}}\omega \varphi _0^2\mathrm{exp}(2\omega t)+H_0`$ (18)
$`a(t)`$ $`=`$ $`a_0\mathrm{exp}\left[{\displaystyle \frac{\varphi _0^2(m+1)}{8}}\mathrm{exp}(2\omega t)+H_ot\right],`$ (19)
$`q(t)`$ $`=`$ $`{\displaystyle \frac{(m+1)}{2}}\left[{\displaystyle \frac{\omega ^2\varphi _0^2\mathrm{exp}(2\omega t)}{\left(H_0\frac{(m+1)\varphi _0^2\omega \mathrm{exp}(2\omega t)}{4}\right)^2}}\right]1,`$ (20)
$`\rho (t)`$ $`=`$ $`{\displaystyle \frac{3(m+1)m\omega ^2\varphi _0^4\mathrm{exp}(4\omega t)}{16}}`$ (21)
$``$ $`{\displaystyle \frac{3m\omega H_0\varphi _0^2\mathrm{exp}(2\omega t)}{2}}+\rho _0,`$ (22)
$`V(\varphi )`$ $`=`$ $`(3H_0^2\rho _0)+{\displaystyle \frac{3(m+1)\omega ^2\varphi ^4}{16}}{\displaystyle \frac{\omega \varphi ^2}{2}}\left(\omega +3H_0\right).`$ (23)
For an expanding Universe we must have $`H>0`$. To ensure this, either $`m<1`$ or $`\omega <0`$. We choose $`\omega <0`$. It may be noted from equation (19) that the proper volume becomes zero at $`t\mathrm{}`$ and hence this is taken to be the initial time for our model. The potential $`V(\varphi )`$ given in equation (22) is the standard renormalizable tree level potential arising in the perturabative regime of quantum field theory. With $`\omega <0`$, the shape of the potential depends on the factor ($`\omega +3H_0`$). When $`H_0>\omega /3`$, it is the most simplified version of the potential for the hybrid inflation . Similarly when $`H_0<\omega /3`$, it is the inverted quadratic potential and has been discussed by many authors for inflationary models (see and references therein). As has been pointed out before, the solution represented by equation (17) is only one of the classes solutions (corresponding to the potential given in equation (23))for which the the initial values of $`\dot{\varphi }`$ and $`\varphi `$ are related by $`\dot{\varphi }_{initial}=\omega \varphi _{initial}`$. Expressing the equation of state as, $`p=(\gamma 1)\rho `$, where $`1\gamma 2`$, the bulk viscous pressure is given by,
$`\pi `$ $`=`$ $`m\omega \varphi _0^2\mathrm{exp}(2\omega t)\left(\omega +{\displaystyle \frac{3\gamma H_0}{2}}\right)`$ (24)
$``$ $`{\displaystyle \frac{3\gamma m(m+1)}{16}}\omega ^2\varphi _0^4\mathrm{exp}(4\omega t)\gamma \rho _0.`$ (25)
### B Behaviour of the solutions
The behaviour of the deceleration parameter is shown in figure 1. The evolution of $`q(t)`$ has the basic feature that we require for an acceptable form for the deceleration parameter. $`q(t)`$ is negative to begin with resulting in an inflationary phase. It increases and subsequently $`q(t)`$ becomes positive. At a later time, it drops below $`0`$ and finally saturates at a constant value below zero. This last feature of $`q(t)`$ at late times when it becomes a negative constant produces the present day accelerated expansion. In fact it asymptotically becomes a de Sitter Universe.
The behaviour of the equation of state for the scalar field $`\mathrm{w}=(\frac{1}{2}\dot{\varphi }^2V(\varphi ))/(\frac{1}{2}\dot{\varphi }^2+V(\varphi ))`$ in figure (2) shows that at late times $`\mathrm{w}=1`$ which essentially supports the existence of a cosmological constant in late time.
An acceptable model should also satisfy constraints arising from cosmological nucleosynthesis and from structure formation. In order to keep these intact, the matter energy density should dominate over the energy density of the scalar field in the early universe. However, at late times the scalar field energy density should be more than that due to matter so that one can explain the missing energy of the universe. In figure (3) we have plotted the energy density for the matter $`(\rho )`$ and for the scalar field $`(\rho _\varphi )`$. We see the above constraints are satisfied. The matter energy density is dominant in the early time and so nucleosynthesis and structure formation are unaffected. But at late time $`\rho _\varphi `$ decreases more slowly than $`\rho `$ and hence ultimately it becomes greater than $`\rho `$ This explains the missing energy associated with accelerated expansion of the universe. It can also be seen that although in early period the energy densities are of different order of magnitude but in late time they are of the same order. This gives a dynamical solution to the cosmic coincidence problem. Also at late times the ratio of these two energy densities becomes constant showing the tracking behaviour. We have also plotted the two density parameter $`\mathrm{\Omega }_m`$ and $`\mathrm{\Omega }_\varphi `$ in figure 4.
The variation of the viscous pressure $`\pi `$ is shown in 5.
### C Stability and attractor structure
The solutions we discussed were those for which $`F(\varphi )=\omega \varphi `$. In other words, for these set of solutions the initial conditions are constrained by the relation $`\dot{\varphi }_i=\omega \varphi _i`$, where $`\varphi _i`$ and $`\dot{\varphi }_i`$ are the initial values of $`\varphi `$ and $`\dot{\varphi }`$, respectively. These solutions are significant only if these are stable, i.e., only if more general initial conditions asymptotically evolve towards these phase space trajectories.
This solution corresponds to the case when the expression for the Hubble parameter ($`H(t)`$), scale factor ($`a(t)`$), deceleration parameter ($`q(t)`$) and the density ($`\rho (t)`$) given by equations (18), (19), (20) and (22). In this section we proceed to investigate whether or not the solutions described are stable and whether there are any constraints on the parameters for this.
The general equation of motion for the potential given in equation (23) is
$$\ddot{\varphi }=3H\dot{\varphi }\frac{3(m+1)}{4}\omega ^2\varphi ^3+\omega ^2\varphi +3H_0\omega \varphi $$
(26)
Together with the equation for the evolution of $`H(t)`$, we can write the following set of three coupled differential equations,
$`\dot{\varphi }`$ $`=`$ $`x`$ (27)
$`\dot{x}`$ $`=`$ $`3Hx{\displaystyle \frac{3(m+1)}{4}}\omega ^2\varphi ^3+\omega ^2\varphi +3H_0\omega \varphi `$ (28)
$`\dot{H}`$ $`=`$ $`{\displaystyle \frac{(m+1)}{2}}x^2`$ (29)
By equating the right hand sides of the equations to $`0`$ we obtain the following three critical points in the ($`\varphi ,\dot{\varphi }`$) plane.
$`(0,0),(\sqrt{{\displaystyle \frac{4(\omega +3H_0)}{3(m+1)\omega }}},0),(\sqrt{{\displaystyle \frac{4(\omega +3H_0)}{3(m+1)\omega }}},0).`$ (31)
The position of the critical points in the phase diagram changes with the value of $`H_0`$. Further, at $`\omega =3H_0`$ all the three critical points merge. There is a transition in the nature of stability of the trajectories in $`(\varphi ,\dot{\varphi })`$ space at this value of $`H_0`$. We discuss below the nature of the trajectories for different values of $`H_0`$. First of all, we should note that $`\omega <0`$.
In figures (6) and (7), we have plotted the phase space trajectories for the scalar field with a variety of initial conditions that are not constrained by the particular choice of the generating function. Figure 6 corresponds to the case of $`H_0=0.2`$. As we have chosen $`\omega `$ to be $`1`$, this is a case in which the potential has a local maximum. On the other hand, figure 7 corresponds to the case of $`H_0=0.4`$ in which $`\varphi =0`$ is a minimum for the potential.
In the first case the phase space trajectories converge to either of the two non-zero values of $`\varphi `$ and is not an attractor like solution as the asymptotic nature of the solution depends on the initial conditions. But in the second case, however, the trajectories converge to $`\varphi =0`$, $`\dot{\varphi }=0`$ independent of the initial conditions and ,the de-Sitter solution is an attractor for this case.
## IV Conclusion
The most important conclusion is that models with a self-interacting scalar field and a matter fluid having a negative pressure in addition to its positive equilibrium pressure can produce a scenario for the cosmological evolution in which one can have an inflationary phase to begin with, an accelerated phase at late times (like the present era) and a decelerating phase in-between. Recently Lopez and Matos have shown that this kind of complete history for the scale factor can be described by a hyperbolic potential. But the physical origin of such potential is still not well known. But here we have shown that such kind of behaviour for the scale factor can be generated with a potential given in (22),which has been widely used by many authors for inflationary models.
The behaviour of $`\rho `$ and $`\rho _\varphi `$ in our model shows that although in early universe, $`\rho `$ is greater than $`\rho _\varphi `$ which is necessary for different physical phenomena like nucleosynthesis and structure formation etc, in the late times, $`\rho _\varphi `$ starts dominating. This feature explains the missing energy density and also the ration of two energy densities becomes a constant in late time showing the “tracking nature”.
One should note that both the assumptions (8) and (16) play crucial role in our model. Given a barotropic equation of state between $`\rho `$ and $`p`$ one can not assume (8) and (16) at the same time if the dissipative pressure $`\pi `$ is zero in our model as t hat will lead to an over-determined problem ( the number of unknowns will be less than the number of independent equations). Even if one assumes these two condition one can check that will lead to an negative equilibrium pressure which is not desirable. H ence the existence of dissipative pressure also plays an important role in our model.
We have also studied the general equation of motion for the scalar field (equation (24)) for the potential (22) and have shown that for the choices of constant for which the potential is minimum at $`\varphi =0`$, the phase space diagram exhibit a attractor behaviour towards the asymptotic de-sitter solution.
We want to mention that previously tracker and attractor solutions have been studied for scalar fields having inverse power law, exponential, cosine potential. But in all of these cases the equation of state $`w`$ is a constant in radiation era as well as in a matter dominated era. It was later shown by Pietro and Demaret that these kind of potential with a constant $`w`$ is not consistent with the field equations. In our case, the equation of state for the scalar field $`w`$ is not a constant but it varies with the cosmic evolution and approaches towards -1 asymptotically showing the existence of a cosmological constant in late times.
|
warning/0005/math0005285.html
|
ar5iv
|
text
|
# The Dirac operator of a commuting 𝑑-tuple
## Introduction
We introduce an abstract notion of Dirac operator in complex dimension $`d=1,2,\mathrm{}`$ and we show that this theory of Dirac operators actually coincides with the theory of commuting $`d`$-tuples of operators on a common Hilbert space $`H`$. The homology and cohomology of Dirac operators is discussed in general terms, and we relate the homological picture to classical spectral theory by describing its application to concrete problems involving the solution of linear equations of the form
$$T_1x_1+T_2x_2+\mathrm{}+T_dx_d=y$$
given $`y`$ and several commuting operators $`T_1,T_2,\mathrm{},T_d`$.
These developments grew out of an attempt to understand the stability properties of a curvature invariant introduced in a previous paper (see , ), and to find an appropriate formula that expresses the curvature invariant as the index of some operator. The results are presented in section 4 (see Theorem B and its corollary).
While there is a large literature concerning Taylor’s cohomological notion of joint spectrum for commuting sets of operators on a Banach space, less attention has been devoted to the Dirac operator that emerges naturally in the context of Hilbert spaces (however, see sections 4 through 6 of , where the operator $`B+B^{}`$ is explicitly related to Taylor invertibility and the Fredholm property). We have made no attempt to compile a comprehensive list of references concerning the Taylor spectrum, but we do call the reader’s attention to work of Albrecht , Curto ,, McIntosh and Pryde , Putinar ,, and Vasilescu ,. A more extensive list of references can be found in the survey . Finally, I want to thank Stephen Parrott for useful remarks based on a draft of this paper, and Hendrik Lenstra for patiently enlightening me on homological issues.
## 1. Preliminaries: Clifford structures and the CARs in dimension $`d`$
Since there is significant variation in the notation commonly used for Clifford algebras and CAR algebras, we begin with explicit statements of notation and terminology as it will be used below.
Let $`H`$ be a complex Hilbert space and let $`d`$ be a positive integer. By a Clifford structure on $`H`$ (of real dimension $`2d`$) we mean a real-linear mapping $`R:^d(H)`$ of the $`2d`$-dimensional real vector space $`^d`$ into the space of self adjoint operators on $`H`$ which satisfies
$$R(z)^2=z^2\mathrm{𝟙},z^d,$$
$`1.1`$
where for a $`d`$ tuple $`z=(z_1,\mathrm{},z_d)`$ of complex numbers, $`z`$ denotes the Euclidean norm
$$z^2=|z_1|^2+\mathrm{}+|z_d|^2.$$
Clifford structures can also be defined as real-linear maps $`R^{}`$ of $`^d`$ into the space of skew-adjoint operators on $`H`$ which satisfy $`R^{}(z)^2=z^2\mathrm{𝟙}`$, and perhaps this is a more common formulation. Note however that such a structure corresponds to a Clifford structure $`R`$ satisfying (1.1) by way of $`R^{}(z)=iR(z)`$.
Letting $`e_1=(1,0,\mathrm{},0),\mathrm{},e_d=(0,\mathrm{},0,1)`$ be the usual unit vectors in $`^d`$ we define operators $`p_1,\mathrm{},p_d,q_1,\mathrm{},q_d(H)`$ by $`p_k=R(e_k)`$, $`q_k=R(ie_k)`$, $`k=1,\mathrm{},d`$. The $`2d`$ operators $`(r_1,\mathrm{},r_{2d})=(p_1,\mathrm{},p_d,q_1,\mathrm{},q_d)`$ are self adjoint, they satisfy
$$r_kr_j+r_jr_k=2\delta _{jk}\mathrm{𝟙},1j,k2d,$$
$`1.2`$
and the complex algebra they generate is a $`C^{}`$-algebra isomorphic to $`M_{2^d}()`$.
While Clifford structures are real-linear maps of $`^d`$ there is an obvious way to complexify them, and once that is done one obtains a (complex-linear) representation of the canonical anticommutation relations. This sets up a bijective correspondence between Clifford structures and reprsentations of the anticommutation relations. The details are as follows. Since the $`2d`$-dimensional real vector space $`^d`$ comes with an a priori complex structure, any real-linear mapping $`R`$ of $`^d`$ into the self adjoint operators of $`(H)`$ is the real part of a unique complex-linear mapping $`C:^d(H)`$ in the sense that
$$R(z)=C(z)+C(z)^{},z^d,$$
$`1.3`$
and $`C`$ is given by $`C(z)=\frac{1}{2}(R(z)iR(iz))`$, $`z^d`$. Corresponding to (1.2) one finds that the operators $`c_k=C(e_k)=\frac{1}{2}(p_kiq_k)`$, $`1kd`$ satisfy the canonical anticommutation relations
$`c_kc_j+c_jc_k`$ $`=0`$ $`1.4`$
$`c_k^{}c_j+c_jc_k^{}`$ $`=\delta _{jk}\mathrm{𝟙}.`$
Equivalently, the complex linear map $`C:^d(H)`$ satisfies
$`C(z)C(w)+C(w)C(z)`$ $`=0,`$ $`1.5`$
$`C(w)^{}C(z)+C(z)C(w)^{}`$ $`=z,w\mathrm{𝟙}`$
for $`z,w^d`$, $`z,w`$ denoting the Hermitian inner product
$$z,w=z_1\overline{w}_1+\mathrm{}+z_d\overline{w}_d.$$
The $``$-algebra generated by the operators $`C(z)`$ contains the identity and is isomorphic to the matrix algebra $`M_{2^d}()`$.
Any two irreducible representations of the CAR algebra (in either of its presentations (1.4) or (1.5)) are unitarily equivalent. The standard irreducible representation of the CAR algebra is defined as follows. Let $`Z`$ be a complex Hilbert space of finite dimension $`d`$, and let $`\mathrm{\Lambda }Z`$ be the exterior algebra over $`Z`$,
$$\mathrm{\Lambda }Z=\mathrm{\Lambda }^0Z\mathrm{\Lambda }^1Z\mathrm{\Lambda }^2Z\mathrm{}\mathrm{\Lambda }^dZ$$
where $`\mathrm{\Lambda }^kZ`$ denotes the $`k`$th exterior power of $`Z`$. By definition, $`\mathrm{\Lambda }^0Z=`$, and the last summand $`\mathrm{\Lambda }^dZ`$ is also isomorphic to $``$. $`\mathrm{\Lambda }^kZ`$ is spanned by vectors of the form $`z_1z_2\mathrm{}z_k`$, $`z_kZ`$, and the natural inner product on $`\mathrm{\Lambda }^kZ`$ satisfies
$$z_1\mathrm{}z_k,w_1\mathrm{}w_k=det(z_i,w_j),$$
the right side denoting the determinant of the $`k\times k`$ matrix of inner products $`a_{ij}=z_i,w_j`$. $`\mathrm{\Lambda }Z`$ is a direct sum of the (complex) Hilbert spaces $`\mathrm{\Lambda }^kZ`$, and it is a Hilbert space of complex dimension $`2^d`$.
For $`zZ`$, the creation operator $`C(z)`$ maps $`\mathrm{\Lambda }^kZ`$ to $`\mathrm{\Lambda }^{k+1}Z`$, and acts on the generators as follows
$$C(z):x_1\mathrm{}x_kzx_1\mathrm{}x_k.$$
$`C:Z(\mathrm{\Lambda }Z)`$ is an irreducible representation of the canonical anticommutation relations (1.5). One obtains the standard irreducible Clifford structure (1.1) by taking the real part of this representation $`R(z)=C(z)+C(z)^{}`$.
###### Remark Remarks
In the next section we will define Dirac operators in terms of Clifford structures. Because of the correspondence cited above, we could just as well have formulated this notion in terms of the anticommutation relations, avoiding Clifford structures entirely. We have chosen to use them because Clifford algebras are associated with the Dirac operators of Riemannian geometry (and perhaps also for reasons of taste, the single equation (1.1) being twice as elegant as the two equations of (1.5)). On the other hand, we have found that proofs seem to go more smoothly with the anticommutation relations (1.5). The preceding observations show that nothing is lost in passing back and forth as needed.
We also want to emphasize that with any representation of either the Clifford relations (1.1) or the anticommutation relations (1.5) on a Hilbert space there are additional objects that are naturally associated with them, namely a gauge group, a number operator, and a $`_2`$-grading of $`H`$. By a $`_2`$-grading of a Hilbert space $`H`$ we simply mean a decomposition of $`H`$ into two mutually orthogonal subspaces
$$H=H_+H_{}.$$
Vectors in $`H_+`$ (resp. $`H_{}`$) are called even (resp. odd). An operator $`A(H)`$ is said to be of odd degree if $`AH_+H_{}`$ and $`AH_{}H_+`$, and the set of all such $`A`$ is a self-adjoint linear subspace of $`(H)`$.
###### Proposition A
Let $`R:^d(H)`$ be a Clifford structure (1.1), and let $`𝒜`$ be the finite dimensional $`C^{}`$-algebra generated by the range of $`R`$. There is a unique strongly continuous unitary representation $`\mathrm{\Gamma }`$ of the circle group $`𝕋`$ on $`H`$ satisfying
$`\mathrm{\Gamma }(𝕋)`$ $`𝒜`$
$`\mathrm{\Gamma }(\lambda )R(z)\mathrm{\Gamma }(\lambda )^{}`$ $`=R(\lambda z),\lambda 𝕋,z^d,`$
and such that the spectrum of $`\mathrm{\Gamma }`$ starts at $`0`$ in the sense that the spectral subspaces
$$H_n=\{\xi H:\mathrm{\Gamma }(\lambda )\xi =\lambda ^n\xi \text{ for all }\lambda 𝕋\},n$$
satisfy $`H_n=\{0\}`$ for negative $`n`$ and $`H_0\{0\}`$.
The number operator $`N`$ is defined as the generator of the gauge group
$$\mathrm{\Gamma }(e^{it})=e^{itN},t,$$
and is a self adjoint element of $`𝒜`$ having spectrum $`\{0,1,2,\mathrm{},d\}`$. The $`_2`$-grading of $`H`$ is defined by
$$H_+=\underset{n\text{ even}}{}H_n,H_{}=\underset{n\text{ odd}}{}H_n.$$
###### Demonstration proof
This is a reformulation of standard results that are perhaps most familiar when formulated in terms of the anticommutation relations. One may check the validity of the proposition explicitly for the irreducible representation on $`\mathrm{\Lambda }^d`$ described above. Since every Clifford structure is unitarily equivalent to a direct sum of copies of this irreducible one, Proposition A persists in the general case.
###### Remark Remark 1.6
One can single out these objects most explicitly in terms of the anticommutation relations $`C:Z(H)`$ (1.5) over any $`d`$-dimensional one-particle space $`Z`$. Here, $`𝒜`$ is the $`C^{}`$-algebra generated by $`C(Z)`$ and $`\mathrm{\Gamma }`$ should satisfy $`\mathrm{\Gamma }(\lambda )C(z)\mathrm{\Gamma }(\lambda )^{}=\lambda C(z)`$ for $`zZ`$, $`\lambda 𝕋`$, along with the two requirements that (1) the spectrum of $`\mathrm{\Gamma }`$ should start at $`0`$ and (2) the gauge automorphisms of $`(H)`$ should be inner in the sense that $`\mathrm{\Gamma }(𝕋)𝒜`$. The number operator and gauge group are given by
$$N=C(e_1)C(e_1)^{}+\mathrm{}+C(e_d)C(e_d)^{},\mathrm{\Gamma }(e^{it})=e^{itN},t$$
$`e_1,\mathrm{},e_d`$ being any orthonormal basis for the complex Hilbert space $`Z`$. The $`_2`$ grading is defined by the spectral subspaces of $`\mathrm{\Gamma }`$ (or equivalently, of $`N`$) as in Proposition A.
## 2. Dirac operators and Taylor invertibility
A Dirac operator is a self-adjoint operator $`D`$ acting on a Hilbert space $`H`$ that has been endowed with a distinguished Clifford structure (1.1), satisfying three additional conditions. In order to keep the bookkeeping explicit, we include the Clifford structure as part of the definition.
###### Definition
A Dirac operator of dimension $`d`$ is a pair $`(D,R)`$ consisting of a bounded self-adjoint operator $`D`$ acting on a Hilbert space $`H`$ and a Clifford structure $`R:^d(H)`$, satisfying
where $`\mathrm{\Gamma }:𝕋(H)`$ is the gauge group associated with $`R`$, and $`𝒜`$ is the $`C^{}`$-algebra generated by the range of $`R`$.
###### Remark Remarks
Let $`H=H_+H_{}`$ be the $`_2`$-grading of $`H`$ induced by the gauge group. (D1) is equivalent to requiring that $`DH_+H_{}`$ and $`DH_{}H_+`$, i.e., that $`D`$ should be an operator of odd degree. (D2) implies that the “Laplacian” $`D^2`$ associated with $`D`$ should be invariant under the action of the gauge group as automorphisms of $`(H)`$. (D3) asserts that the “partial derivatives” of $`D`$ must commute with the operators in $`R(^d)`$.
We have already pointed out that Clifford structures are interchangeable with representations $`C`$ of the anticommutation relations (1.5). In terms of $`C`$, the definition of Dirac operator would be similar except that (D3) would be replaced with the following: $`C(z)D+DC(z)𝒜^{}`$, for every $`z^d`$.
There is a natural notion of isomorphism for Dirac operators, namely $`(D,R)`$ (acting on $`H`$) is isomorphic to $`(D^{},R^{})`$ (acting on $`H^{}`$) if there is a unitary operator $`U:HH^{}`$ such that $`UD=D^{}U`$ and $`UR(z)=R^{}(z)U`$ for every $`z^d`$. Notice that the spectrum and multiplicity function of a Dirac operator are invariant under isomorphism, but of course the notion of isomorphism involves more than simple unitary equivalence of the operators $`D`$ and $`D^{}`$.
We first show how to construct a Dirac operator, starting with a multioperator $`(T_1,\mathrm{},T_d)`$. Let $`T_1,\mathrm{},T_d(H)`$ be a commuting $`d`$-tuple of bounded operators, let $`Z`$ be a $`d`$-dimensional Hilbert space (which may be thought of as $`^d`$), and let $`C_0:Z\mathrm{\Lambda }Z`$ be the irreducible representation of the anticommutation relations (1.5) that was described in section 1.
Consider the Hilbert space $`\stackrel{~}{H}=H\mathrm{\Lambda }Z`$ and let $`C(z)=\mathrm{𝟙}_HC_0(z)`$, $`zZ`$. $`C`$ obviously satisfies (1.5). Fix any orthonormal basis $`e_1,\mathrm{},e_d`$ for $`Z`$ and define an operator $`B`$ on $`\stackrel{~}{H}`$ as follows
$$B=T_1C_0(e_1)+\mathrm{}+T_dC_0(e_d).$$
The pair $`(D,R)`$ is defined as follows
$$D=B+B^{},R(z)=C(z)+C(z)^{},zZ.$$
$`2.1`$
If we use the orthonormal basis to identify $`Z`$ with $`^d`$, the discussion of section 1 shows that $`R`$ satisfies (1.1).
###### Proposition
$`(D,R)`$ is a Dirac operator on $`\stackrel{~}{H}`$. For $`\lambda =(\lambda _1,\mathrm{},\lambda _d)^d`$, the Dirac operator of the translated $`d`$-tuple $`(T_1\lambda _1\mathrm{𝟙},\mathrm{},T_d\lambda _d\mathrm{𝟙})`$ is $`(D_\lambda ,R)`$, where $`D_\lambda =DR(\lambda )`$.
###### Demonstration proof
Noting that the gauge group $`\mathrm{\Gamma }`$ is related to $`B`$ by way of
$$\mathrm{\Gamma }(\lambda )B\mathrm{\Gamma }(\lambda )^{}=\lambda B,\lambda 𝕋,$$
$`2.2`$
we find that
$$\mathrm{\Gamma }(\lambda )D\mathrm{\Gamma }(\lambda )^{}=\lambda B+\overline{\lambda }B^{}$$
from which (D1) follows. (D3) follows after a straightforward computation using the anticommutation relations (1.4). In order to check (D2), notice first that $`B^2=0`$. Indeed, one has
$$B^2=\underset{i,j=1}{\overset{d}{}}T_iT_jC_0(e_i)C_0(e_j).$$
Since $`T_iT_j=T_jT_i`$ whereas $`C_0(e_i)C_0(e_j)=C_0(e_j)C_0(e_i)`$, this sum must vanish.
It follows that $`D^2=B^{}B+BB^{}`$. By (2.2), both $`BB^{}`$ and $`B^{}B`$ commute with the gauge group, hence so does $`D^2`$. The last sentence is immediate from (2.1).
###### Remark Remark
A routine verification shows that the isomorphism class of this Dirac operator $`(D,R)`$ does not depend on the choice of orthonormal basis, and depends only on the commuting $`d`$-tuple $`\overline{T}=(T_1,\mathrm{},T_d)`$. For this reason we sometimes write $`D_{\overline{T}}`$ rather than $`(D,R)`$, for the Dirac operator constructed from a multioperator $`\overline{T}`$.
###### Remark Comments on homology, cohomology and the Taylor spectrum
Joseph Taylor introduced a notion of invertibility (and therefore joint spectrum) for commuting $`d`$-tuples of operators $`T_1,\mathrm{},T_d`$ acting on a complex Banach space. Taylor’s notion of invertibility can be formulated as follows. Let
$$\stackrel{~}{H}=\stackrel{~}{H}_0\stackrel{~}{H}_1\mathrm{}\stackrel{~}{H}_d$$
be the natural decomposition of $`\stackrel{~}{H}=H\mathrm{\Lambda }Z`$ induced by the decomposition of the exterior algebra $`\mathrm{\Lambda }Z`$ into homogeneous forms of degree $`k=0,1,\mathrm{},d`$
$$\stackrel{~}{H}_k=H\mathrm{\Lambda }^kZ.$$
The operator $`B=T_1c_1+\mathrm{}+T_dc_d`$ of formula (1.6) satisfies
$$B\stackrel{~}{H}_k\stackrel{~}{H}_{k+1}$$
and as we have already pointed out, $`B^2=0`$. Thus, the pair $`\stackrel{~}{H},B`$ defines a complex (the Koszul complex of the $`[z_1,\mathrm{},z_d]`$-module $`H`$), and when the range of $`B`$ is closed and of finite codimension in $`\mathrm{ker}B`$, we can define the cohomology of this complex. Taylor defines the underlying $`d`$-tuple to be invertible if the cohomology is trivial: $`B\stackrel{~}{H}=\mathrm{ker}B`$. As we will see presently, for Hilbert spaces invertibility becomes a concrete property of the Dirac operator: a $`d`$-tuple of commuting operators on $`H`$ is Taylor-invertible if and only if its Dirac operator $`D`$ is invertible in $`(H\mathrm{\Lambda }^d)`$.
The Taylor spectrum of a commuting $`d`$-tuple $`\overline{T}=(T_1,\mathrm{},T_d)`$ is defined as the set of all complex $`d`$-tuples $`\lambda =(\lambda _1,\mathrm{},\lambda _d)^d`$ with the property that the translated $`d`$-tuple
$$(T_1\lambda _1\mathrm{𝟙},\mathrm{},T_d\lambda _d\mathrm{𝟙})$$
is not invertible. In terms of the Dirac operator $`(D,R)`$ of $`\overline{T}`$, this is the set of all $`\lambda ^d`$ such that $`DR(\lambda )`$ is not invertible. The relation between this “Clifford spectrum” and the ordinary spectrum of $`D`$ is not very well understood.
The Taylor spectrum and Taylor’s notion of invertibility are important not only because they lead to the “right” theorems about the spectrum in multivariable operator theory (see ), but also and perhaps more significantly, because they embody the correct multivariable generalization of classical spectral theory as it is defined in terms of solving linear equations.
In order to discuss the latter it is necessary to cast Taylor’s cohomological picture of the joint spectrum into a homological picture; once that is done, a clear interpretation of the Taylor spectrum will emerge in terms of solving linear equations. In more detail, consider the canonical anticommutation relations in the form (1.4) and let $`c_1,\mathrm{},c_d`$ be the irreducible representation described in section 1, where $`c_i`$ acts as follows on the generators of $`\mathrm{\Lambda }^k^d`$
$$c_i:z_1\mathrm{}z_ke_iz_1\mathrm{}z_k,$$
$`e_1,\mathrm{},e_d`$ denoting an orthonormal basis for $`^d`$. Starting with a commuting $`d`$-tuple $`T_1,\mathrm{},T_d(H)`$, we have defined a cohomological boundary operator on $`H\mathrm{\Lambda }^d`$ by
$$B=T_1c_1+\mathrm{}+T_dc_d.$$
Instead, let us consider the homological boundary operator
$$\stackrel{~}{B}=T_1c_1^{}+\mathrm{}+T_dc_d^{}.$$
$`2.3`$
Formula (2.1) defines a Dirac operator $`(D,R)`$, and we now show that the operators
$$\stackrel{~}{D}=\stackrel{~}{B}+\stackrel{~}{B}^{},\stackrel{~}{R}(z)=R(\overline{z}),z^d$$
also define a Dirac operator $`(\stackrel{~}{D},\stackrel{~}{R})`$, $`R`$ being the Clifford structure of (2.1) and $`\overline{z}`$ denoting the natural conjugation in $`^d`$, for $`z=(z_1,\mathrm{},z_d)`$, $`\overline{z}=(\overline{z}_1,\mathrm{},\overline{z}_d)`$.
###### Proposition: homology vs. cohomology
The pair $`(\stackrel{~}{D},\stackrel{~}{R})`$ is a Dirac operator on $`H\mathrm{\Lambda }^d`$, and it is isomorphic to the Dirac operator $`(D,R)`$ of (2.1). The gauge group $`\stackrel{~}{\mathrm{\Gamma }}`$ of $`(\stackrel{~}{D},\stackrel{~}{R})`$ is related to the gauge group $`\mathrm{\Gamma }`$ of $`(D,R)`$ by $`\stackrel{~}{\mathrm{\Gamma }}(\lambda )=\lambda ^d\mathrm{\Gamma }(\lambda ^1)`$.
###### Demonstration proof
Consider the annihilation opertors $`a_k=c_k^{}`$, $`1kd`$. Obviously, the operators $`a_1,\mathrm{},a_d`$ and their adjoints form an irreducible set of operators satisfying (1.4), hence there is a unitary operator $`U(\mathrm{\Lambda }^d)`$ such that $`Uc_kU^{}=c_k^{}`$, $`k=1,\mathrm{},d`$. Letting $`C_0`$ and $`\stackrel{~}{C}_0`$ be the corresponding anticommutation relations in the form (1.5),
$$C_0(z)=z_1c_1+\mathrm{}+z_dc_d,\stackrel{~}{C}_0(z)=z_1c_1^{}+\mathrm{}+z_dc_d^{},$$
we have $`\stackrel{~}{C}_0(z)=C_0(\overline{z})^{}`$, and moreover $`\stackrel{~}{C}_0(z)=UC_0(z)U^{}`$, $`z^d`$. It follows that the unitary operator $`W=\mathrm{𝟙}_HU(H\mathrm{\Lambda }^d)`$ satisfies $`WC(z)W^{}=C(\overline{z})^{}`$, $`z^d`$. Since
$$\stackrel{~}{R}(z)=R(\overline{z})=C(\overline{z})+C(\overline{z})^{}=W(C(z)^{}+C(z))W^{}=WR(z)W^{}$$
and since $`\stackrel{~}{B}=WBW^{}`$, $`W`$ implements an isomorphism of the pair $`(D,R)`$ and the pair $`(\stackrel{~}{D},\stackrel{~}{R})`$. Thus, $`(\stackrel{~}{D},\stackrel{~}{R})`$ is a Dirac operator isomorphic to $`(D,R)`$.
Letting $`C_k=\mathrm{𝟙}c_k`$, $`k=1,\mathrm{},d`$ the number operators $`\stackrel{~}{N}`$ and $`N`$ of $`(\stackrel{~}{D},\stackrel{~}{R})`$ and $`(D,R)`$ are seen to be
$$\stackrel{~}{N}=C_1^{}C_1+\mathrm{}+C_d^{}C_d,N=C_1C_1^{}+\mathrm{}+C_dC_d^{},$$
so by the anticommutation relations (1.4) we have $`\stackrel{~}{N}=d\mathrm{𝟙}N`$, and the formula relating $`\stackrel{~}{\mathrm{\Gamma }}`$ to $`\mathrm{\Gamma }`$ follows from Remark 1.6.
In particular, the preceding proposition implies that the Taylor spectrum can be defined in either cohomological terms (using $`(D,R)`$ and its associated coboundary operator $`B`$) or in homological terms (using $`(\stackrel{~}{D},\stackrel{~}{R})`$ and its boundary operator $`\stackrel{~}{B}`$). It is the homological formulation that leads to the following interpretation.
Classical spectral theory starts with the problem of solving linear equations of the form $`Tx=y`$, where $`T`$ is a given operator in $`(H)`$, $`y`$ is a given vector in $`H`$, and $`x`$ is to be found; $`T`$ is said to be invertible when for every $`y`$ there is a unique $`x`$. Taylor’s notion of invertibility in its homological form provides the correct generalization to higher dimensions of this fundamental notion in dimension one. In dimension two for example, one has a pair $`T_1,T_2`$ of commuting operators acting on a Hilbert space $`H`$, and one is interested in solving equations of the form
$$T_1x_1+T_2x_2=y,$$
$`2.4`$
where $`y`$ is a given vector in $`H`$. Of course the pair $`(x_1,x_2)`$ is never uniquely determined by $`y`$, since if $`(x_1,x_2)`$ solves this equation then so does $`(x_1^{},x_2^{})`$ where $`x_1^{}=x_1+T_2\zeta `$ and $`x_2^{}=x_2T_1\zeta `$ where $`\zeta H`$ is arbitrary. Equivalently,
$`x_1^{}`$ $`=x_1+T_1\xi _{11}+T_2\xi _{12}`$
$`x_2^{}`$ $`=x_2+T_1\xi _{21}+T_2\xi _{22},`$
where the vectors $`\xi _{ij}`$, $`1i,j2`$ satisfy $`\xi _{ji}=\xi _{ij}`$ for all $`i,j`$ but are otherwise arbitrary (of course, we must have $`\xi _{11}=\xi _{22}=0`$ and $`\xi _{12}=\zeta `$). Such perturbations $`(x_1^{},x_2^{})`$ can be written down independently of any properties of the given operators $`T_1`$, $`T_2`$ (beyond commutativity, of course), and for that reason we will call them tautological perturbations of the given solution $`x_1,x_2`$. In order to understand how to solve such equations one needs to determine what happens modulo tautological perturbations, and for that one must look at the homology of (2.4).
Since we are in dimension two we can write
$$H\mathrm{\Lambda }^2=\mathrm{\Omega }_0\mathrm{\Omega }_1\mathrm{\Omega }_2$$
where $`\mathrm{\Omega }_0=H`$, $`\mathrm{\Omega }_1=\{(x_1,x_2):x_kH\}`$, and $`\mathrm{\Omega }_2`$ is parameterized as a space of “antisymmetric” sequences as follows
$$\mathrm{\Omega }_2=\{(\xi _{ij}):1i,j2,\xi _{ij}=\xi _{ji}\text{ for all }i,j\}.$$
Of course, $`\mathrm{\Omega }_2`$ is isomorphic to $`H`$ by way of the map which associates to a vector $`\zeta H`$ the antisymmetric sequence $`\xi _{11}=\xi _{22}=0,\xi _{12}=\zeta ,\xi _{21}=\zeta `$. The homological boundary operator $`B=T_1c_1^{}+T_2c_2^{}`$ of the complex
$$0\mathrm{\Omega }_0\mathrm{\Omega }_1\mathrm{\Omega }_20$$
$`2.5`$
acts as follows. On $`\mathrm{\Omega }_1`$, $`B(x_1,x_2)=T_1x_1+T_2x_2`$, and on $`\mathrm{\Omega }_2`$
$$B(\xi _{ij})=(T_1\xi _{11}+T_2\xi _{12},T_1\xi _{21}+T_2\xi _{22})=(T_2\xi _{12},T_1\xi _{12}).$$
Apparently, (2.4) has a solution iff $`y`$ belongs to $`B\mathrm{\Omega }_1=T_1H+T_2H`$. Given a solution $`(x_1,x_2)`$ of (2.4) and another pair of vectors $`(x_1^{},x_2^{})`$, $`(x_1^{},x_2^{})`$ is also a solution iff the difference $`(x_1x_1^{},x_2x_2^{})`$ belongs to $`\mathrm{ker}B`$. Given that $`(x_1^{},x_2^{})`$ is a solution, then it is a tautological perturbation of $`(x_1,x_2)`$ iff the difference $`(x_1x_1^{},x_2x_2^{})`$ belongs to $`B\mathrm{\Omega }_2`$. Finally, the kernel of the boundary operator at $`\mathrm{\Omega }_2`$ is identified with $`\mathrm{ker}T_1\mathrm{ker}T_2`$. We conclude that the complex (2.5) is exact iff a) $`T_1H+T_2H=H`$, b) $`\mathrm{ker}T_1\mathrm{ker}T_2=\{0\}`$, and c) solutions of (2.4) are unique up to tautological perturbations. While the algebra is more subtle in higher dimensions the fundamental issues are the same, and that is why the Taylor spectrum is important in multivariable spectral theory.
We will not have to delve into homological issues here; but the above comments do show that the theory of abstract Dirac operators is rooted in concrete problems of linear algebra that are associated with solving linear equations involving commuting sets of operators.
Taylor’s definition of invertibility can be reformulated in terms of the Dirac operator $`D_{\overline{T}}`$, and then extended to define Fredholm $`d`$-tuples and their index. In more detail, in the proof of the previous proposition we have already pointed out that $`D^2=B^{}B+BB^{}`$; and since $`B\stackrel{~}{H}`$ and $`B^{}\stackrel{~}{H}`$ are orthogonal, we conclude that $`B\stackrel{~}{H}=\mathrm{ker}B`$ iff $`D^2`$ is invertible.
Conclusion: A commuting $`d`$-tuple $`(T_1,\mathrm{},T_d)`$ is invertible if and only if its Dirac operator is invertible.
By a Fredholm $`d`$-tuple we mean one whose Dirac operator $`(D,R)`$ is Fredholm in the sense that the self-adjoint operator $`D`$ has closed range and finite dimensional kernel. The index of a Fredholm $`d`$-tuple is defined as follows. By property (D1) we have $`D\stackrel{~}{H}_+\stackrel{~}{H}_{}`$ and $`D\stackrel{~}{H}_{}\stackrel{~}{H}_+`$. Thus we may consider the operator
$$D_+=D_{H_+}(\stackrel{~}{H}_+,\stackrel{~}{H}_{}),$$
whose adjoint is given by
$$D_+^{}=D_H_{}(\stackrel{~}{H}_{},\stackrel{~}{H}_+).$$
For a Fredholm $`d`$-tuple $`\overline{T}=(T_1,\mathrm{},T_d)`$, $`D_+`$ is a Fredholm operator from $`\stackrel{~}{H}_+`$ to $`\stackrel{~}{H}_{}`$, and the index of $`\overline{T}`$ is defined by
$$\text{ind}(\overline{T})=dim\mathrm{ker}(D_+)dim\mathrm{ker}(D_+^{}).$$
One can define semi-Fredholm $`d`$-tuples similarly, but we do not require the generalization here.
## 3. Dirac operators and Hilbert modules over $`[z_1,\mathrm{},z_d]`$
In this section we prove the following result, which implies that Dirac operators $`(D,R)`$ contain exactly the same geometric information as multioperators $`\overline{T}`$.
###### Theorem A
For every $`d`$-dimensional Dirac operator $`(D,R)`$ there is a commuting $`d`$-tuple $`\overline{T}=(T_1,\mathrm{},T_d)`$ acting on a Hilbert space $`H`$ such that $`(D,R)`$ is isomorphic to $`D_{\overline{T}}`$. If $`\overline{T}^{}=(T_1^{},\mathrm{},T_d^{})`$ is another commuting $`d`$-tuple acting on $`H^{}`$, then $`D_{\overline{T}}`$ and $`D_{\overline{T}^{}}`$ are isomorphic if and only if there is a unitary operator $`U:HH^{}`$ such that $`UT_k=T_k^{}U`$ for every $`k=1,\mathrm{},d`$.
###### Demonstration proof
Let $`K`$ be the underlying Hilbert space of $`(D,R)`$, so that $`D=D^{}(K)`$ and $`R:^d(K)`$ is a Clifford structure (1.1) which satisfy (D1), (D2), (D3).
Consider the map $`C:^d(K)`$ defined by $`C(z)=(1/2)(R(z)iR(iz))`$. The discussion of section 1 implies that $`C`$ satisfies the anticommutation relations (1.5), and $`R(z)=C(z)+C(z)^{}`$. $`C`$ is unitarily equivalent to a direct sum of copies of the standard irreducible representation $`C_0`$ of the anticommutation relations on $`\mathrm{\Lambda }^d`$; thus by replacing $`(D,R)`$ with an isomorphic copy we may assume that there is a Hilbert space $`H`$ such that $`K=H\mathrm{\Lambda }^d`$ and that $`R(z)=C(z)+C(z)^{}`$ where $`C(z)`$ is defined on $`H\mathrm{\Lambda }^d`$ by
$$C(z)=\mathrm{𝟙}_HC_0(z),z^d.$$
We must exhibit a commuting set of operators $`T_1,\mathrm{},T_d`$ on $`H`$ so that $`D=B+B^{}`$ where
$$B=T_1C_0(e_1)+\mathrm{}+T_dC_0(e_d),$$
$`e_1,\mathrm{},e_d`$ being the usual orthonormal basis for $`^d`$.
To that end, let $`𝒜`$ be the finite dimensional $`C^{}`$-algebra $`𝒜=\mathrm{𝟙}_H(\mathrm{\Lambda }^d)`$. The $`C^{}`$-algebras generated by $`R(^d)`$ and $`C(^d)`$ are the same, and in fact
$$C^{}(R(^d))=C^{}(C(^d))=𝒜.$$
$`3.1.`$
By (D1), $`R(z)D+DR(z)`$ must commute with $`𝒜`$ for every $`z^d`$ and, in view of the relation $`C(e_k)^{}=2(R(e_k)+iR(ie_k))`$ we have $`C(e_k)^{}D+DC(e_k)^{}𝒜^{}`$. Thus for every $`k`$ there is a unique operator $`T_k(H)`$ such that
$$C(e_k)^{}D+DC(e_k)^{}=T_k\mathrm{𝟙}_{\mathrm{\Lambda }^d}.$$
$`3.2`$
For each $`k=1,\mathrm{},d`$, let $`c_k=C_0(e_k)(\mathrm{\Lambda }^d)`$, and consider the operator
$$B=T_1c_1+\mathrm{}+T_dc_d(H\mathrm{\Lambda }^d).$$
$`3.3`$
In order to show that $`D=B+B^{}`$ we will make use of
###### Lemma
Let $`R:^d(K)`$ be a Clifford structure (1.1) on $`K`$ and let $`\mathrm{\Gamma }:𝕋(K)`$ be its gauge group. Every operator $`A(K)`$ satisfying $`R(z)A+AR(z)=0`$ for every $`z^d`$ admits a decomposition $`A=A_0\mathrm{\Gamma }(1)`$, where $`A_0`$ belongs to the commutant of $`C^{}(R(^d))`$. In particular, such an operator must also be gauge invariant in the sense that $`\mathrm{\Gamma }(\lambda )A\mathrm{\Gamma }(\lambda )^{}=A`$, $`\lambda 𝕋`$.
###### Demonstration proof
Since $`\mathrm{\Gamma }(1)R(z)\mathrm{\Gamma }(1)^{}=R(z)=R(z)`$ it follows that $`\mathrm{\Gamma }(1)`$ anticommutes with $`R(z)`$ for every $`z^d`$. Since $`A`$ also anticommutes with $`R(z)`$, the operator $`A_0=A\mathrm{\Gamma }(1)`$ must commute with $`R(z)`$, and we have $`A=A\mathrm{\Gamma }(1)^2=A_0\mathrm{\Gamma }(1)`$ as required. The last assertion follows from this decomposition, because for every $`\lambda 𝕋`$, $`\mathrm{\Gamma }(\lambda )`$ belongs to the $`C^{}`$-algebra generated by the range of $`R`$ and hence commutes with both factors $`A_0`$ and $`\mathrm{\Gamma }(1)`$.
We now show that for $`B`$ as in (3.3) we have $`D=B+B^{}`$. Indeed, since $`C(e_k)=\mathrm{𝟙}_Hc_k`$ and the $`c_k`$ satisfy the anticommutation relations (1.4) we have
$`C(e_k)B+BC(e_k)`$ $`={\displaystyle \underset{j=1}{\overset{d}{}}}T_j(c_kc_j+c_jc_k)=0,`$
$`C(e_k)^{}B+BC(e_k)^{}`$ $`={\displaystyle \underset{j=1}{\overset{d}{}}}T_j(c_k^{}c_j+c_jc_k^{})={\displaystyle \underset{j=1}{\overset{d}{}}}T_j\delta _{jk}\mathrm{𝟙}=T_k\mathrm{𝟙}.`$
Using the definition of $`T_k`$ (3.2) it follows from the preceding calculation that the difference $`DBB^{}`$ must anticommute with all of the operators $`C(e_j),C(e_k)^{}`$, $`1j,kd`$. Since $`R(z)=C(z)+C(z)^{}`$ it follows that $`DBB^{}`$ anticommutes with $`R(z)`$ for every $`z^d`$.
By the Lemma, there is a (necessarily unique) operator $`X(H)`$ such that
$$DBB^{}=X\mathrm{\Gamma }_0(1),$$
$`3.4`$
where $`\mathrm{\Gamma }_0:𝕋(\mathrm{\Lambda }^d)`$ is the natural gauge action on $`\mathrm{\Lambda }^d`$ and $`\mathrm{\Gamma }(\lambda )=\mathrm{𝟙}_H\mathrm{\Gamma }_0(\lambda )`$. We want to show that $`X=0`$. For that, recall that $`D`$ is odd (property (D1)) and $`B`$ is clearly odd by its definition (3.3). Hence $`DBB^{}`$ is odd, so it must anticommute with the unitary operator $`\mathrm{\Gamma }(1)=P_{H_+}P_H_{}`$. On the other hand (3.4) implies that it commutes with $`\mathrm{\Gamma }(1)`$. Since $`\mathrm{\Gamma }(1)`$ is invertible, $`DBB^{}=0`$.
What remains to be proved is that the operators $`T_k`$ of (3.2) commute with each other. Indeed, we claim first that $`B^2=0`$. Since we have established that $`D=B+B^{}`$ we can write
$$D^2=B^{}B+BB^{}+B^2+B_{}^{}{}_{}{}^{2}.$$
$`3.5`$
From the definition of $`B`$ (3.3) we have
$$\mathrm{\Gamma }(\lambda )B\mathrm{\Gamma }(\lambda )^{}=\underset{k=1}{\overset{d}{}}T_k\mathrm{\Gamma }_0(\lambda )c_k\mathrm{\Gamma }_0(\lambda )^{}=\lambda \underset{k=1}{\overset{d}{}}T_kc_k=\lambda B,\lambda 𝕋.$$
It follows that $`B^{}B`$ and $`BB^{}`$ are invariant under the action of the gauge group, and that $`\mathrm{\Gamma }(\lambda )B^2\mathrm{\Gamma }(\lambda )^{}=\lambda ^2B^2`$. Thus
$$\mathrm{\Gamma }(\lambda )D^2\mathrm{\Gamma }(\lambda )^{}=B^{}B+BB^{}+\lambda ^2B^2+\overline{\lambda }^2B_{}^{}{}_{}{}^{2}.$$
$`3.6`$
Because of (D2), the left side of (3.6) does not depend on $`\lambda `$. Hence by equating Fourier coefficients on left and right we find that $`B^2=B_{}^{}{}_{}{}^{2}=0`$.
We can now show that the operators $`T_k`$ defined by (3.2) mutually commute. Consider the operator $`C`$ defined on $`H\mathrm{\Lambda }^d`$ by
$$C=\underset{1j<kd}{}(T_jT_kT_kT_j)c_jc_k.$$
$`3.7`$
Since the operators $`\{c_jc_k:1j<kd\}(\mathrm{\Lambda }^d)`$ are linearly independent, it is enough to show that $`C=0`$. To see this, we use the anticommutation relations $`c_kc_j+c_jc_k=\delta _{jk}\mathrm{𝟙}`$ to write
$`C`$ $`={\displaystyle \underset{1j<kd}{}}T_jT_kc_jc_k{\displaystyle \underset{1j<kd}{}}T_kT_jc_jc_k`$
$`={\displaystyle \underset{1j<kd}{}}T_jT_kc_jc_k+{\displaystyle \underset{1j<kd}{}}T_kT_jc_kc_j`$
$`={\displaystyle \underset{1p,qd}{}}T_pT_qc_pc_q=B^2=0.`$
That completes the proof that every Dirac operator is associated with a commuting $`d`$-tuple.
Suppose now that we are given two commuting $`d`$-tuples $`\overline{T}`$ and $`\overline{T}^{}`$, acting on Hilbert spaces $`H`$ and $`H^{}`$. It is obvious that if $`U:HH^{}`$ is a unitary operator satisfying $`UT_k=T_k^{}U`$ for every $`k=1,\mathrm{},d`$, then $`W=U\mathrm{𝟙}:H\mathrm{\Lambda }^dH^{}\mathrm{\Lambda }^d`$ is a unitary operator which implements an isomorphism of the respective Dirac operators.
Conversely, let $`W:H\mathrm{\Lambda }^dH^{}\mathrm{\Lambda }^d`$ be a unitary operator implementing an isomorphism of the respective Dirac operators $`(D,R)`$ and $`(D^{},R^{})`$ associated with $`\overline{T}`$ and $`\overline{T}^{}`$. Let $`R_0:^d\mathrm{\Lambda }^d`$ be the irreducible Clifford structure defined in section 1. Since $`R(z)=\mathrm{𝟙}_HR_0(z)`$ and $`R^{}(z)=\mathrm{𝟙}_H^{}R_0(z)`$, it follows that $`H`$ and $`H^{}`$ have the same dimension (namely the common multiplicity of the unitarily equivalent Clifford structures $`R`$ and $`R^{}=WRW^{}`$). Thus by replacing $`\overline{T}^{}`$ with a unitarily equivalent $`d`$-tuple, we can assume that $`H=H^{}`$, i.e., that both $`d`$-tuples act on the same Hilbert space $`H`$.
In these “coordinates”, the relation
$$W(\mathrm{𝟙}_HR_0(z))W^{}=\mathrm{𝟙}_HR_0(z),z^d$$
implies that $`W`$ commutes with $`\mathrm{𝟙}_H(\mathrm{\Lambda }^d)`$, the $`C^{}`$-algebra generated by $`R(^d)`$. Thus $`W`$ decomposes $`W=U\mathrm{𝟙}_{\mathrm{\Lambda }^d}`$ where $`U`$ is a uniquely determined unitary operator on $`H`$. Now according to the definition of Dirac operators (2.1), we have $`D=B+B^{}`$, $`D^{}=B^{}+B_{}^{}{}_{}{}^{}`$, where
$$B=T_1c_1+\mathrm{}+T_dc_d,B^{}=T_1^{}c_1+\mathrm{}+T_d^{}c_d,$$
$`c_1,\mathrm{},c_d`$ being the irrecucible representation of the canonical anticommutation relations (1.4) associated with $`R_0`$. Letting $`C_k=\mathrm{𝟙}_Hc_k`$ and using (1.4), a routine calculation gives
$$C_k^{}D+DC_k^{}=T_k\mathrm{𝟙},C_k^{}D^{}+D^{}C_k^{}=T_k^{}\mathrm{𝟙},k=1,\mathrm{},d.$$
Since $`U\mathrm{𝟙}=W`$ commutes with all $`C_k^{}`$ and satisfies $`WDW^{}=D^{}`$, it follows that for every $`k=1,\mathrm{},d`$ we have
$$UT_kU^{}\mathrm{𝟙}=W(C_k^{}D+DC_k^{})W^{}=C_kD^{}+D^{}C_k^{}=T_k^{}\mathrm{𝟙},$$
and hence $`U`$ implements a unitary equivalence of $`\overline{T}`$ and $`\overline{T}^{}`$. That completes the proof of Theorem A.
###### Remark Remark 3.8
It is worth pointing out that the proof of Theorem A shows how one may go directly from a Dirac operator $`(D,R)`$ (acting on $`H`$) to the Koszul complex of its underlying $`d`$-tuple $`\overline{T}`$ (the operators $`T_1,\mathrm{},T_d`$ acting on some other Hilbert space) without making explicit reference to $`\overline{T}`$. Indeed, considering the spectral representation of the gauge group of $`R`$
$$\mathrm{\Gamma }(\lambda )=\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\lambda ^nE_n=\underset{n=0}{\overset{d}{}}\lambda ^nE_n,\lambda 𝕋$$
and the operator $`B=_nE_{n+1}DE_n`$, from the proof of Theorem A one finds that
$$B^2=0,D=B+B^{}.$$
$`3.9`$
Moreover, the spectral subspaces $`H_n=E_nH`$ satisfy $`BH_nH_{n+1}`$, $`B^{}H_nH_{n1}`$, and the Koszul complex is given by
$$0H_0H_1\mathrm{}H_d0$$
with cohomology defined by $`B`$.
## 4. Stability of the Curvature invariant: graded case
Recall from that a commuting $`d`$-tuple of operators $`(T_1,\mathrm{},T_d)`$ on a Hilbert space $`H`$ is said to be graded if it is circularly symmetric in the sense that there is a strongly continuous unitary representation $`\mathrm{\Gamma }:𝕋(H)`$ such that
$$\mathrm{\Gamma }(\lambda )T_k\mathrm{\Gamma }(\lambda )^{}=\lambda T_k,k=1,\mathrm{},d,\lambda 𝕋.$$
Many examples of graded $`d`$-contractions were described in ; in particular, all examples of $`d`$-contractions that were associated with projective algebraic varieties (and their finitely generated modules) are graded.
It was shown in (, see Theorem B) that the curvature invariant of a pure graded finite rank $`d`$-contraction is an integer, namely the Euler characteristic of a certain finitely generated algebraic module over $`[z_1,\mathrm{},z_d]`$ that is associated naturally with $`\stackrel{~}{T}`$. However, in the ungraded case this formula fails (both sides of this formula still make sense in the ungraded case, but examples are given in for which they are unequal). This led us to ask in , if $`K(\overline{T})`$ is an integer even when $`\overline{T}`$ is ungraded. That has been recently proved by Greene, Richter and Sundberg , and in fact the results of show that the integer $`K(\overline{T})`$ can be identified as the (almost everywhere constant) rank of the boundary values of a certain operator-valued “inner” function that is naturally associated with $`\overline{T}`$ via dilation theory (a fuller discussion of this inner operator and its relation to $`\overline{T}`$ can be found in ). It is fair to say that the rank of this inner function is not easily computed in terms of the operator theory of $`\overline{T}`$, and thus we were led to ask if there is a formula that relates the curvature invariant more directly to the geometry of $`\overline{T}`$…preferably in terms of an expression that is obviously an integer.
It is also noteworthy that the asymptotic formula for the curvature (Theorem C of ) implies that it has certain stability properties; for example, the curvature is stable under the operation of restricting to an invariant subspace of finite codimension. But nothing was known about stability of the curvature invariant under compact perturbations.
These questions led us to search for another formula for the curvature invariant that looks more like an index theorem in the sense that it equates the curvature invariant to the index of some operator. Such a formula would presumably lead to stability under compact perturbations, it would imply that the curvature invariant is in all cases an integer, and it would more closely resemble the Gauss-Bonnet-Chern formula in its modern incarnation as an index theorem (for example, see p. 311 of ). As a first step in this direction, we offer the following.
###### Theorem B
Let $`\overline{T}=(T_1,\mathrm{},T_d)`$ be a pure $`d`$-contraction of finite rank acting on a Hilbert space $`H`$. Assume that $`\overline{T}`$ is graded and let $`(D,R)`$ be its Dirac operator. Then both $`\mathrm{ker}D_+`$ and $`\mathrm{ker}D_+^{}`$ are finite dimensional and
$$(1)^dK(\overline{T})=dim\mathrm{ker}D_+dim\mathrm{ker}D_+^{}.$$
###### Remark Remark
Notice that we have not assumed that $`D`$ is a Fredholm operator. However, when it is Fredholm we have the following stability.
###### Corollary
Let $`\overline{T}=(T_1,\mathrm{},T_d)`$ and $`\overline{T}^{}=(T_1^{},\mathrm{},T_d^{})`$ be two pure $`d`$-contractions of finite rank acting on respective Hilbert spaces $`H`$, $`H^{}`$. Assume that both $`\overline{T}`$ and $`\overline{T}^{}`$ are graded, that $`\overline{T}`$ is Fredholm, and that they are unitarily equivalent modulo compacts in the sense that there is a unitary operator $`U:HH^{}`$ such that
$$UT_kT_k^{}U\text{ is compact},k=1,\mathrm{},d.$$
Then $`K(\overline{T})=K(\overline{T}^{})`$.
###### Demonstration proof of Corollary
Let $`(D,R)`$ and $`(D^{},R^{})`$ be the Dirac operators of $`\overline{T}`$ and $`\overline{T}^{}`$, acting on respective Hilbert spaces $`\stackrel{~}{H}=H\mathrm{\Lambda }^d`$ and $`\stackrel{~}{H}^{}=H^{}\mathrm{\Lambda }^d`$. The hypothesis implies that the unitary operator $`W:U\mathrm{𝟙}:\stackrel{~}{H}\stackrel{~}{H}^{}`$ satisfies $`WR(z)=\stackrel{~}{R}(z)W`$ for all $`z^d`$, and $`WDD^{}W`$ is compact. The first of these two relations implies that $`W`$ implements an equivalence of the respective gauge groups $`W\mathrm{\Gamma }(\lambda )=\mathrm{\Gamma }^{}(\lambda )W`$, and hence $`W`$ carries the $`_2`$-grading of $`\stackrel{~}{H}`$ to that of $`\stackrel{~}{H}^{}`$. It follows that the restrictions of $`W`$ to the even and odd subspaces of $`\stackrel{~}{H}`$ implement a unitary equivalence modulo compact operators of the two operators $`D_+`$ and $`D_+^{}`$. Since $`D_+`$ is Fredholm by hypothesis, $`D_+^{}`$ must be Fredholm as well, and moreover they must have the same index. From Theorem B we conclude that $`K(\overline{T})=K(\overline{T}^{})`$.
Before giving the proof of Theorem B, we recall some algebraic preliminaries. Let $`𝒜`$ be the complex polynomial algebra $`[z_1,\mathrm{},z_d]`$. By an $`𝒜`$-module we mean a complex vector space $`M`$ which is endowed with a commuting $`d`$-tuple of linear operators $`T_1,\mathrm{},T_d`$, the module structure being defined by $`f\xi =f(T_1,\mathrm{},T_d)\xi `$, $`f𝒜`$, $`\xi M`$. $`M`$ is said to be finitely generated if there is a finite set $`\xi _1,\mathrm{},\xi _s`$ of vectors in $`M`$ such that
$$M=\{f_1\xi _1+\mathrm{}+f_s\xi _s:f_1,\mathrm{},f_s𝒜\}.$$
The free module of rank $`1`$ is defined to be $`𝒜`$ itself, with the module action associated with multiplication of polynomials. The free module of rank $`r=1,2,\mathrm{}`$ is the direct sum of $`r`$ copies of the free module of rank $`1`$, with the obvious module action on $`r`$-tuples of polynomials.
Hilbert’s Syzygy theorem implies that every finitely generated $`𝒜`$-module has a finite free resolution in the sense that there is an exact sequence of $`𝒜`$-modules
$$0F_n\mathrm{}F_1M0$$
$`4.1`$
where each $`F_k`$ is a free module of finite rank. In , we defined the Euler characteristic of $`M`$ in terms of finite free resolutions (4.1) as follows
$$\chi (M)=\underset{k=1}{\overset{n}{}}(1)^{k+1}\text{rank}(F_k).$$
$`4.2`$
This integer does not depend on the particular resolution of $`M`$ chosen to define it.
We must relate $`\chi (M)`$ to the alternating sum of the Betti numbers of the Koszul complex of $`M`$; since the latter is also called the Euler characteristic, we distinguish it from $`\chi (M)`$ by calling it the Euler number of $`M`$ and by writing it as $`e(M)`$. The Euler number is defined as follows.
The Koszul complex of an $`𝒜`$-module $`M`$ is defined as the $`𝒜`$-module
$$M\mathrm{\Lambda }^d=\mathrm{\Omega }^0\mathrm{\Omega }^1\mathrm{}\mathrm{\Omega }^d,$$
where $`\mathrm{\Omega }^k=M\mathrm{\Lambda }^k^d`$ is the submodule of $`k`$-forms, with coboundary operator
$$B=T_1c_1+\mathrm{}+T_dc_d$$
exactly as we have done above in the case where $`M`$ is a Hilbert space and the $`T_k`$ are bounded linear operators. Letting $`B_k`$ be the restriction of $`B`$ to $`\mathrm{\Omega }^k`$ we have a corresponding cohomology space $`H^k(M)=\mathrm{ker}B_k/\text{ran}B_{k1}`$ for $`1kd`$, $`H^0(M)=\mathrm{ker}B_0`$, which may or may not be finite dimensional. $`M`$ is said to be of finite type if $`H^k(M)`$ is finite dimensional for every $`0kd`$, and in that case the Euler number is defined by
$$e(M)=\underset{k=0}{\overset{d}{}}(1)^kdimH^k(M).$$
$`4.3`$
Taking $`M`$ to be the free module $`𝒜`$ of rank one, it is well-known that $`H^k(𝒜)=0`$ for $`0kd1`$ and that $`H^d(𝒜)=𝒜/(z_1𝒜+\mathrm{}+z_d𝒜)`$ is one-dimensional. It follows that for a free module $`F`$ of arbitrary finite rank, we have
$$e(F)=(1)^d\text{rank}F.$$
$`4.4`$
The following result is part of the lore of commutative algebra; we sketch a proof for the reader’s convenience.
###### Lemma 1
Let $`0KLM0`$ be a short exact sequence of $`𝒜`$ modules, some two of which are of finite type. Then all are of finite type and we have
$$e(L)=e(K)+e(M).$$
###### Demonstration proof
Letting $`\kappa (N)`$ denote the Koszul complex of an $`𝒜`$-module N, one sees that $`\kappa (N)`$ has $`d+1`$ nonzero terms, and the corresponding sequence of complexes
$$0\kappa (K)\kappa (L)\kappa (M)0$$
is exact. Thus by fundamental principles we obtain a long exact sequence of cohomology spaces which contains at most $`3d+3`$ nonzero terms. Two of any three consecutive terms in the latter sequence are finite dimensional because two of the three modules $`K,L,M`$ are assumed to have finite dimensional cohomology. By exactness all cohomology spaces are finite dimensional and the alternating sum of their dimensions must be zero. The asserted formula follows.
###### Lemma 2
Every finitely generated $`𝒜`$-module $`M`$ is of finite type, and
$$e(M)=(1)^d\chi (M).$$
###### Demonstration proof
Choose a finite free resolution of $`M`$ in the form (4.1)
$$0F_n\mathrm{}F_1M0.$$
Let $`R_kF_{k1}`$ be the image of $`F_k`$ in $`F_{k1}`$, $`2kn`$, and let $`R_1M`$ be the image of $`F_1`$. Starting at the left of (4.1) we have a short exact sequence of modules
$$0F_nF_{n1}R_{n1}0,$$
the first two of which are of finite type. By Lemma 1, $`R_{n1}`$ is of finite type and
$$e(R_{n1})=e(F_{n1})e(F_n).$$
Moving one step to the right, the same argument applied to
$$0R_{n1}F_{n2}R_{n2}0$$
shows that $`R_{n2}`$ is of finite type and
$$e(R_{n2})=e(F_{n2})e(R_{n1})=e(F_{n2})e(F_{n1})+e(F_n).$$
Continuing in this way to the end of the sequence, we arrive at the conclusion that $`M`$ is of finite type and
$$e(M)=\underset{k=1}{\overset{n}{}}(1)^{k+1}e(F_k)=(1)^d\underset{k=1}{\overset{n}{}}(1)^{k+1}\text{rank}(F_k)=(1)^d\chi (M),$$
where in the second equality we have made use of (4.4).
###### Demonstration proof of Theorem B
We are assuming that $`\overline{T}`$ is graded; this means that there is a continuous unitary representation of the circle group $`U:𝕋(H)`$ such that
$$U_\lambda T_kU_\lambda ^{}=\lambda T_k,1kd.$$
$`4.5`$
Let $`\mathrm{\Delta }=(\mathrm{𝟙}T_1T_1^{}\mathrm{}T_dT_d^{})^{1/2}`$ be the defect operator of $`\overline{T}`$. By hypothesis, $`\mathrm{\Delta }`$ is of finite rank, and the canonical algebraic module $`M_H`$ associated with $`\overline{T}`$
$$M_H=\text{span}\{f(T_1,\mathrm{},T_d)\zeta :f𝒜,\zeta \mathrm{\Delta }H\}$$
is a finitely generated $`𝒜`$ module. Because $`\overline{T}`$ is pure, $`M_H`$ is dense in $`H`$ (see , Proposition 5.4).
It follows that $`M_H\mathrm{\Lambda }^d`$ is dense in $`\stackrel{~}{H}=H\mathrm{\Lambda }^d`$. Let $`D(\stackrel{~}{H})`$ be the Dirac operator of $`\overline{T}`$. We will show that both $`\mathrm{ker}D_+`$ and $`\mathrm{ker}D_+^{}`$ are finite dimensional subspaces of $`M_H\mathrm{\Lambda }^d`$, and that in fact we have
$$dim\mathrm{ker}(D_+)=\underset{k\text{ even}}{}dimH^k(M_H),dim\mathrm{ker}(D_+^{})=\underset{k\text{ odd}}{}dimH^k(M_H)$$
$`4.6`$
where $`M_H\mathrm{\Lambda }^d`$ is viewed as the Koszul complex of $`M_H`$. Assuming for the moment that (4.6) has been established we find that
$$dim\mathrm{ker}D_+dim\mathrm{ker}D_+^{}=\underset{k=0}{\overset{d}{}}(1)^kdimH^k(M_H)=e(M_H),$$
and by Lemma 2 the right side is $`(1)^d\chi (M_H)`$. By Theorem B of , the latter is $`(1)^dK(H)`$, and the proof of Theorem B above will be complete.
In order to establish (4.6) we make use of the grading as follows. Let $`c_1,\mathrm{},c_d`$ be operators on $`^d`$ satisfying the anticommutation relations (1.4) and let
$$B=T_1c_1+\mathrm{}+T_dc_d$$
be the coboundary operator on $`\stackrel{~}{H}`$. Since $`D^2=B^{}B+BB^{}`$, the kernel of $`D`$ is given by $`\mathrm{ker}D=\mathrm{ker}B\mathrm{ker}B^{}`$. Let $`V:𝕋(\stackrel{~}{H})`$ be the unitary representation corresponding to $`U`$, $`V_\lambda =U_\lambda 1_{\mathrm{\Lambda }^d}`$, $`\lambda 𝕋`$. By (4.5) we have
$$V_\lambda BV_\lambda ^{}=\lambda B,$$
and it follows that both $`\mathrm{ker}B`$ and $`\mathrm{ker}B^{}`$ are invariant under the action of $`V`$. Since the spectral subspaces of $`U`$ and $`V`$
$$H_k=\{\xi H:U_\lambda \xi =\lambda ^k\xi ,\lambda 𝕋\},\stackrel{~}{H}_k=\{\zeta \stackrel{~}{H}:V_\lambda \zeta =\lambda ^k\zeta ,\lambda 𝕋\}$$
are related by $`\stackrel{~}{H}_k=H_k\mathrm{\Lambda }^d`$, it follows that both $`\mathrm{ker}B`$ and $`\mathrm{ker}B^{}`$ decompose into orthogonal sums
$$\mathrm{ker}B=\underset{k}{}\mathrm{ker}B\stackrel{~}{H}_k,\mathrm{ker}B^{}=\underset{k}{}\mathrm{ker}B^{}\stackrel{~}{H}_k.$$
We conclude that
$$\mathrm{ker}D=\underset{k}{}\mathrm{ker}D\stackrel{~}{H}_k=\underset{k}{}\mathrm{ker}B\mathrm{ker}B^{}\stackrel{~}{H}_k.$$
It was shown in , Proposition 5.4, that each $`H_k`$ is a finite dimensional subspace of $`M_H`$, hence $`\stackrel{~}{H}_k`$ is a finite dimensional subspace of $`M_H\mathrm{\Lambda }^d`$. Since the restriction $`B_{M_H}`$ of $`B`$ to $`M_H\mathrm{\Lambda }^d`$ is the boundary operator of the Koszul complex of $`M_H`$ it follows that for the restriction $`B_k`$ of $`B`$ to $`M_H\stackrel{~}{H}_k`$ we have
$$dim(\mathrm{ker}D\stackrel{~}{H}_k)=dim(\mathrm{ker}B\mathrm{ker}B^{}\stackrel{~}{H}_k)=dim(\mathrm{ker}B_k/\text{ran }B_{k1}).$$
Summing on $`k`$ we find that
$$dim\mathrm{ker}D=dim(\mathrm{ker}B_{M_H}/\text{ran }B_{M_H}).$$
The right side of the preceding formula is finite, because the Koszul complex of $`M_H`$ has finite dimensional cohomology by Lemma 2.
By restricting this argument respectively to the even and odd subspaces of $`\stackrel{~}{H}`$, one finds in the same way that $`dim\mathrm{ker}D_+`$ and $`dim\mathrm{ker}D_+^{}`$ are, respectively, the total dimensions of the even and odd cohomology of the Koszul complex of $`M_H`$, and that gives the two formulas of (4.6).
###### Remark Concluding remarks and examples
It is natural to ask if Theorem B remains valid when one drops the hypothesis that $`\overline{T}`$ is graded. On the surface, this may appear a foolish question since it is not known if the Dirac operator associated with a finite rank pure $`d`$-contraction is Fredholm, and if it is not Fredholm then what does the index of $`D_+`$ mean? The Dirac operator is known to be Fredholm for a class of concrete examples (this is an unpublished result of the author’s), but the issue of Fredholmness for general pure finite rank $`d`$-contractions remains somewhat mysterious.
Nevertheless, Stephen Parrott has proved a result for single operators that implies that Theorem B is true verbatim for the one-dimensional case $`d=1`$ and an arbitrary pure contraction $`T`$ of finite rank, graded or not. His result implies that $`T`$ is necessarily a Fredholm operator.
Finally, we describe a class of examples of finite rank pure $`d`$-contractions $`\overline{T}=(T_1,\mathrm{},T_d)`$ in arbitrary dimension $`d=1,2,\mathrm{}`$. Most are ungraded. We show that these examples are Fredholm and we compute all three integer invariants (the index of the Dirac operator, the curvature invariant $`K(\overline{T})`$, and the Euler characteristic $`\chi (\overline{T})`$ of ). For some of these examples the formula $`K(\overline{T})=\chi (\overline{T})`$ of (, Theorem B) holds, but for most of them it fails. On the hand, in all cases the formula of Theorem B above
$$(1)^dK(\overline{T})=dim\mathrm{ker}D_+dim\mathrm{ker}D_+^{}$$
$`4.7`$
is satisfied. Indeed, we know of no examples for which (4.7) fails.
Fix $`d=1,2,\mathrm{}`$ and let $`r`$ be a positive integer. Following the notation and terminology of , we will consider the $`d`$-shift $`\overline{S}=(S_1,\mathrm{},S_d)`$ of multiplicity $`r+1`$. $`\overline{S}`$ acts on the Hilbert space $`(r+1)H^2`$, a direct sum of $`r+1`$ copies of the basic free Hilbert module $`H^2=H^2(^d)`$. We consider certain invariant subspaces $`M(r+1)H^2`$ and their quotient Hilbert modules $`H=(r+1)H^2/M`$. The $`d`$-shift compresses to a pure $`d`$-contraction $`\overline{T}=(T_1,\mathrm{},T_d)`$ acting on $`H`$, and the rank of $`\overline{T}`$ is at most $`r+1`$. For the examples below, the rank is $`r+1`$ and $`\overline{T}`$ will have the properties asserted above. The subspaces $`M`$ are defined as follows. Let $`\varphi _1,\varphi _2,\mathrm{},\varphi _r`$ be a set of multipliers of $`H^2`$ and set
$$M=\{(f,\varphi _1f,\varphi _2f,\mathrm{},\varphi _rf):fH^2\}(r+1)H^2.$$
###### Proposition
Assume that the set of $`r+1`$ functions $`\{1,\varphi _1,\varphi _2,\mathrm{},\varphi _r\}`$ is linearly independent, and let $`\overline{T}=(T_1,\mathrm{},T_d)`$ be the $`d`$-tuple of operators associated with the quotient Hilbert module $`H=(r+1)H^2/M`$. $`\overline{T}`$ is a pure $`d`$-contraction of rank $`r+1`$, it is Fredholm, and its index and curvature invariant are given by
$$dim\mathrm{ker}(D_+)dim\mathrm{ker}D_+^{}=(1)^dr,K(\overline{T})=r.$$
If each $`\varphi _k`$ is a homogeneous polynomial of some degree $`n_k`$ then the Euler characteristic is also given by $`\chi (\overline{T})=r`$. On the other hand, if $`M`$ contains no nonzero element $`(p_1,p_2,\mathrm{},p_{r+1})`$ with polynomial components $`p_k`$, then $`\chi (\overline{T})=r+1`$.
###### Remark Remark
For example, if each $`\varphi _k`$ is the exponential of some nontrivial polynomial (or more generally, if no $`\varphi _k`$ is a rational function), then the only $`r+1`$-tuple of polynomials $`(p_0,p_1,\mathrm{},p_r)`$ that belongs to $`M`$ is the zero $`r+1`$-tuple.
###### Demonstration proof
We merely sketch the key elements of the proof.
We deal first with the Euler characteristic. This invariant is associated with the finitely generalted algebraic $`[z_1,\mathrm{},z_d]`$-module
$$M_H=\text{span}\{f(T_1,\mathrm{},T_d)\zeta :f[z_1,\mathrm{},z_d],\zeta \mathrm{\Delta }H\},$$
$`\mathrm{\Delta }`$ being the finite rank defect operator $`\mathrm{\Delta }=(\mathrm{𝟙}T_1T_1^{}\mathrm{}T_dT_d^{})^{1/2}`$. Realizing the quotient $`H=(r+1)H^2/M`$ as the orthogonal complement $`M^{}(r+1)H^2`$, let $`E_0((r+1)H^2)`$ be the projection onto the $`r+1`$-dimensional space of constant vector functions. The operators $`T_1,\mathrm{},T_d`$ are obtained by compressing $`S_1,\mathrm{},S_d`$ to $`M^{}`$, and a straghtforward computation shows that $`\mathrm{\Delta }`$ is identified with the square root of the compression of $`E_0`$ to $`M^{}`$. This operator is of rank $`r+1`$ because of the linear independence hypothesis on $`\{1,\varphi _1,\mathrm{},\varphi _r\}`$ (for example, see 8.4.3 of ). It follows that $`M_H`$ is identified with the projection onto $`M^{}`$ of the space of all vector polynomials
$$S=\{(p_0,p_1,\mathrm{},p_r):p_k[z_1,\mathrm{},z_d]\}.$$
The $`[z_1,\mathrm{},z_d]`$-module action of a polynomial $`f[z_1,\mathrm{},z_d]`$ on $`M_H`$ is given by
$$fP_M^{}(p_0,p_1,\mathrm{},p_r)=P_M^{}(fp_0,fp_1,\mathrm{},fp_r).$$
Now assume that $`M`$ contains no nonzero element having polynomial components, and let $`F=(r+1)[z_1,\mathrm{},z_d]`$ denote the free $`[z_1,\mathrm{},z_d]`$-module of rank $`r+1`$. Consider the linear map
$$L:(f_0,f_1,\mathrm{},f_r)FP_M^{}(f_0,f_1,\mathrm{},f_r)M_H.$$
$`L`$ is injective by hypothesis, its range is all of $`M_H`$, and it is obviously a homomorphism of $`[z_1,\mathrm{},z_d]`$-modules. Hence $`M_H`$ is a free module of rank $`r+1`$ and its Euler characteristic is $`r+1`$. This shows that $`\chi (\overline{T})=r+1`$ in this case.
On the other hand, if each $`\varphi _k`$ is a homogeneous polynomial then one may extend the argument of the proof of (, Proposition 7.4) (which addresses the case $`r=1`$ explicitly) in straightforward way to show that $`\overline{T}`$ is a graded $`d`$-contraction. It follows from Theorem B of that $`\chi (\overline{T})=K(\overline{T})`$. We will show momentarily that in all cases we have $`K(\overline{T})=r`$, and this calculates the Euler characteristic for the asserted cases.
We show next that $`\overline{T}`$ is Fredholm of index $`(1)^dr`$. For that, it is enough to show that $`\overline{T}`$ is similar to a Fredholm $`d`$-tuple whose index is known to be $`(1)^dr`$. The latter $`d`$-tuple is the $`d`$-shift of multiplicity $`r`$. In more detail, consider the linear mapping $`A:(r+1)H^2rH^2`$ defined by
$$A(f_0,f_1,f_2,\mathrm{},f_r)=(f_1\varphi _1f_0,f_2\varphi _2f_0,\mathrm{},f_r\varphi _rf_0).$$
It is clear that $`A`$ is bounded, surjective, has kernel $`M`$, and intertwines the action of $`\overline{S}`$ (acting on $`(r+1)H^2`$) and the multiplicity $`r`$ $`d`$-shift acting on $`rH^2`$. Thus $`A`$ promotes to an isomorphism of Hilbert spaces $`\stackrel{~}{A}:HrH^2`$ which implements a similarity of $`\overline{T}`$ and the $`d`$-shift of multiplicity $`r`$. The latter is known to be Fredholm and has index $`(1)^dr`$.
Finally, in order to calculate $`K(\overline{T})`$ we appeal to a result of Greene, Richter and Sundberg as follows. Identifying $`H`$ with $`M^{}(r+1)H^2`$, we have already seen that the natural projection $`L=P_M^{}:(r+1)H^2H`$ is the minimal dilation of $`H`$ in the sense of , and obviously $`L`$ is a co-isometry with $`L^{}L=\mathrm{𝟙}P_M`$. Now if one evaluates all of the functions in $`M`$ at a point $`z`$ in the open unit ball of $`^d`$, one obtains the following linear subspace of $`^{r+1}`$
$$M(z)=\{(\lambda ,\lambda \varphi _1(z),\lambda \varphi _2(z),\mathrm{},\lambda \varphi _r(z)):\lambda \}.$$
This is a one-dimensional space having codimension $`(r+1)1=r`$, and the same assertion is valid for almost every point $`z`$ on the boundary of the unit ball. By the results of , the codimension of $`M(z)^{r+1}`$ is equal to $`K(\overline{T})`$ for almost every $`z`$ in the boundary of the unit ball. Thus, $`K(\overline{T})=r`$.
|
warning/0005/hep-th0005251.html
|
ar5iv
|
text
|
# CU-TP-977 HU-EP 00/23 UPR-889T hep-th/0005251 Local Anomaly Cancellation, M-Theory Orbifolds and Phase-Transitions
## 1 Introduction
M-theory harbors a broad spectrum of phenomena which can be systematically probed by analyzing anomalies in effective quantum field theories. In the case of orbifold compactifications of eleven-dimensional supergravity, a wide range of topological restrictions can be resolved, and the states localized on orbifold planes determined, by imposing factorization criteria on anomaly polynomials. The basic paradigm was espoused in by analyzing the $`S^1/𝐙_2`$ compactification. This was successfully applied to the $`T^5/𝐙_2`$ compactification in . In each of these cases, the orbifold planes comprise isolated, non-intersecting submanifolds. In more general situations, the orbifold planes can intersect, which gives rise to a number of novel features. In this paper, we describe local anomaly cancellation on $`S^1/𝐙_2\times T^4/𝐙_M`$ compactifications of M-theory, with $`M=2,3,4`$ or $`6`$. These correspond to special points in the moduli space of $`S^1/𝐙_2\times K3`$ compactifications. In such situations, the orbifold planes do intersect. These issues were first discussed in and later in . Here, we greatly extend the analysis of local anomaly cancellation in orbifolds of this type, particularly emphasizing results pertaining to the simplest of these cases, corresponding to $`M=2`$.
In our previous work , we described the general features of $`S^1/𝐙_2\times T^4/𝐙_M`$ $`M`$-theory orbifolds. In this paper, we expand those results, incorporating two subtle technical points which were not addressed in complete generality in our previous work. The first is an issue pertaining to the periodicity of the four-form $`G`$ that has recently been more thoroughly described in . The second issue concerns the precise normalization of the $`CGG`$ term in the supergravity action. Each of these impinge numerically on our analyses, both in and in this current paper, by changing the overall coefficient of the anomaly inflow due to the classical variation of the $`CGG`$ term. We treat this coefficient as a parameter, to be determined by consistency arguments, in a manner similar to the approach described in . In order that $`M`$-fivebranes have unit magnetic charge, we choose a scale for the three-form potential $`C`$ such that, upon integration over dimensions transverse to a fivebrane, we obtain $`𝑑G=1`$. This leaves only one coefficient in the basic Chern-Simons interactions of the effective field theory unconstrained by supersymmetry and by the requirement of fivebrane anomaly cancellation. However, this one parameter is uniquely fixed by the additional requirement of consistent orbifold compactifications and gives rise to the particular coefficient cited below for the $`CGG`$ inflow anomaly. As a result of these two described changes we now find a whole class of solutions of the local anomaly matching conditions, which in general consist of vector, hyper and tensor multiplets confined on the local orbifold six-planes and which also, for global consistency requirements, contain fivebranes, free to move in the 11-dimensional bulk (see also ref. for discussion of solutions of the local anomaly equations).
This situation is interesting since it is the simplest scenario which involves fivebrane-mediated phase transitions in which the gauge group is nontrivially influenced by local tensor couplings. Specifically, if a fivebrane hits one of the local orbifold six-planes it will be described by a torsion free sheaf being equivalent to a small gauge instanton. Due to the presence of this instanton the original gauge group at the local orbifold plane will be broken to some subgroup, where also the number of tensor and hypermultiplets gets changed. In this way elements of a certain class of orbifold compactifications are related to each other by fivebrane-mediated phase transitions, where the associated magnetic charges at the six-dimensional orbifold planes are changing by one unit. In addition we also discuss the possibility of phase transitions with half-integer change of magnetic charge at the local six-planes. We suggest that these phase transitions are due to fivebranes which split at interconnecting ten-planes and are then transmuted to half-integrally charged gauge instantons.
The orbifold compactifications which we discuss in this paper are of interest for other reasons as well. For example, in a series of papers, the $`S^1/𝐙_2`$ orbifold was compactified on smooth Calabi-Yau threefolds producing realistic “brane universe” theories of particle physics and cosmology . In future work, we will explore both the formal and phenomenological aspects of different $`M`$-theory orbifold compactifications.
## 2 $`S^1/𝐙_2\times T^4/𝐙_M`$ Orbifolds
The $`S^1/𝐙_2\times T^4/𝐙_M`$ orbifolds each involve a pair of ten-dimensional hyperplanes fixed under the $`𝐙_2`$ projection, which we denote by $`\alpha `$, and a set of distinct seven-dimensional hyperplanes fixed under the $`𝐙_M`$ projection, which we denote by $`\beta `$. Each of the $`\beta `$-planes transversally intersects each of the $`\alpha `$-planes once, at particular six-dimensional hyperplanes invariant under both $`\alpha `$ and $`\beta `$. Chiral anomalies are induced on the $`\alpha `$-planes and (separately) on the $`\alpha \beta `$-planes, due to localized chiral projections of fields. Cancellation of the ten-dimensional $`\alpha `$-plane anomalies is uniquely accomplished by an additional ten-dimensional $`E_8`$ Yang-Mills supermultiplet on each of the two $`\alpha `$-planes, as is well-known. In this paper, we concern ourselves with the six-dimensional $`\alpha \beta `$-plane anomalies.
Although some of our discussion will be more general, it is helpful to have a specific orbifold in mind to help visualize the basic geometric setting. Our prototype is the simplest of the orbifolds described above, namely, the $`S^1/𝐙_2\times T^4/𝐙_2`$ orbifold. In this case, spacetime has topology $`𝐑^6\times T^5`$ and each of the five compact coordinates takes values on the interval $`[\pi ,\pi ]`$ with the endpoints identified. The nontrivial projections are $`\alpha :(x^\mu ,x^i,x^{11})(x^\mu ,x^i,x^{11})`$ and $`\beta :(x^\mu ,x^i,x^{11})(x^\mu ,x^i,x^{11})`$, where $`x^\mu `$ parameterizes the six noncompact dimensions, while $`x^i`$ and $`x^{11}`$ parameterize the $`T^4`$ and $`S^1`$ factors respectively. The element $`\alpha `$ leaves invariant the two ten-planes defined by $`x^{11}=0`$ and $`x^{11}=\pi `$, while $`\beta `$ leaves invariant the sixteen seven-planes defined when the four coordinates $`x^i`$ individually assume the values $`0`$ or $`\pi `$. Finally, $`\alpha \beta `$ leaves invariant the thirty-two six-planes defined when all five compact coordinates individually assume the values $`0`$ or $`\pi `$. The $`\alpha \beta `$ six-planes coincide with intersections of the $`\alpha `$ ten-planes with the $`\beta `$ seven-planes. The global structure is depicted in Figure 1.
There are several magnetic and electric sources for $`G`$ necessary to resolve chiral anomalies in these orbifolds. The basic Chern-Simons terms include the $`CGG`$ interaction and also the higher-derivative $`GX_7`$ interaction, where $`X_8dX_7`$ is the eight-form describing the worldvolume anomaly generated by the fivebrane zero modes. The worldvolume anomaly is cancelled by inflow mediated by the $`GX_7`$ interaction, provided the fivebrane acts as a magnetic source for $`G`$, in the sense described above. Next, each of the two $`\alpha `$-planes supports a ten-dimensional $`E_8`$ super-Yang-Mills multiplet. They also provide magnetic sources for $`G`$ due to the presence of terms $`\delta ^{(1)}I_4`$ in the $`dG`$ bianchi identity, where $`\delta ^{(1)}`$ is a one-form brane-current localized on the $`\alpha `$-plane and $`I_4`$ is a four-form polynomial involving the Lorentz-valued curvature $`R`$ and the local $`E_8`$ field-strength $`F`$.
The seven-dimensional $`\beta `$-planes provide electric sources for $`G`$ via Chern-Simons interactions $`\delta ^{(4)}GY_3^0`$, where $`\delta ^{(4)}`$ is a four-form brane-current localized on the seven-plane and $`Y_4=dY_3^0`$ is a gauge-invariant four-form polynomial. This polynomial involves the curvature $`R`$ and also a field strength $``$ associated with additional adjoint super-gauge fields localized on the seven-plane (with the gauge group determined by anomaly cancellation in a manner which we will describe). This coupling gives rise to an “I-brane” effect via interplay with the ten-dimensional magnetic source (involving $`I_4`$). This contributes additional inflow localized on the six-dimensional intersection of the ten-dimensional $`\alpha `$-plane and the seven-dimensional $`\beta `$-plane <sup>1</sup><sup>1</sup>1See Section 5 of for a description of this effect.
The magnetic and electric sources described so far are encapsulated by the following three polynomials
$`X_8(R)`$ $`=`$ $`{\displaystyle \frac{1}{(2\pi )^3\mathrm{\hspace{0.17em}4}!}}\left(\frac{1}{8}\mathrm{tr}R^4\frac{1}{32}(\mathrm{tr}R^2)^2\right)`$
$`I_4(R,F)`$ $`=`$ $`{\displaystyle \frac{1}{16\pi ^2}}\left(\frac{1}{2}\mathrm{tr}R^2+\mathrm{tr}F^2\right)`$
$`Y_4(R,)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}\left(\frac{1}{32}\eta \mathrm{tr}R^2+\rho \mathrm{tr}^2\right).`$ (1)
The precise forms of $`X_8`$ and $`I_4`$ are fixed by fivebrane consistency and ten-dimensional anomaly cancellation, respectively. The polynomial $`Y_4`$ is parameterized by two rational coefficients $`\eta `$ and $`\rho `$. These are determined by further requirements described below.
Finally, the six-dimensional $`\alpha \beta `$-planes carry a magnetic charge. This appears in the $`dG`$ Bianchi identity as a term $`g\delta ^{(5)}`$, where $`\delta ^{(5)}`$ is a five-form brane-current localized on the $`\alpha \beta `$-plane and $`g`$ is a rational magnetic charge subject to a quantization condition. For the case $`M=2`$, the magnetic charge $`g`$ should be quarter-integer, as explained in . The Bianchi identity for the four-form field strength $`G`$ is, therefore, given by
$`dG={\displaystyle \underset{i=1}{\overset{2}{}}}I_{4(i)}\delta _{M_i^{10}}^{(1)}+{\displaystyle \underset{i=1}{\overset{2f}{}}}g_i\delta _{M_i^6}^{(5)}+{\displaystyle \underset{i=1}{\overset{N_5}{}}}\delta _{W_i^6}^{(5)},`$ (2)
where we have included all of the magnetic sources described above. The manifold $`M_i^{10}`$ is the $`i`$th $`\alpha `$-plane, while $`M_i^6`$ is the $`i`$th $`\alpha \beta `$-plane and $`W_i^6`$ is the worldvolume of the $`i`$th fivebrane <sup>2</sup><sup>2</sup>2The pervasive use of the label $`i`$ is merely convenient, and does not imply any specific correlation between these manifolds.. In most of this paper, our expressions apply to a particular $`\alpha \beta `$-plane. Hence, the label $`i`$ is implicit but omitted.
The projection $`\beta `$ can independently break the ten-dimensional $`E_8`$ gauge groups on the $`\alpha \beta `$-planes to maximal subgroups. Since a ten-dimensional vector supermultiplet decomposes into one six-dimensional $`N`$=1 vector and one six-dimensional hypermultiplet, the breaking pattern will be characterized by an integer number $`V_B`$ of vector multiplets and another integer number $`H_B`$ of hypermultiplets, each transforming according to some representation $``$ of the residual maximal subgroup of $`E_8`$. The projection $`\beta `$ necessarily removes half of the $`E_8`$ degrees of freedom. But the identity of which half depends on how $`\beta `$ acts on the $`E_8`$ root lattice.
Chiral projection of the supergravity fields results in further contributions to the local $`\alpha \beta `$ anomalies. These derive from “untwisted” fields comprising one universal $`N`$=1 tensor multiplet and some number $`h`$ of hypermultiplets. The value of $`h`$ depends on which $`Z_M`$ orbifold is being considered. For the cases $`M`$=2 and $`M`$=3, we have $`h`$=4 and $`h`$=2 respectively. Furthermore, since the local anomaly due to the supergravity and residual $`E_8`$ fields arises from the coupling of fields which are not themselves localized on the $`\alpha \beta `$-planes, it involve fractions which reflect the multiplicies of the fixed planes. We parameterize this by another integer $`f`$ corresponding to the number of $`\beta `$-planes associated with the orbifold in question. For the cases $`M`$=2 and $`M`$=3, we have $`f`$=16 and $`f`$=9 respectively. These correspond to the respective sixteen and nine fixed-planes in the $`Z_2`$ and $`Z_3`$ orbifold limits of the $`K3`$ manifold <sup>3</sup><sup>3</sup>3The $`Z_4`$ and $`Z_6`$ orbifolds involve additional subtlety which we will not discuss in this paper..
Finally, we allow for as yet unspecified $`N`$=1 supermatter localized on each $`\alpha \beta `$-plane. We call this matter “twisted”, since it is analogous to twisted sector matter in superstring orbifolds. This matter assembles into $`n_T`$ tensor multiplets, $`n_V`$ vector multiplets and $`n_H`$ hypermultiplets, and involves an as yet undetermined “twisted” gauge group $`\stackrel{~}{𝒢}`$. The vector multiplets transform in the adjoint representation, while the hypermultiplets transform in an unspecified representation $`\stackrel{~}{}`$.
We focus on a particular $`\alpha \beta `$-plane, and assemble each of the contributions to the six-dimensional anomaly localized on this plane. There are three classical (inflow) contributions, described by the polynomials
$`I_8(CGG)`$ $`=`$ $`\pi gI_4(R,F)^2`$
$`I_8(GX_7)`$ $`=`$ $`gX_8(R)`$
$`I_8(IB)`$ $`=`$ $`I_4(R,F)Y_4(R,).`$ (3)
The first two arise from the variation of the $`CGG`$ and $`GX_7`$ terms. The third arises from the variation of $`\delta ^{(4)}GY_3^0`$ and describes the “I-brane” anomaly <sup>4</sup><sup>4</sup>4The necessity for the “I-brane” contribution in an orbifold context was first recognized in , and was inspired by an analogous effect on intersecting D-branes, introduced in .. There are also three quantum contributions
$`I_8(SG)`$ $`=`$ $`{\displaystyle \frac{1}{2f}}\left(I_{GRAV}^{(3/2)}(R)(1+h)I_{GRAV}^{(1/2)}(R)\right)`$
$`I_8(𝒢)`$ $`=`$ $`{\displaystyle \frac{1}{f}}\left((V_BH_B)I_{GRAV}^{(1/2)}(R)+I_{MIXED}^{(1/2)}(R,F)_{}+I_{GAUGE}^{(1/2)}(F)_{}\right)`$
$`I_8(\stackrel{~}{𝒢})`$ $`=`$ $`(n_Vn_Hn_T)I_{GRAV}^{(1/2)}(R)n_TI_{GRAV}^{(3\mathrm{form})}(R)`$ (4)
$`+I_{MIXED}^{(1/2)}(R,)_\stackrel{~}{}+I_{GAUGE}^{(1/2)}()_\stackrel{~}{}.`$
The factors $`I_{\mathrm{GRAV}}`$, $`I_{MIXED}`$ and $`I_{GAUGE}`$ which appear in the quantum anomalies describe one-loop gravitational, mixed and pure-gauge anomalies. They are attributable to the type of chiral fields with spin indicated by the superscripts. These are determined by index theorems and are listed explicitly in appendix C of . The anomaly $`I_8(𝒢)`$ describes the local anomaly involving whatever subgroup $`𝒢E_8`$ is left unbroken by $`\beta `$ on the relevant $`\alpha \beta `$-plane. The subscript $``$ indicates the representation content of the $`E_8`$ residual subgroup. Hence, the traces over the gauge factors in $`I_{MIXED}^{(1/2)}(R,F)_{}`$ and $`I_{GAUGE}^{(1/2)}(F)_{}`$ are traces over $``$. Similar comments apply to the $`\alpha \beta `$ twisted sector with gauge group $`\stackrel{~}{𝒢}`$, where the representation content is indicated by the subscript $`\stackrel{~}{}`$. <sup>5</sup><sup>5</sup>5The technical aspects involved in determining (3) and (4) are described in . The determination of $`I_8(CGG)`$ is more subtle than indicated in that paper, however, for reasons mentioned previously. See for a more thorough description of this one term. Note that the twisted gauge group $`\stackrel{~}{𝒢}`$ can include factors which coincide with some factors in the $`E_8`$ residual group $`𝒢`$.
There are numerous unspecified parameters involved in the six contributions in (3) and (4). To begin with, there are ten integers characterizing the global geometry of the orbifold, the local $`E_8`$ breaking pattern and multiplicities in the twisted and untwisted spectra. For instance, the orbifold geometry is partially encoded in the number $`f`$ of $`\beta `$-fixed planes and the number $`h`$ of universal untwisted hypermultiplets. The magnetic charge $`g`$ is an integer times a basic quantization unit (which also reflects the orbifold geometry). The two parameters $`\eta `$ and $`\rho `$ describe electrical charges of the seven-planes, as defined in equation (1). The $`E_8`$ breaking pattern is encoded in the multiplicities $`V_B`$ and $`H_B`$. Finally, the local twisted spectrum is encoded in the multiplicities $`n_V,n_H`$ and $`n_T`$.
Furthermore, there are six sets of rational parameters which characterize the representation content of the matter fields. These are given by “representation indices”, which allow one to relate traces over a given representation to traces over the fundamental representation. The relevant representation indices are denoted $`_2()`$, $`_{2,2}()`$ and $`_4()`$, and are defined by
$`\mathrm{tr}_{}F^2`$ $`=`$ $`_2()\mathrm{tr}F^2`$
$`\mathrm{tr}_{}F^4`$ $`=`$ $`_{2,2}()(\mathrm{tr}F^2)^2+_4()\mathrm{tr}F^4,`$ (5)
where all traces on the right-hand side are over the fundamental representation. Some useful representation indices are listed in Table 1 <sup>6</sup><sup>6</sup>6The evaluation of the representation indices for arbitrary representations of the classical Lie groups is a complicated problem. See for a discussion of this issue..
Remarkably, each and every one of the above parameters can be resolved, in a sense to be made clear, by imposing necessary factorization properties on the net anomaly obtained by summing the six contributions listed in (3) and (4). That result describes the total local anomaly on a particular $`\alpha \beta `$-plane due to the sources which we have described so far. Since the total local anomaly must vanish, there are two possibilities. The first is that the net anomaly vanishes identically. The remaining possibility is that this net anomaly is non-zero, but is cancelled by a contribution arising from special couplings of local tensor fields which, thereby, provide a local Green-Schwarz mechanism.
Local tensor fields reside in twisted $`N`$=1 tensor multiplets on the $`\alpha \beta `$-plane. The two-form fields in such a multiplet are anti-self-dual, in the sense that the associated gauge-invariant three-form field-strength satisfies $`H=H`$. As a result of this, a special Chern-Simons interaction $`\delta ^{(5)}HZ_3^0`$, where $`\delta ^{(5)}`$ is the five-form brane-current with support on the $`\alpha \beta `$-plane and $`Z_4=dZ_3^0`$ is a gauge-invariant four-form polynomial, will lead to two consequences. First, due to the anti-self-duality, the electric coupling described by the Chern-Simons interaction implies a magnetic coupling described by the Bianchi identity $`dH=Z_4`$. The magnetic coupling involves the same polynomial as the electric coupling because anti-self-duality is equivalent to an electric-magnetic duality <sup>7</sup><sup>7</sup>7This is easy to see by taking the exterior derivative of the Euler-Lagrange equation $`H=Z_3`$, and then replacing $`H`$ with $`H`$.. Secondly, the Chern-Simons coupling generates an inflow anomaly described by $`I_8(GS)=Z_4Z_4`$, where $`GS`$ designates Green-Schwarz. Note that this polynomial is a perfect square due to the duality. This is, in turn, a consequence of $`N`$=1 supersymmetry. Thus, the net anomaly can be cancelled by specialized local tensor dynamics, provided that the net anomaly reduces to a sum of perfect squares with one term for each available tensor field.
## 3 Local Anomaly Cancellation
As a result of the above discussion, a program for analyzing the local anomaly on a given $`\alpha \beta `$-plane becomes apparent. First, we assemble the net anomaly by summing the six terms in (3) and (4). We will call this result $`I_8`$. Then we sequence through the possibilities $`n_T`$=0,1,2,…, in each case imposing that $`I_8`$ satisfies the appropriate factorization requirement. For the case $`n_T`$=0, since there are no local tensors, we require $`I_8`$=0. For the case $`n_T`$=1, we impose that $`I_8`$ is proportional to a complete square, that is, $`I_8(Z_4)^2`$. (In this second case, we simultaneously determine the form of the electric and magnetic couplings of the local tensor.) For the case $`n_T`$=2, we require that $`I_8`$ is the sum of two perfect squares. And so forth. These factorization requirements prove to be marvelously restrictive. For each choice of $`n_T`$, there results unambiguous values for each and every one of the previously unspecified geometric and topological parameters, including the values of the magnetic charge, electric charges, and the identity of the gauge groups and representation content of the twisted sectors.
Since M-fivebranes carry unit magnetic charge, as well as the zero mode fields described previously, we infer that a fivebrane moving onto (or off of) a given $`\alpha \beta `$-plane deposits (or removes) charge and twisted modes in the process of doing so. Since the fivebrane carries one $`N`$=1 tensor multiplet and one hypermultiplet, we expect that the solutions to our factorization constraints will assemble into hierarchies linked by incrementing the magnetic charge and the local tensor and hyper multiplicities as $`gg+1`$, $`n_Tn_T+1`$ and $`n_Hn_H+1`$. This is precisely what we find. Mathematically, this can be understood as follows. A single fivebrane touching an orbifold fixed plane is described by a singular object called a “torsion free sheaf” . This sheaf carries one extra unit of magnetic charge and has one tensor multiplet and one hypermultiplet as zero modes, identical to the analogous fivebrane data. This accounts for the increase in each of these quantities by unity when the fivebrane is moved to a fixed plane.
At least in some cases, the torsion free sheaf can be shown to be the singular “small instanton” limit of a smooth gauge instanton . Smoothing out the sheaf into an instanton represents a true phase transition, where the fivebrane data disappears and is replaced by a vector bundle . In this process, the unit magnetic charge of the sheaf is replaced by a unit increase in the second Chern number of the vector bundle. The zero modes of the vector bundle are, in general, quite different from those of the torsion free sheaf. Most importantly, the appearance of a smooth, non-trivial vector bundle signals the breakdown of the original twisted sector gauge group to a smaller group $`𝒢E_8`$. Therefore, after this phase transition, we expect to have a smaller twisted sector gauge group with identical topological charge but different numbers of tensor and hypermultiplets. Be that as it may, this theory remains anomaly free. As we will see, locally anomaly free orbifold planes do exist that could be related to each other through small instanton phase transitions. Thus, factorization of the local anomaly polynomial yields an extra bonus.
In addition to the torsion free sheaf transitions described earlier, we expect another interesting grouping of our factorization solutions. In this second grouping, we expect small instanton transitions between sets of solutions with identical local magnetic charge, but with different numbers of zero modes and different local gauge groups.
For clarity, let us recapitulate the fivebrane transitions discussed in the previous two paragraphs. If a fivebrane moves to a particular $`\alpha \beta `$-plane, it should increment the local magnetic charge by one, and add one local tensor and one local hypermultiplet to the associated twisted spectrum. We would then attribute one unit of the total local magnetic charge to the latent magnetic charge of the fivebrane, now interpreted as a torsion-free sheaf. The associated tensor would be available to help mediate anomaly cancellation through a local Green-Schwarz mechanism. Now, assume this configuration is a small instanton and can be deformed to a smooth vector bundle. Then, since the instanton does not have a tensorial zero mode, there would be one less tensor available to mediate the anomaly cancellation. The anomaly polynomial should, therefore, reconfigure so as to ensure continued anomaly cancellation, but with a modified factorization criterion. Thus, by classifying independent solutions to the factorization requirements into sets involving identical values of $`g`$ but different numbers of tensor fields, one can infer such nontrivial phase transitions.
We begin our analysis by considering a particular $`\alpha \beta `$ plane, corresponding to one of the solid dots in Figure 1, representing the unique six-dimensional intersection of a particular ten-dimensional $`\alpha `$-plane and a particular seven-dimensional $`\beta `$-plane. To be more concrete, we focus on one of the intersection points on the lower of the two $`\alpha `$-planes in Figure 1, so that the local geometry is depicted as in Figure 2.
In Figure 2, the horizontal line represents the $`\alpha `$-plane, the vertical line represents the $`\beta `$-plane and, finally, the point of intersection represents the $`\alpha \beta `$-plane. The intersection supports the local anomaly in which we are interested and has magnetic charge $`g`$. The $`\alpha `$-plane supports $`E_8`$ super Yang-Mills fields, as described above. This $`E_8`$ group is, in general, broken on the $`\alpha \beta `$-plane to some subgroup depending of the action of $`\beta `$ on the $`E_8`$ root lattice. In the vertex-diagrams below, we indicate, to the right of the horizontal lines, the subgroup of $`E_8`$ left unbroken at the intersection. Thus, Figure 2 indicates a scenario in which the full $`E_8`$ is left unbroken. This figure also indicates the presence of additional gauge structure with group $`\stackrel{~}{𝒢}_7`$ localized on the $`\beta `$-plane and further gauge structure on the $`\alpha \beta `$-plane with group $`\stackrel{~}{𝒢}_6`$. These correspond to additional seven-dimensional and six-dimensional fields, respectively.
The only multiplet in seven dimensions is a vector multiplet, which transforms in the adjoint of $`\stackrel{~}{𝒢}_7`$. This decomposes into one six-dimensional $`N`$=1 vector multiplet and one six-dimensional hypermultiplet <sup>8</sup><sup>8</sup>8As a point of interest, this is the same decomposition enjoyed by a ten-dimensional vector multiplet.. An important fact is that the seven-dimensional fields are chirally projected by $`\alpha `$ onto the embedded six-dimensional $`\alpha \beta `$-plane. It, thereby, contributes to the local $`\alpha \beta `$ anomaly.
If $`\stackrel{~}{𝒢}_7`$ does not coincide with some factor of the broken subgroup of $`E_8`$, then six-dimensional gauge invariance dictates that the hypermultiplet is the part projected out by $`\alpha `$. Thus, the $`\stackrel{~}{𝒢}_7`$ gauge fields remain at the intersection to enforce local $`\stackrel{~}{𝒢}_7`$ invariance. The $`\stackrel{~}{𝒢}_7`$ adjoint gauginos fill out a six-dimensional $`N`$=1 vector multiplet, which is indicated by the $`V`$ next to the downward arrow in Figure 2. This tells us that the “vector” part survives the $`\alpha `$ projection as we move down the vertical line in that diagram and land on the intersection point. Since the $`N`$=1 vector multiplet is chiral, the $`\stackrel{~}{𝒢}_7`$ gauginos will contribute to the $`\alpha \beta `$ anomaly. However, since this anomaly results from the coupling of fields not localized on the $`\alpha \beta `$-plane, this anomaly must be divided by two (since there are two $`\alpha \beta `$-planes embedded within a given $`\beta `$-plane) compared to a similar anomaly due to six-dimensional $`\stackrel{~}{𝒢}_7`$ adjoint gauginos.
If $`\stackrel{~}{𝒢}_7`$ coincides with a factor in the unbroken subgroup of $`E_8`$, then the $`\stackrel{~}{𝒢}_7`$ gauge fields on the $`\alpha \beta `$-plane may be supplied by the ten-dimensional gauge fields which survive the $`\beta `$ projection. Consequently, the projection $`\alpha `$ should remove the “vector” part of the seven-dimensional adjoint matter, so that the other part, corresponding to an adjoint hypermultiplet, survives the $`\alpha `$ projection. We indicate this alternate situation by an $`H`$ next to the downward arrow in the corresponding diagram. In this case, the surviving hyperinos would contribute to the local anomaly. This anomaly would include a division by two compared to a similar anomaly due to six-dimensional $`\stackrel{~}{𝒢}_7`$ adjoint hyperinos, for reasons identical to those described in the preceeding paragraph <sup>9</sup><sup>9</sup>9The possibility of alternatively projecting out the vector or hypermultiplet parts of the seven-dimensional fields on the six-dimensional planes was also mentioned in ..
A third possibility is that the group $`\stackrel{~}{𝒢}_7`$ is broken, by $`\beta `$, to some maximal subgroup $`\stackrel{~}{𝒢}_7`$ on the $`\alpha \beta `$-plane. In this case, the seven-dimensional fields would decompose into various six-dimensional fields transforming according to representations determined by the appropriate branching rule. These would include fields transforming in the adjoint of $``$ and other fields transforming in other representations of $``$. The vector part of the adjoint fields would survive the $`\beta `$ projection while the hypermultiplet part of the remaining fields would survive. We indicate this hybrid situation by replacing the $`V`$ in Figure 2 with the relevant subgroup $`\stackrel{~}{𝒢}_7`$ which survives the projection.
In resolving the factorization criteria necessary to explain local anomaly cancellation on a given $`\alpha \beta `$-plane, one requires factors of two which can only be explained by seven-dimensional matter in the manner we have just described. Thus, the identity of seven-dimensional matter is indicated by anomaly cancellation on an embedded sub-plane. This is interesting because the seven-plane itself cannot support a local anomaly since it is odd-dimensional. (The situation is analogous to the fact that eleven-dimensional supergravity is needed by the $`E_8`$ super-gauge multiplets on the two $`\alpha `$-planes to render those ten-planes anomaly-free.)
There is one more subtle factor of one-half which needs explanation. This relates to hypermultiplets in the six-dimensional twisted sector. It is possible for $`r`$ hypermultiplets to transform according to a $`2r`$ dimensional representation $`\stackrel{~}{}`$ of the gauge group $`\stackrel{~}{𝒢}`$, provided the representation is “pseudoreal” in a sense to be clarified. In this case, the $`4r`$ scalar fields assemble into $`r`$ quaternions represented as $`\varphi _i^\alpha `$, where $`i=1,2`$ is an index which spans the 2 represention of an $`Sp(1)`$ automorphism of the supersymmetry algebra and $`\alpha =1,\mathrm{},2r`$ spans $`\stackrel{~}{}`$. The group $`\stackrel{~}{𝒢}`$ acts as $`\delta \varphi _i^\alpha =\theta ^a(T_a)_\beta ^\alpha `$, where $`(T_a)_\beta ^\alpha `$ are antihermitian generators. There exists a real invariant tensor $`\rho ^{\alpha \beta }=\rho _{\alpha \beta }`$ which, by suitable field redefinition, can be put into block-diagonal form $`\rho =\mathrm{diag}(i\sigma _2\mathrm{}i\sigma _2)`$, where $`\sigma _2`$ is the second Pauli matrix. The representation is pseudoreal if $`(T_a^{})=\rho T_a\rho `$. See for a more comprehensive discussion of hypermultiplets. In this case, we refer more properly to $`2r`$ half-hypermultiplets, since the number of hypermultiplets is half the dimensionality of the representation. Each such half-hypermultiplet then contributes one-half of the anomaly which we would normally attribute to $`2r`$ antichiral spinors transforming in $`\stackrel{~}{}`$ via naive application of index theorems.
Next, we should explain how to algebraically characterize the possible branching patterns describing the projection of the two $`E_8`$ factors by $`\beta `$. The simplest possibility is the one indicated in Figure 2, where the relevant $`E_8`$ factor remains unbroken. In terms of six-dimensional $`N`$=1 multiplets, the ten-dimensional $`E_8`$ vector multiplet decomposes into one vector multiplet and one hypermultiplet. In this case, it is the hypermultiplet components which are projected out by $`\beta `$ on the $`\alpha \beta `$-plane. This leaves us with the gauge fields necessary to enforce local $`E_8`$ invariance on the $`\alpha \beta `$-plane. This also fixes two of our parameters, $`(V_B,H_B)=(248,0)`$. In this case, all $`E_8`$ traces which appear in both the inflow anomalies (3) and the quantum anomalies (4) are each taken at face value. Thus, using the representation indices in Table 1, we can use the results $`\mathrm{Trace}_{\mathrm{𝟐𝟒𝟖}}F^230\mathrm{tr}F^2`$ and $`\mathrm{Trace}_{\mathrm{𝟐𝟒𝟖}}F^4=9(\mathrm{tr}F^2)^2`$ to express the traces which appear in the quantum anomalies (4) in terms of the fundamental ($`\mathrm{tr}`$) traces <sup>10</sup><sup>10</sup>10Note that the traces which appear in the inflow anomaly (3), through the implicit dependence of (1), are fundamental traces to begin with..
More generally, the $`E_8`$ factors will be broken by $`\beta `$, on the $`\alpha \beta `$-planes, to some maximal subgroup with a branching pattern which can be found from the tables in . In this case, when we determine our anomaly polynomial $`I_8`$ by adding up the six contributions in (3) and (4), we replace the various $`E_8`$ traces by traces over the relevant representations of the residual subgroup. We then relate these to traces over fundamental representations of the factors in this subgroup by using representation indices, such as those listed in Table 1. However, there is a subtle difference between the inflow contributions (3) and the quantum contributions (4) which should be properly accounted for, and which we now describe. Since the inflow anomalies are classical expressions, we can apply the group theoretic reduction directly on the traces which appear in (3). However, in the general case, the six-dimensional quantum anomalies derive from both chiral and antichiral fields. The chiral fields, which satisfy $`\mathrm{\Gamma }_7\psi =\psi `$, appear in the six-dimensional $`N`$=1 vector multiplets. The antichiral fields, which satisfy $`\mathrm{\Gamma }_7\psi =\psi `$, occur in hypermultiplets. Since chiral and antichiral fields contribute one-loop anomalies with opposite sign, there is an extra minus sign associated with all quantum anomalies arising from hypermultiplet couplings. We illustrate this with two explicit examples.
As a first example, we choose the breaking pattern $`E_8E_7\times SU(2)`$. In this case, we have the branching rule $`\mathrm{𝟐𝟒𝟖}=(\mathrm{𝟏𝟑𝟑},\mathrm{𝟏})(\mathrm{𝟏},\mathrm{𝟑})(\mathrm{𝟓𝟔},\mathrm{𝟐})`$. We determine that the surviving six-dimensional fields comprise 133 $`N`$=1 vector multiplets transforming as the adjoint of $`E_7`$, another three vector multiplets transforming as the adjoint of $`SU(2)`$ and 112 hypermultiplets transforming as a bifundamental representation. Thus, $`(V_B,H_B)=(136,112)`$. In this case, we reduce the $`E_8`$ traces which occur in the inflow anomaly as follows
$`\mathrm{tr}F^2`$ $`=`$ $`\frac{1}{30}\mathrm{Tr}_{\mathrm{𝟐𝟒𝟖}}F^2`$ (6)
$`=`$ $`\frac{1}{30}\left(\mathrm{Tr}_{\mathrm{𝟏𝟑𝟑}}F_a^2+2\mathrm{tr}_{\mathrm{𝟓𝟔}}F_a^2+\mathrm{Tr}_\mathrm{𝟑}F_b^2+56\mathrm{tr}_\mathrm{𝟐}F_b^2\right)`$
$`=`$ $`\frac{1}{6}\mathrm{tr}F_a^2+2\mathrm{tr}F_b^2,`$
where we have used the indices in Table 1. In (6), the subscripts $`a`$ and $`b`$ denote $`E_7`$ and $`SU(2)`$ respectively. In the final line, we have dropped the labels from the fundamental $`\mathrm{𝟓𝟔}`$ and $`\mathrm{𝟐}`$ traces. Thus, to describe the inflow anomaly in the case where the $`E_8`$ factor is broken by $`\beta `$ to $`E_7\times SU(2)`$, we substitute the identity (6) for the $`\mathrm{tr}F^2`$ in the inflow anomaly (3). In the quantum anomaly, on the other hand, hypermultiplets and vector multiplets contribute with opposite signs. As a result, when computing the local one-loop anomaly in the case where $`E_8E_7\times SU(2)`$, we we should replace the factor $`\mathrm{trace}F^2`$ which appears in $`I_{\mathrm{MIXED}}^{(1/2)}`$ using the following <sup>11</sup><sup>11</sup>11See equation (C.2) of for the explicit polynomial corresponding to $`I_{MIXED}^{(1/2)}`$, as well as all of the other quantum anomaly polynomials referred to in this paper.
$`\mathrm{trace}F^2`$ $`=`$ $`\mathrm{Tr}_{\mathrm{𝟐𝟒𝟖}}F^2`$ (7)
$`=`$ $`\mathrm{Tr}_{\mathrm{𝟏𝟑𝟑}}F_a^22\mathrm{tr}_{\mathrm{𝟓𝟔}}F_a^2+\mathrm{Tr}_\mathrm{𝟑}F_b^256\mathrm{tr}_\mathrm{𝟐}F_b^2`$
$`=`$ $`\mathrm{tr}F_a^252\mathrm{tr}F_b^2.`$
This derivation differs from (6) by the minus signs on terms relating to hypermultiplet couplings. As described above, these minus signs reflect the antichirality of hyperinos. Similar comments apply to the term $`\mathrm{trace}F^4`$ which appears in $`I_{\mathrm{GAUGE}}^{(1/2)}`$. In total, to describe the case $`E_8E_7\times SU(2)`$ we should make the following replacements
$`\mathrm{tr}F^2`$ $`=`$ $`\frac{1}{6}\mathrm{tr}F_a^2+2\mathrm{tr}F_b^2`$
$`\mathrm{trace}F^2`$ $`=`$ $`\mathrm{tr}F_a^252\mathrm{tr}F_b^2`$
$`\mathrm{trace}F^4`$ $`=`$ $`\frac{1}{12}(\mathrm{tr}F_a)^220(\mathrm{tr}F_b)^26\mathrm{tr}F_a^2\mathrm{tr}F_b^2.`$ (8)
The first of these should be substituted in the classical (inflow) anomaly, while the second two should be substituted in the quantum anomaly. Note our mnemonic that traces which appear in the quantum anomaly are designated “trace”, whereas classical traces are abbreviated “tr”.
As a second example, we choose the breaking pattern $`E_8SO(16)`$. In this case, we have the branching rule $`\mathrm{𝟐𝟒𝟖}\mathrm{𝟏𝟐𝟎}\mathrm{𝟏𝟐𝟖}`$. We determine that the surviving six-dimensional fields comprise 120 $`N`$=1 vector multiplets transforming as the adjoint and 128 hypermultiplets transforming as the spinor of $`SO(16)`$. Thus, $`(V_B,H_B)=(120,128)`$. In this case, we reduce the $`E_8`$ traces which occur in the inflow anomaly and in the quantum anomaly in a manner similar to that described in our previous example, making use of the representation indices in Table 1. The appropriate reductions are
$`\mathrm{tr}F^2`$ $`=`$ $`\mathrm{tr}F_a^2`$
$`\mathrm{trace}F^2`$ $`=`$ $`2\mathrm{tr}F_a^2`$
$`\mathrm{trace}F^4`$ $`=`$ $`3(\mathrm{tr}F_a)^2+16\mathrm{tr}F_a^4,`$ (9)
where the subscript $`a`$ now denotes $`SO(16)`$. Once again, the first of these should be substituted in the classical (inflow) anomaly, while the second two should be substituted in the quantum anomaly.
Using the three distinct cases which we have so far addressed, corresponding to the choices where $`\beta `$ breaks $`E_8`$ to $`E_8`$, $`E_7\times SU(2)`$ or $`SO(16)`$, we have enough data to completely determine an interesting set of solutions to our anomaly factorization problem. These solutions conform to our expectations by assembling into hierachies as described previously.
On a given $`\alpha \beta `$-plane, such as that depicted by the intersection point in Figure 1, the net six-dimensional anomaly $`I_8`$ is determined by adding up all six terms in (3) and (4). One then substitutes identities, such as (8) and (9), relevant to the particular $`E_8`$ breaking pattern being considered, in the manner explained above. What results is a polynomial with terms proportional to the each of $`\mathrm{tr}R^4`$, $`(\mathrm{tr}R^2)^2`$, $`\mathrm{tr}R^2\mathrm{tr}F^2`$, $`\mathrm{tr}R^2\mathrm{tr}F^2`$, $`(\mathrm{tr}F^2)^2`$, $`\mathrm{tr}F^2\mathrm{tr}^2`$, $`\mathrm{tr}F^4`$, and $`\mathrm{tr}^4`$, where $`F`$ stands generically for factors in the residual group $`𝒢E_8`$ which survives the $`\beta `$ projection (we have denoted these $`F_a`$ and $`F_b`$ above) and $``$ stands generically for factors in any gauge group associated with twisted matter.
Note that we have allowed for twisted fields which are either six or seven dimensional. In the former case, we refer to $`N`$=1 fields living exclusively on the $`\alpha \beta `$-plane under consideration. In the latter, we refer to vector adjoint supermultiplets living on the seven-dimensional $`\beta `$-plane which intersects this $`\alpha \beta `$-plane. The seven-dimensional fields will contribute to the anomaly with a tell-tale factor of two, as described above. Computationally, we accomodate both of these cases simultaneously by formally allowing $`n_T`$ tensors, $`n_V`$ vectors and $`n_H`$ hypermultiplets, where $`n_V`$ and $`n_H`$ can assume half-integral values. The (formal) appearance of half-integer numbers of multiplets then indicates that the associated matter is seven-dimensional.
Keeping the twisted matter arbitrary, we determine $`I_8`$ by adding up all six contributions in (3) and (4). For any choice of $`𝒢E_8`$, we reduce the various $`E_8`$ traces to $`𝒢`$ traces according to the scheme described above. This process provides us with a provisional form of the local anomaly. It remains provisional since the twisted contribution remains to be resolved. Nevertheless, we can extract our first bits of useful information. Since anomaly cancellation is possible only if $`I_8`$ either vanishes identically or reduces to a sum of perfect squares, it follows that any nonfactorizable terms in $`I_8`$ must vanish. The vanishing of the $`\mathrm{tr}R^4`$ term requires
$`n_Hn_V=30g29n_T+\frac{1}{2f}(244h)+\frac{1}{f}(V_BH_B).`$ (10)
This constraint is the local version of the global constraint $`N_HN_V+29N_T=273`$, where $`N_H`$, $`N_V`$ and $`N_T`$ are the total number of hyper, vector and tensor multiplets in the entire orbifold, including all twisted and untwisted contributions. The relationship between the local constraint (10) and the global version is described in . Note that (10) is invariant when $`gg+1`$, $`n_Tn_T+1`$, and $`n_Hn_H+1`$, consistent with expectations described previously.
Henceforth, we concentrate on the $`S^1/𝐙_2\times T^4/𝐙_2`$ orbifold. In this case, we have $`(f,h)=(16,4)`$, as described above, so that (10) becomes
$`n_Hn_V=30g29n_T+\frac{15}{2}+\frac{1}{16}(V_BH_B).`$ (11)
Since the left-hand side must be either an integer or a half-integer, it follows that $`(V_BH_B)/16`$ must be integer or half-integer as well, since in this case $`g`$ is quantized in quarter-integer units. Each of the three $`E_8`$ breaking patterns which we have addressed, $`𝒢=E_8,E_7\times SU(2)`$ and $`SO(16)`$, corresponding to $`(V_B,H_B)=(248,0),(136,112)`$ and $`(120,128)`$ respectively, respect this constraint. We restrict our study to these three breaking patterns.
The sytematic analysis of local anomaly cancellation proceeds as outlined above. We first seek solutions with no twisted tensor multiplets, so that $`n_T=0`$. In this case, $`I_8`$ must vanish identically. We consider each of the three possible $`E_8`$ breaking patterns described above, using the relevant values of $`V_B`$ and $`H_B`$ in each case. For these three possibilities, when $`n_T=0`$ equation (11) reduces to the constraints indicated in Table 2.
Thus, if there are no local twisted tensor multiplets, anomaly cancellation implies the indicated correlations between the local magnetic charge and the multiplicities of twisted hyper and vector multiplets. Note that, in each case, extra twisted matter is required since the indicated multiplicites can not be made to vanish with any properly quantized choice of $`g`$. The challenge in the case $`n_T`$=0 is not only to identify the multiplicities of twisted states, but also to identify the twisted gauge groups, the representation content of the twisted matter, as well as values of $`g`$, $`\eta `$ and $`\rho `$ which satisfy the restrictions in Table 2 (which ensures cancellation of the local $`\mathrm{tr}R^4`$ anomaly). Furthermore, these choices must provide for complete cancellation of all other terms in the full polynomial $`I_8`$. Satisfying all of these requirements is a highly restrictive demand.
We first look for a “basic” solution where $`𝒢=E_8`$. In this case, we find a unique solution to all of our constraints. This solution requires $`g=3/4`$ and $`\stackrel{~}{𝒢}_7=SU(2)`$, which is broken as $`SU(2)U(1)`$ on the $`\alpha \beta `$-plane. In this case, the three adjoint $`SU(2)`$ fields provide one six-dimensional vector multiplet and two hypermultiplets on the $`\alpha \beta `$ plane. There are no further twisted fields. Thus, the gauge structure on the $`\alpha \beta `$-plane is $`E_8\times U(1)`$, under which the twisted fields transform as follows
| Vectors: | $`\frac{1}{2}\mathbf{\hspace{0.17em}1}_{(\mathrm{𝟎})}`$ | $`n_V=1/2`$ |
| --- | --- | --- |
| Hypers: | $`\frac{1}{2}\mathbf{\hspace{0.17em}1}_{(+\mathrm{𝟏})}\frac{1}{2}\mathbf{\hspace{0.17em}1}_{(\mathrm{𝟏})}`$ | $`n_H=1/2+1/2=1`$ |
| Tensors: | None | $`n_T=0`$ , |
where the $`U(1)`$ charges are indicated in the parenthetical subscripts. This solution also requires $`(\eta ,\rho )=(1,0)`$ <sup>12</sup><sup>12</sup>12The requirement that $`\rho =0`$ and the need for $`U(1)`$ gauge factors in the “basic” solution to $`S^1/𝐙_2\times T^4/𝐙_2`$ orbifolds was also discussed in .. In this case, the anomaly vanishes completely, as it should since there are no tensors to provide a local Green-Schwarz mechanism. The factor of one-half accompanying the twisted field representions indicate that these describe seven-dimensional fields living on the $`\beta `$-plane, contributing via a chiral projection onto the embedded $`\alpha \beta `$-plane. We represent this local solution with the diagram shown in Figure 3.
In Figure 3, the twisted gauge group $`SU(2)`$ is indicated next to the vertical line, signifying that this corresponds to $`\stackrel{~}{𝒢}_7`$. The magnetic charge associated with the intersection is indicated by the $`3/4`$. There is no local $`U(1)`$ anomaly because the charges cancel, leaving a net $`U(1)`$ charge of zero.
Our basic solution has a two notable aspects. First, using the multiplicities listed above, we compute $`n_Hn_V=1/2`$, which is precisely the value specified in Table 2 for the case of unbroken $`E_8`$ and for the choice $`g=3/4`$. The second notable aspect concerns the magnetic charge. For reasons described in , we attribute a topological significance to this number. Specifically, $`g`$ corresponds to an $`E_8`$ instanton number (associated with an instanton residing on the $`\alpha \beta `$-plane) minus the local contribution due to the nontrivial Euler character of the $`K3`$ manifold. (We attribute the second of these to a local gravitational instanton, associated with a pointlike version of the ALE space needed to blow up the orbifold.) Since the Euler number of $`K3`$ is 24, we divide this evenly over the 32 $`\alpha \beta `$-planes, so that the local gravitational contribution to $`g`$ should be exactly $`3/4`$. Since we are considering the case of unbroken $`E_8`$, we assume there are no local gauge instantons. Therefore, the only contribution to $`g`$ should be the gravitational result of $`3/4`$. It is gratifying that this number is required by our independent anomaly cancellation requirements.
When we impose $`n_T`$=0 and $`I_8=0`$ on the cases where $`\beta `$ breaks $`E_8`$ to $`𝒢=E_7\times SU(2)`$ or $`𝒢=SO(16)`$, we also find unique solutions with specialized values of $`g`$ and with specific twisted matter content. These solutions are described by the diagrams in the left-hand column of Figure 4. Note that, in each case, we require seven-dimensional $`SU(2)`$ vector multiplets. In the $`𝒢=E_7\times SU(2)`$ case, the seven-dimensional $`SU(2)`$ factor is identified with the $`SU(2)`$ factor in $`𝒢`$. This identification is indicated by the asterix on the two $`SU(2)`$ factors in the relevant diagram. As described previously, under these circumstances, $`\alpha `$ projects out the “vector” component of the seven-dimensional matter but preserves the “hyper”component. This is represented by the $`H`$ next to the arrow in the same diagram. In contrast to our “basic” solution, both of these new solutions involving local breakdown of $`E_8`$ require $`(\eta ,\rho )=(1,1)`$. The nonvanishing of $`\rho `$ indicates that the $`\beta `$-planes support non-zero $`SU(2)`$ electric charge in these cases. The physics of this observation may prove interesting.
We interpret the $`\beta `$-induced $`E_8`$ symmetry breakdown as a reflection of instantons residing on the $`\alpha \beta `$-plane. Since there are two classes of gauge instantons which could reside on the $`\alpha \beta `$-planes, one contributing integer magnetic charge and the other contributing half-integer magnetic charge, we infer that solutions could exist with $`g`$ taking values at half-integer increments greater than the “basic” value of $`3/4`$. This is precisely what we find; the $`n_T`$=0 solutions for $`𝒢=E_7\times SU(2)`$ and $`𝒢=SO(16)`$ require $`g=1/4`$ and $`g=+1/4`$, corresponding to half-integer and integer valued instantons, respectively.
In the next phase of our systematic search for local anomaly-free vertices, we study the cases with $`n_T`$=1. In these cases, we impose that the anomaly factorizes as a complete square, $`I_8(Z_4)^2`$, so that it can be cancelled by dynamics involving the self-dual tensor in the local twisted spectrum. Equation (11), which enforces the vanishing of the $`\mathrm{tr}R^4`$ term in $`I_8`$, is still important since $`\mathrm{tr}R^4`$ cannot factorize. Once again, we consider each of the three $`E_8`$ breaking patterns discussed above, using the relevant values of $`V_B`$ and $`H_B`$ in each case. For these three possibilities, when $`n_T`$=1, equation (11) reduces to the constraints indicated in Table 3. Thus, if there is one local twisted tensor multiplets then local anomaly factorization implies the indicated correlations between the local magnetic charge and the multiplicities of twisted hyper and vector multiplets. Note that, in each case, extra twisted matter is still required since the indicated multiplicites can not be made to vanish with any properly quantized choice of $`g`$. The challenge, in the case $`n_T`$=1, is not only to identify multiplicities of twisted states, but also the twisted gauge groups, the represention content of the twisted fields, as well as values of $`g`$, $`\eta `$ and $`\rho `$ which can satisfy the restrictions in Table 3 (which ensures cancellation of the local $`\mathrm{tr}R^4`$ anomaly). Furthermore, they must provide the appropriate factorization of $`I_8`$. Satisfying these requirements is, again, a highly restrictive demand.
As before, we start with the case where $`E_8`$ remains unbroken. We again find a unique solution to our constraints, but this time with $`g=+1/4`$. Once again $`\stackrel{~}{𝒢}_7=SU(2)`$, and the seven dimensional gauge group is broken by $`\beta `$ on the $`\alpha \beta `$-plane as $`SU(2)U(1)`$. The only change to the twisted spectrum in the analogous $`n_T=0`$ solution is the addition of one singlet hypermultiplet to the local six-dimensional spectrum. Thus, the twisted fields transform under $`E_8\times U(1)`$ as follows
| Vectors: | $`\frac{1}{2}\mathbf{\hspace{0.17em}1}_{(\mathrm{𝟎})}`$ | $`n_V=1/2`$ |
| --- | --- | --- |
| Hypers: | $`\frac{1}{2}\mathbf{\hspace{0.17em}1}_{(+\mathrm{𝟏})}\frac{1}{2}\mathbf{\hspace{0.17em}1}_{(\mathrm{𝟏})}\mathrm{𝟏}_{(0)}`$ | $`n_H=1/2+1/2+1=2`$ |
| Tensors: | $`\mathrm{𝟏}_{(0)}`$ | $`n_T=0`$ , |
where the $`U(1)`$ charges are again indicated in the parenthetical subscripts. This solution also requires $`(\eta ,\rho )=(1,0)`$ The factors of one-half indicate that these fields describe a projection of a seven-dimensional multiplet, living on the $`\beta `$-fixed plane, via a chiral projection onto its boundary. This solution is shown in the upper right-hand diagram in Figure 4. In our diagrams, we indicate the presence of a twisted tensor multiplet by an . In this case, the anomaly does not vanish, but is given by
$`I_8={\displaystyle \frac{1}{(2\pi )^34}}{\displaystyle \frac{3}{16}}\left(\mathrm{tr}R^22\mathrm{tr}F^2\right)^2.`$ (12)
Since this is proportional to a perfect square, it can be removed by a local Green-Schwarz mechanism mediated by the anti-self-dual tensor in the twisted tensor multiplet.
When we impose $`n_T`$=1 and $`I_8(Z_4)^2`$ on the cases where $`\beta `$ breaks $`E_8`$ to $`𝒢=E_7\times SU(2)`$ or $`𝒢=SO(16)`$, we also find unique solutions with specialized values of $`g`$ and specific twisted matter content. The set of $`n_T`$=1 solutions corresponding to each of the our three choices for $`𝒢`$ are described by the diagrams in the right-hand column of Figure 4. In each of these cases, the presence of a twisted tensor multiplet is indicated by the on the relevant vertex. In these cases, the anomaly $`I_8`$ does not vanish but, rather, is given by a complete square. Therefore, the twisted tensor involves interesting dynamics. Notably, the cases where $`E_8`$ is broken require $`\rho =1`$. This is in contrast to the situation involving unbroken $`E_8`$, where $`\rho =0`$.
It is useful to compare the top right vertex-diagram in Figure 4 with the lower left vertex-diagram in that same figure. Since these two vertices have identical magnetic charge, we infer a transition whereby the fivebrane connects smoothly to an instanton, locally breaking $`E_8`$ to $`SO(16)`$. A subtle point concerns the electric charge of the associated seven-dimensional $`\beta `$-plane (the vertical line in these diagrams). In the unbroken ($`E_8`$) phase we require $`\rho =0`$, so the seven-plane is not electrically charged. However, in the broken ($`E_7\times SU(2)`$ or $`SO(16)`$) phases we require $`\rho =1`$. This process is shown by the two-diagram sequence depicted in Figure 5.
## 4 The Global Structure
Now that we have tabulated some consistent local solutions, as classified in Figure 4, we can attempt to assemble these into a coherent global orbifold. There are extra constraints on this procedure, however, which need to be taken into account. The first of these is implied by the exactness of $`dG`$ and follows from integrating the Bianchi identity (2) over the five compact dimensions. Since this region has no boundary, the left-hand side of the integrated version of (2) vanishes due to Stokes theorem since the integrand is a total derivative. This implies that the net magnetic charge of the entire orbifold is zero. Without loss of generality, we can concentrate all of the magnetic sources either on the $`\alpha \beta `$-planes or on fivebrane worldvolumes. We therefore determine that
$`N_5+{\displaystyle \underset{i=1}{\overset{32}{}}}g_i=0,`$ (13)
where $`N_5`$ is the number of fivebranes not residing on $`\alpha \beta `$-planes. Note that $`N_5`$ is necessarily a positive integer.
There is a unique “basic” global configuration satisfying this contraint for which neither of the $`E_8`$ factors is broken at any intersection. In this case, each $`\alpha \beta `$-plane carries a magnetic charge of $`3/4`$. Since there are thirty-two $`\alpha \beta `$-planes, the proper magnetic balance is minimally achieved by including 24 fivebranes distributed randomly in the bulk of the orbifold. This situation is depicted by the first diagram in Figure 6.
The diagrams in Figure 6 depict a portion of the orbifold in which only four of the thirty-two $`\alpha \beta `$-planes are shown. (The entire orbifold would be represented as in Figure 1.) One should think of these ladder-diagrams as assemblies of the individual vertex-diagrams represented in Figure 4. Thus, vertical lines represent seven-dimensional $`\beta `$-planes and horizontal lines represent ten-dimensional $`\alpha `$-planes. It is further understood that each of the two ten-planes supports local $`E_8`$ matter. The explicit factors of $`E_8`$ shown in the first diagram of Figure 6 indicate that the ten-dimensional $`E_8`$ matter is completely unbroken on each of the four $`\alpha \beta `$ intersections shown in that diagram. Similarly, any explicit group shown at a vertex in a ladder-diagram indicates the subgroup $`𝒢E_8`$ which remains unbroken by $`\beta `$ at that indicated vertex. The ’s indicate fivebranes. Each has a worldvolume which fills the six noncompact dimensions extending out of the plane of the diagram and carries unit magnetic charge.
The fivebranes are free to move about the diagrams. Notably, due to the consistent vertex indicated at the top of the right-hand column in Figure 4, the fivebranes are free to move to, and wrap, any of the vertices in the first diagram of Figure 6. Such a procedure is shown in the second diagram in Figure 6, where the arrow indicates that the fivebrane has moved to, and wrapped, the indicated vertex. Note that, in this process, the local magnetic charge is increased by one unit, from $`3/4`$ to $`+1/4`$. In addition, the twisted spectrum at that vertex is augmented by one tensor and one hypermultiplet. All of this is consistent with the absorption of a fivebrane. Transitions of this type can occur, without further constraints, at any vertex of a global orbifold configuration.
We would now like to consider the case where a fivebrane moves to a vertex and metamorphizes into a gauge instanton via a small instanton phase transition. As discussed above, this results in the breaking of $`E_8`$ to one of its subgroups. Previously in this paper, such transitions have been analyzed at a single vertex only. However, within the context of a global orbifold configuration, it is necessary to insure that such a phase transition is compatible with the structure of the surrounding vertices. This puts additional, rather strong, constraints on the allowed phase transitions. The pertinent issue involves the electric charge $`\rho `$ of the $`\beta `$-planes. Since the unbroken ($`E_8`$) phases correspond to $`\rho =0`$ and the broken phases ($`E_7\times SU(2)`$ or $`SO(16)`$) correspond to $`\rho =1`$, and since $`\rho `$ is associated with an entire seven-dimensional $`\beta `$-plane (i.e. an entire vertical line in one of our ladder-diagrams), it would seem that instanton transitions on vertices should only occur in pairs. According to this interpretation, the lone $`n_T`$=1 vertex shown in the second diagram in Figure 6 can smoothly connect to an instanton only if another fivebrane first moves to the complementary six-plane on the top of the ladder-diagram, as depicted in the third diagram in Figure 6.
This enables each of the fivebranes to then smoothly connect to instantons, through processes of the sort shown in Figure 6, simultaneously turning on an electric charge $`\rho =1`$ on the interpolating seven-plane. Thus, the sequence depicted in Figure 5 describes a fivebrane-mediated transition which smoothly connects an $`E_8\times E_8`$ phase of the moduli space to an $`SO(16)\times SO(16)`$ phase.
It is puzzling that the value of the electric parameter $`\rho `$, which is ostensibly derived from a seven-dimensional Chern-Simons coupling, changes from zero to one when the described transitions take place. This is puzzling because if $`\rho `$ is a mere coupling constant, its value should not be subject to intermittent change. (On the other hand, we are able to justify a related change in the magnetic charge $`g`$, since this has an understandable topological origin, which allows us to resolve such a change in the manner described previously.) We suppose that this issue has an interesting resolution. This curiosity was independently noticed and commented on in . For the time being we allow a situation-dependent $`\rho `$ as an allowed rule. We hope to discuss this issue further in a future paper.
It is less clear how the half-integer instantons can emerge via smooth transitions involving fivebranes. One picture which suggests itself, however, is the following. We can imagine a fivebrane moving to one of the ten-planes, which smoothly connects in moduli space to a small ten-dimensional $`E_8`$ gauge instanton. Such a small instanton could then grow until it encompasses each of two fixed six-planes within the ten plane. We could imagine that this instanton then splits into two half-integer instantons, one localized on each of the two adjacent $`\alpha \beta `$-planes. This process is shown by a sequence of diagrams in Figure 7. We abbreviate this transition by the two-step sequence depicted in Figure 8. We can describe this process heuristically by saying that half of a fivebrane has wrapped each of the two involved $`\alpha \beta `$-planes. If such a thing were possible, then we could describe another phase transition as indicated by the sequence of diagrams in Figure 9. This depicts a fivebrane-mediated transition which smoothly connects an $`E_8\times E_8`$ phase of moduli space to an $`SO(16)\times E_7\times SU(2)`$ phase.
## 5 Conclusions
We have described some of the technology needed to resolve microscopic consistency issues in M-theory orbifolds. We have applied these techniques particularly to the $`S^1/𝐙_2\times T^4/𝐙_2`$ orbifold, and have presented an interesting set of consistent vertices, describing the twisted states residing on orbifold planes at intersecting points. By assembling consistent vertices we are able to build up consistent global orbifolds which describe different phases of moduli space. Our construction rather nicely indicates the possibility of phase transitions involving M-fivebranes as mediators. There remain intriguing conceptual issues which we intend to discuss and hope to resolve in forthcoming papers.
By resolving a larger set of consistent local vertices, thereby enlarging the number of diagrams in Figure 4, we should be able to significantly enrich our understanding of the possible allowed global configurations, and of the implied interconnectedness amongst phases in the moduli spaces associated with various orbifold compactifications indicated by the transitions which our diagrams suggest. It should prove interesting to apply our techniques to other orbifolds as well.
Acknowledgements
This work is supported by the European Commission TMR programme ERBFMRX-CT96-0045, and the work of B.A.O. is supported by the Alexander von Humboldt-foundation. We would like to acknowledge helpful emails from Luis Alvarez-Gaume, Paul Aspinwall, Bernard de Wit and Bert Schellekens, and helpful comments by Vadim Kaplunovsky. In addition, we are grateful to Stefan Theisen for pointing out an error in the previous version of this paper pertaining to the pseudoreality of a hypermultiplet representation.
|
warning/0005/cond-mat0005379.html
|
ar5iv
|
text
|
# Abstract
## Abstract
We introduce a nonextensive entropic measure $`S_\chi `$ that grows like $`N^\chi `$, where $`N`$ is the size of the system under consideration. This kind of nonextensivity arises in a natural way in some $`N`$-body systems endowed with long-range interactions described by $`r^\alpha `$ interparticle potentials. The power law (weakly nonextensive) behavior exhibited by $`S_\chi `$ is intermediate between (1) the linear (extensive) regime characterizing the standard Boltzmann-Gibbs entropy and the (2) the exponential law (strongly nonextensive) behavior associated with the Tsallis generalized $`q`$-entropies. The functional $`S_\chi `$ is parametrized by the real number $`\chi [1,2]`$ in such a way that the standard logarithmic entropy is recovered when $`\chi =1`$ . We study the mathematical properties of the new entropy, showing that the basic requirements for a well behaved entropy functional are verified, i.e., $`S_\chi `$ possesses the usual properties of positivity, equiprobability, concavity and irreversibility and verifies Khinchin axioms except the one related to additivity since $`S_\chi `$ is nonextensive. For $`1<\chi <2`$, the entropy $`S_\chi `$ becomes superadditive in the thermodynamic limit. The present formalism is illustrated by a numerical study of the thermodynamic scaling laws of a ferromagnetic Ising model with long-range interactions.
Keywords: entropy, nonextensivity, long-range interactions.
PACS: 05.20.-y;05.20.Gg;02.70.Lq;75.10.HK;05.70.Ce
## 1. Introduction
There is nowadays an intense research activity on the mathematical properties and physical applications of new versions of the Maximum Entropy Principle based on generalized or alternative entropic functionals . This line of inquiry has been greatly stimulated by the work of Tsallis, who showed that it is possible to build up a mathematically consistent and physically meaningful generalization of the standard Boltzmann-Gibbs-Jaynes thermostatistical formalism on the basis of a nonextensive entropic measure . The main motivation behind Tsallis proposal is that there are important physical scenarios, such as self-gravitating systems , electron-plasma two dimensional turbulence, among many others, which are characterized by a nonextensive behavior: due to the long range of the relevant interactions some of the thermodynamic variables usually regarded as additive, such as the internal energy, lose their extensive character. This suggests that a nonextensive (non-additive) entropy functional might be appropriate for their thermostatistical description. The Jaynes MaxEnt approach to Statistical Mechanics suggests in a natural way the possibility of incorporating alternative entropy functionals to the variational principle. The new entropy functional introduced by Tsallis has the form
$$S_q=\frac{1}{q1}\left(1\underset{i=1}{\overset{w}{}}p_i^q\right),$$
(1)
where $`(p_i,i=1,\mathrm{},w)`$ are the microstate probabilities describing the system under consideration and the entropic index $`q`$ is any real number. The standard Boltzmann-Gibbs entropy $`S=_{i=1}^wp_i\mathrm{ln}p_i`$ is recovered in the limit $`q1`$. The measure $`S_q`$ is nonextensive: The entropy of a composite system $`AB`$ constituted by two subsystems $`A`$ and $`B`$, independent in the sense that $`p_{ij}^{(AB)}=p_i^{(A)}p_j^{(B)}`$, verifies the Tsallis’ $`q`$-additive relation
$$S_q(AB)=S_q(A)+S_q(B)+(1q)S_q(A)S_q(B).$$
(2)
We see from the above equation that the Tsallis’ parameter $`q`$ can be regarded as a measure of the degree of nonextensivity. Many relevant mathematical properties of the standard thermostatistics are verified by the Tsallis’ generalized formalism or can be appropriately generalized . Self-gravitating systems constituted the first physical problem discussed within Tsallis’ nonextensive thermostatistics and Tsallis’ theory has recently been applied to other physical problems . One of the main consequences of the intensive effort devoted in recent years to the study of Tsallis theory is that there is now a growing consensus that there are many problems in statistical physics, biology, economics, and other areas, where a generalization of the standard approach based on Boltzmann-Gibbs-Jaynes extensive entropy might be useful. A comprehensive bibliography on the current research literature on Tsallis’ theory and the statistical physics of nonextensive systems can be found in .
Inspired on Tsallis’ pioneering proposal, various nonextensive entropic measures endowed with interesting properties have been recently advanced . Moreover, it has been proved that some physically relevant mathematical properties are shared by large families of entropic measures . The aim of the present work is to explore the possibility of developing a thermostatistical formalism based on a nonadditive entropic functional characterized by a degree of nonextensivity weaker than the one exhibited by Tsallis measure. As we shall see, Tsallis entropy varies exponentially with the size of the system under consideration. The new measure introduced here only scales as a power of the size of the system. That is to say, our proposal is tantamount of considering a nonextensive regime intermediate between (1) the standard extensive one associated with Boltzmann-Gibbs entropy, and (2) the exponentially nonextensive one described by Tsallis formalism.
There is an important physical motivation for introducing an entropy endowed with power law nonextensivity. Systems with long range interactions constitute the physical scenarios where the need of a generalization of the standard thermostatistical formalism can be more clearly appreciated. Consider a system of $`N`$ particles in a $`d`$-dimensional (one particle) configuration space. If the dependence of the interparticle potential with the interparticle distance $`r`$ is given by $`r^\alpha `$, it can be shown that the system’s energy levels scale as
$$N\stackrel{~}{N}=N\frac{N^{1\alpha /d}\alpha /d}{1\alpha /d}.$$
(3)
Hence, for large $`N`$, the internal energy scales as a power of the size of the system. That is,
$$EN^{2\alpha /d}$$
(4)
In the case of extensive systems governed by short-range interactions, the internal energy $`E`$ and the the entropy $`S`$ scale in the same way: they both grow linearly with $`N`$. On the other hand, the temperature $`T`$ is an intensive variable and does not change with $`N`$. How can these scaling laws be generalized to the non-extensive setting? A possible path towards the alluded generalization starts with the Helmholtz free energy,
$$F=ETS.$$
(5)
From the above expression it seems reasonable to expect both $`E`$ and $`TS`$ to scale in the same fashion . The only way to fulfill this expectation, if the standard extensive entropy is used, is to require that the temperature scales as
$$T\stackrel{~}{N}N^{1\alpha /d},$$
(6)
losing its intensive character.
It would be very appealing to have, within the nonextensive scenario, an entropy $`S_\chi `$ endowed with the same scaling law as the one exhibited by the energy. If, as it occurs for extensive systems, the entropy and the energy behave in the same way, the temperature would still be an intensive quantity. What we need in order to have a thermostatistical formalism with these nice properties is an entropy functional with power law nonextensivity, scaling as
$$S_\chi N^\chi N^{2\alpha /d}.$$
(7)
From the above discussion it is clear that we are going to assume that the exponent $`\chi `$ appearing in the entropic scaling law is related with $`d`$ and $`\alpha `$ by
$$\chi =\mathrm{\hspace{0.17em}2}\alpha /d,$$
(8)
so that the physically motivated range of vales for $`\chi `$ is
$$1\chi 2.$$
(9)
The purpose of this work is to study the basic properties of a possible candidate for a weakly-nonextensive thermostatistics, and to consider its application to a magnetic system with long-range interactions. The paper is organized as follows. In Section II the exponential behavior of Tsallis entropy is analyzed and a power–law weakly nonextensive entropy is introduced. The basic mathematical properties of the new measure are studied in Section III. Two state systems are considered in Section IV. In Section V the weakly nonextensive entropy is applied to a ferromagnetic Ising model with long-range interactions. Finally, some conclusions are drawn in Section VI.
## 2. Exponential Vs. Power Law Nonextensivity
### Tsallis Entropy and Exponential Nonextensivity
The nonextensive behavior of Tsallis functional, encapsulated in equation (2), is the most important single property distinguishing Tsallis measure from the standard additive logarithmic entropy. An important consequence of relation (2) that has not been fully appreciated in the literature is that Tsallis entropy may varies exponentially with the size of the system. In order to clarify the above assertion let us consider a system consisting of $`N`$ independent two-state subsystems. For simplicity we also assume that the system is described by an equiprobability distribution. That is, each of the $`w=2^N`$ possible microstate of the system has probability $`1/w`$. The associated Tsallis measure adopts then the value $`S_q^{equiprob.}=\frac{1w^{1q}}{q1}`$, so that, for $`q<1`$ and large $`N`$,
$$S_q\frac{1}{1q}\mathrm{\hspace{0.17em}\hspace{0.17em}2}^{(1q)N}.$$
(10)
If $`q>1`$ the $`q`$-entropy tends to the constant limit value $`\frac{1}{q1}`$ as $`N\mathrm{}`$. We are going to consider only the $`q<1`$ regime. In that case the $`q`$-entropy exhibits an exponential behavior as a function of the size $`N`$ of the system. In general, the entropy $`S_q(N)`$ of a composite system $`A=_{j=1}^NA^{(j)}`$ consisting on $`N`$ identical and independent subsystems $`\{A^{(j)},j=1,\mathrm{}N\}`$, is given by
$$S_q(N)=\frac{1}{q1}\left\{1[1+(1q)S_q(1)]^N\right\},$$
(11)
where $`S_q(1)`$ stands for the individual entropy of each constituent subsystem $`A^{(j)}`$. Notice that in order to obtain the above expression it is not necessary to assume that each subsystem $`A^{(j)}`$ is described by an equiprobability distribution.
### Power Law Nonextensive Entropy
We believe that it is worth exploring the possibility of a thermostatistical formalism based on a nonextensive entropy $`S_\chi `$ growing as a power $`N^\chi `$ of the size of the system, instead of growing exponentially. In accord with the physical arguments discussed in the Introduction (see equations (8-9)), we shall restrict our considerations to values of $`\chi [1,2]`$ . Let us assume that the functional $`S_\chi `$ associated with a given discrete probability distribution $`\{p_i,i=1,\mathrm{},w\}`$ is given by
$$S_\chi =\underset{i=1}{\overset{w}{}}\varphi _\chi (p_i),$$
(12)
$`\varphi _\chi (p_i)`$ being an appropriate function of the individual microstate probabilities $`p_i`$. In order to find a suitable expression for the function $`\varphi _\chi (z)`$ it is enough to consider again the equiprobability distribution associated with a collection of $`N`$ identical independent two state subsystems. In that case we have $`w=2^N`$ and
$`S_\chi `$ $`=`$ $`w\varphi _\chi (1/w)`$ (13)
$`=`$ $`\mathrm{\hspace{0.17em}2}^N\varphi _\chi (2^N)`$ (14)
$``$ $`N^\chi ,`$ (15)
which implies
$$\varphi _\chi (2^N)\mathrm{\hspace{0.17em}2}^NN^\chi .$$
(16)
The simplest choice for a function $`\varphi _\chi (z)`$ complying with the above relation is
$$\varphi _\chi (z)=z(\mathrm{ln}z)^\chi .$$
(17)
Unfortunately, this function is not adequate for our purposes. Its second derivative $`d^2\varphi _\chi /dz^2`$ lacks a definite sign within the relevant range of values of $`z`$ and $`\chi `$, leading thus to an entropy functional without a definite concavity. However, since we are only interested in the large $`N`$ asymptotic regime, any function $`\varphi _\chi (z)`$ behaving like (17) in the limit $`z0`$ would do. As we shall see, the function
$$\varphi _\chi (z)=\frac{1}{2}\left[z(\mathrm{ln}z)^\chi +z(\mathrm{ln}z)^{1/\chi }\right]$$
(18)
leads to the measure
$$S_\chi =\frac{1}{2}\underset{i=1}{\overset{w}{}}\left[p_i(\mathrm{ln}p_i)^\chi +p_i(\mathrm{ln}p_i)^{1/\chi }\right],$$
(19)
which complies with all the basic properties of a well behaved entropy. It is clear that in the case $`\chi =1`$ the standard entropy $`S_1={\displaystyle \underset{i=1}{\overset{w}{}}}p_i\mathrm{ln}p_i`$ is recovered.
The optimization of $`S_\chi `$ under the constraints imposed by normalization,
$$\underset{i=1}{\overset{w}{}}p_i=\mathrm{\hspace{0.33em}1},$$
(20)
and the mean values
$$𝒜^{(r)}\underset{i=1}{\overset{w}{}}p_i𝒜_i^{(r)}=𝒜_\chi ^{(r)}(r=1,\mathrm{},R)$$
(21)
of a given set $`\{𝒜^{(r)}\}`$ of observable leads to the variational problem
$$\delta \left\{S_\chi \underset{r=1}{\overset{R}{}}\beta _r𝒜^{(r)}\alpha \underset{i=1}{\overset{w}{}}p_i\right\}=\mathrm{\hspace{0.17em}0},$$
(22)
whose solution is of the form
$$p_i=F\left(\alpha +\underset{r=1}{\overset{R}{}}\beta _r𝒜_i^{(r)}\right)i=1,\mathrm{},w.$$
(23)
Here $`\alpha `$ and $`\{\beta _r\}`$ are appropriate Lagrange multipliers and $`F(z)`$ is the inverse function of
$$\varphi _\chi ^{}(z)=\frac{1}{2}\left\{(\mathrm{ln}z)^\chi +(\mathrm{ln}z)^{\frac{1}{\chi }}\chi (\mathrm{ln}z)^{\chi 1}\frac{1}{\chi }(\mathrm{ln}z)^{\frac{1}{\chi }1}\right\},$$
(24)
which verifies the relation
$$F[\varphi _\chi ^{}(z)]=\varphi _\chi ^{}[F(z)]=z.$$
(25)
The function $`F(z)`$ is well defined because the second derivative of $`s_\chi (z)`$ is negative,
$`{\displaystyle \frac{\mathrm{d}^2s_\chi }{\mathrm{d}z^2}}`$ $`=`$ $`{\displaystyle \frac{1}{2z}}[\chi (\chi 1)(\mathrm{ln}z)^{\chi 2}+{\displaystyle \frac{1}{\chi }}({\displaystyle \frac{1}{\chi }}1)(\mathrm{ln}z)^{\frac{1}{\chi }2}`$ (27)
$`\chi (\mathrm{ln}z)^{\chi 1}+{\displaystyle \frac{1}{\chi }}(\mathrm{ln}z)^{\frac{1}{\chi }1}]`$
$`<`$ $`0,`$ (28)
for $`0<z<1`$, and $`1<\chi <2`$. Actually, $`s_\chi ^{\prime \prime }(z)`$ has a definite (negative) sign within the larger interval $`[1/\chi _c,\chi _c]`$, where $`\chi _c2.1762`$. This can be appreciated in Fig. 1, where the real roots of $`s_\chi ^{\prime \prime }(z)=0`$ are depicted as a function of $`\chi `$. We see that there are no real roots $`\chi [1/\chi _c,\chi _c]`$.
It is important to realize that the lack of a simple analytical expression for the function $`F(z)`$ does not constitute a serious conceptual problem, nor does it pose any practical difficulty for the numerical treatment of problems involving the MaxEnt distributions (23). The form of the function $`F(z)`$ is shown in Fig. 2.
The canonical ensemble probability distribution associated with the entropy $`S_\chi `$,
$$p_i=F(\alpha +\beta ϵ_i),$$
(29)
is obtained when the $`S_\chi `$ is extremalized under the constraints of normalization and the mean value of the energy,
$$U=\underset{i}{}p_iϵ_i,$$
(30)
where $`ϵ_i`$ stands for the energy of the microstate $`i`$. The Lagrange multiplier $`\beta `$ appearing in (29) is the one associated with the energy constraint and corresponds, within the present thermostatistical formalism, to the inverse temperature. That is, taking Boltzmann constant $`k=1`$, we have $`\beta =1/T`$.
## 3. Main Properties of the New Entropy
### 3.1 Khinchin axioms
Khinchin proposed a set of four axioms, which are usually regarded as reasonable requirements for a well behaved information measure. Our entropy measure $`S_\chi `$ verifies the first three of them:
(i) $`S_\chi =S_\chi (p_1,\mathrm{},p_w)`$,
i.e., the entropy is a function of the probabilities $`p_i`$ only.
(ii) $`S_\chi (p_1,\mathrm{},p_w)S_\chi (\frac{1}{w},\mathrm{},\frac{1}{w})S_\chi ^{equipr.}(w)`$,
i.e., $`S_\chi `$ adopts its extreme at equiprobability (this property will be proved in subsection 3.2).
(iii) $`S_\chi (p_1,\mathrm{},p_w)=S_\chi (p_1,\mathrm{},p_w,0)`$,
this property, known as expansibility, is clearly verified since $`\varphi _\chi (0)=0`$.
(iv) The fourth Khinchin axiom concerns the behavior of the entropy of a composite system in connection to the entropies of the subsystems. We will comment on this axiom later.
### 3.2 General mathematical properties
Let us now consider the properties related to positivity, certainty, concavity, equiprobability, additivity and irreversibility, which are some of the most important features characterizing an information or entropic measure .
#### 3.2.1 Positivity
It is plain from equation (18) that $`\varphi _\chi (0)=\varphi _\chi (1)=0`$ and also that $`\varphi _\chi (p)0`$ for $`p[0,1]`$. This implies the positivity condition
$$S_\chi 0.$$
(31)
#### 3.2.2 Certainty
The equality symbol in Eq. (31) holds only at certainty, i.e.,
$$S_\chi (1,0,\mathrm{},0)=\mathrm{\hspace{0.33em}0}.$$
(32)
Indeed, $`S_\chi `$ vanishes if and only if we have certainty.
#### 3.2.3 Concavity
Considering $`(p_1,\mathrm{},p_w)`$ as independent variables, the second partial derivatives of $`S_\chi `$ are
$$\frac{^2S_\chi }{p_jp_k}=\frac{^2S_\chi }{^2p_j}\delta _{jk}<0\text{for}\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}0}<p_j<1.$$
(33)
Expression (33) guarantees definite concavity over probability space (see, for instance, ).
#### 3.2.4 Equiprobability
The probability distribution that extremizes $`S_\chi `$ under the normalization constraint is
$$p_i=F(\alpha )=\mathrm{\hspace{0.17em}1}/w$$
(34)
Therefore, since $`S_\chi `$ has negative concavity, it is maximal at equiprobability. A well behaved entropy should also be, at equiprobability, a monotonically increasing function of the number of states $`w`$. For large $`w`$ we have $`S_\chi (w)(\mathrm{ln}w)^\chi `$. Therefore, for large values of $`w`$, $`S_\chi ^{equipr.}(w)`$ is an increasing function of $`w`$.
#### 3.2.5 Nonextensivity
The nonextensive behavior of the entropy $`S\chi `$ is determined by the relation of the entropy of a composite system with the individual entropies of its constituent subsystems. Let us consider systems $`A`$ and $`B`$ with associated probabilities $`\{a_i,i=1,\mathrm{},w_A\}`$ and $`\{b_j,j=1,\mathrm{},w_B\}`$, respectively. If systems $`A`$ and $`B`$ are independent, i.e., the composite system $`AB`$ has associated probabilities $`\{a_ib_j;i=1,\mathrm{},w_A;j=1,\mathrm{},w_B\}`$, then the entropy $`S_\chi (AB)`$ of the composite system minus the sum of the entropies of its subsystems,
$$\mathrm{\Delta }S_\chi (A,B)S_\chi (AB)S_\chi (A)S_\chi (B)$$
(35)
is the quantity characterizing the nonextensive features of the measure $`S_\chi `$. When $`\mathrm{\Delta }S_\chi (A,B)>0`$($`\mathrm{\Delta }S_\chi (A,B)<0`$) we have superaditivity (subadditivity). From the examination of particular examples we conclude that $`\mathrm{\Delta }S_\chi (A,B)`$ does not have always the same sign. However, the region of probability space where $`\mathrm{\Delta }S_\chi (A,B)`$ is positive is much larger than the region where that quantity is negative. Furthermore, if we consider $`N`$ identical subsystems (instead of just two of them), the region of probability space corresponding to subadditive behavior tends to vanish as $`N`$ grows. consequently, the entropy $`S_\chi `$ becomes superextensive in the thermodynamic limit. Particular examples illustrating these features of the measure $`S_\chi `$ are provided in Section IV. Moreover, in Section V we are going to consider the nonextensivity of $`S_\chi `$ in connection with the thermodynamic properties of an Ising model endowed with long–range interactions.
#### 3.2.6 Irreversibility
One of the most important roles played by entropic functionals within theoretical physics is to characterize the “arrow of time”. When they verify an $`H`$-theorem, they provide a quantitative measure of macroscopic irreversibility. We will now show, for some simple systems, that the present measure $`S_\chi `$ satisfies an $`H`$-theorem, i.e., its time derivative has a definite sign.
Let us calculate the time derivative of $`S_\chi `$
$$\frac{\mathrm{d}S_\chi }{\mathrm{d}t}=\underset{i=1}{\overset{w}{}}\frac{\mathrm{d}p_i}{\mathrm{d}t}s_\chi ^{}(p_i),$$
(36)
for a system whose probabilities $`p_i`$ evolve according to the master equation
$$\frac{\mathrm{d}p_i}{\mathrm{d}t}=\underset{j=1}{\overset{w}{}}[P_{ji}p_jP_{ij}p_i],$$
(37)
where $`P_{ij}`$ is the transition probability per unit time between microscopic configurations $`i`$ and $`j`$. Assuming a system with a uniform equilibrium distribution and detailed balance, i.e., $`P_{ij}=P_{ji}`$, we obtain from (36)
$$\frac{\mathrm{d}S_\chi }{\mathrm{d}t}=\frac{1}{2}\underset{i=1}{\overset{w}{}}\underset{j=1}{\overset{w}{}}P_{ij}(p_ip_j)\left(s_\chi ^{}(p_j)s_\chi ^{}(p_i)\right).$$
(38)
Since $`s_\chi ^{\prime \prime }(p_i)<0`$, the quantities $`(p_ip_j)`$ and $`(s_\chi ^{}(p_j)s_\chi ^{}(p_i))`$ have the same sign. Then we obtain
$$\frac{\mathrm{d}S_\chi }{\mathrm{d}t}\mathrm{\hspace{0.33em}0}.$$
(39)
The equality holds for equiprobability, i.e., at equilibrium, while in any other cases the entropy $`S_\chi `$ increases with time. Therefore, $`S_\chi `$ exhibits irreversibility.
#### 3.2.7 Jaynes thermodynamic relations
It is noteworthy that, within the present MaxEnt formalism, the usual thermodynamical relations involving the entropy, the relevant mean values, and the associated Lagrange multipliers, i.e.,
$$\frac{S_\chi }{𝒜^{(r)}}=\beta _r$$
(40)
are verified. Hence, our formalism exhibits the usual thermodynamical Legendre transform structure. Actually, this property is verified by a wide family of entropy functionals. A particular important example of (40) is furnished by the canonical ensemble thermodynamic relation
$$\frac{S_\chi }{U}=\beta =\frac{1}{T}.$$
(41)
## 4. Two-state systems
In order to illustrate some of the above properties, we consider a two-state system (with associated probabilities $`\{p,1p\}`$). In this case, $`S_\chi `$ only depends on the variable $`p`$. In fact, from its definition, we have
$`S_\chi (p)={\displaystyle \frac{1}{2}}`$ $`[p(\mathrm{ln}p)^\chi +p(\mathrm{ln}p)^{\frac{1}{\chi }}`$ (43)
$`+(1p)(\mathrm{ln}(1p))^\chi +(1p)(\mathrm{ln}(1p))^{\frac{1}{\chi }}],`$
The shape of $`S_\chi (p)`$ for different values of $`\chi `$ is shown in Fig. 3, which exhibits the positivity and concavity of $`S_\chi `$. In fact, from expression (43), the first derivative of $`S_\chi (p)`$ vanishes at $`p=1/2`$ and $`\mathrm{d}^2S_\chi /\mathrm{d}p^2<0p[0,1]`$. Since the second derivative is always negative, $`S_\chi (p)`$ is maximal at equiprobability. Moreover, as shown in the general case, taking into account the concavity of $`S_\chi `$ and that $`S_\chi `$ vanishes at the certainty, then $`S_\chi `$ is positive for all $`p`$.
The non-additivity of $`S_\chi `$ is illustrated in Fig. 4 for two independent two-state systems $`A`$ (with probabilities $`p`$ and $`1p`$) and $`B`$ (with probabilities $`q`$ and $`1q`$), through the plot of the relative difference
$$(\mathrm{\Delta }S_\chi )_{rel.}=[S_\chi (AB)(S_\chi (A)+S_\chi (B))]/S_\chi (AB)$$
(44)
as a function of $`p`$ and $`q`$. For most values of $`p`$ and $`q`$ the nonextensive measure $`S_\chi `$ behaves in a superadditive fashion. Only for values lying in a small region near the edges of the $`(p,q)`$square does $`S_\chi `$ become subextensive.
The nonextensive behavior of $`S_\chi `$ in the thermodynamic limit can be illustrate by recourse to a system constituted by $`N`$ two-state subsystem. Let us first assume that each one of the subsystems is described by the same probabilities $`p`$ and $`(1p)`$. Consequently, the entropy associated with the composite system is
$$S_\chi (N)=\frac{1}{2^N}\underset{k=0}{\overset{N}{}}\left(\begin{array}{c}N\\ k\end{array}\right)p^k(1p)^{Nk}[(\mathrm{ln}(p^k(1p)^{Nk})^\chi +(\mathrm{ln}(p^k(1p)^{Nk})^{\frac{1}{\chi }}]$$
(45)
It can be shown after some algebra that, for $`0<p<1`$,
$$[\mathrm{ln}(1p)]^\chi N^\chi <S_\chi (N)<[\mathrm{ln}p]^\chi N^\chi +[\mathrm{ln}p]^{1/\chi }N^{1/\chi }.$$
(46)
Hence, for any given value of $`p(0,1/2)`$ there exist an $`M`$ such that
$$N>MS_\chi (N)>NS_\chi (1),$$
(47)
which means that $`S\chi `$ becomes superadditive for large enough values of $`N`$. This is shown in Fig. 5a , where the quantity
$$(\mathrm{\Delta }S_\chi )_{rel.}=[S_\chi (N)NS_\chi (1)]/S_\chi (N)$$
(48)
is depicted for different values of $`N`$. In a similar way, Fig. 5b shows the behavior of $`(\mathrm{\Delta }S_\chi )_{rel.}`$ as a function of $`p`$ for a composite system consisting of $`N1`$ two-state subsystems each with probabilities $`\{\frac{1}{2},\frac{1}{2}\}`$ along with one extra subsystem with probabilities $`p`$ and $`(1p)`$.
## 5. Ferromagnetic Ising Model with Long-Range Interactions
In this section we apply the canonical thermostatistics associated with the entropy $`S_\chi `$ to an $`N`$-body system described by an $`r^\alpha `$ interparticle potential. The energy of this kind of systems scales as a power of the number of particles . The consideration of systems exhibiting this power-law nonextensivity constituted our main motivation for introducing the measure $`S_\chi `$. As argued in Section I, an entropy endowed with the same non-extensive behavior as the one associated with the energy may lead to a thermostatistical description preserving the intensive character of the temperature. In order to illustrate these ideas we are going to study a long-range Ising model described by the Hamiltonian
$$=J\underset{i,j=1}{\overset{N}{}}\frac{1S_iS_j}{r_{ij}^\alpha },(S_i=\pm 1,i),$$
(49)
where $`J`$ is an appropriate coupling constant, and the sum runs over all the pairs of sites on a $`d`$-dimensional lattice with periodic boundary conditions and $`r_{ij}`$ stands for the distance between the sites $`i`$ and $`j`$. It is clear that the range of the interaction is determined by the value of the exponent $`\alpha `$. In particular, the standard (short-range) first-neighbors interaction is recovered in the limit $`\alpha \mathrm{}`$, while the mean field approximation is obtained when $`\alpha =0`$ (replacing $`J`$ by $`J/N`$). These extreme cases illustrate, respectively, the two possible thermodynamic behaviors that, according to the value of $`\alpha `$, are admitted by the system (49). On the one hand we have the extensive regime corresponding to $`\alpha >d`$. On the other hand we have the non-extensive regime associated with $`\alpha <d`$. In order to clarify this let us estimate the internal energy per particle at zero temperature. We have
$$\frac{E}{N}_1^{\mathrm{}}𝑑rr^{d1}r^\alpha .$$
(50)
It is easy to see that the above integral converges if $`\alpha >d`$ and diverges if $`0\alpha d`$. The following is a standard notation used in previous works:
$$\stackrel{~}{N}1+\frac{1}{d}_1^{N^{1/d}}𝑑rr^{d1}r^\alpha =\frac{N^{1\alpha /d}\alpha /d}{1\alpha /d},$$
(51)
Here we consider the non–extensive regime for the one–dimensional case $`d=1`$ and $`\alpha =0.8`$. We have performed numerical simulations by using a novel approach recently introduced to study systems with long range interactions governed by generalized entropies as the one considered here. The method relies upon the calculation of the number of states, $`\mathrm{\Omega }(ϵ_k)`$, with a given energy $`ϵ_k`$.
Note that, the number of possible configurations and associated probabilities $`p_i`$ is in general very large, $`W=2^N`$ for Ising models for example. However the number of permitted energies or energy levels $`K`$ is not so large, because there is a large number of states $`\mathrm{\Omega }(ϵ_k)`$ sharing the same energy $`ϵ_k`$. We can rewrite the sums in Eqs. (19,30) taking into account the $`K`$ energy levels weighted by $`\mathrm{\Omega }(ϵ_k)`$:
$`S_\chi `$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{k=1}{\overset{K}{}}}\mathrm{\Omega }(ϵ_k)p(ϵ_k)[(\mathrm{ln}(p(ϵ_k))^\chi +(\mathrm{ln}(p(ϵ_k))^{1/\chi }],`$ (52)
$`<O>`$ $`=`$ $`{\displaystyle \underset{k=1}{\overset{K}{}}}\mathrm{\Omega }(ϵ_k)p(ϵ_k)O(ϵ_k).`$ (53)
Hence, the knowledge of the number of states $`\mathrm{\Omega }(ϵ_k)`$ allows, by the use of these expressions, the calculation of the entropy and any averages of interest.
To calculate the probabilities $`p(ϵ_k)`$ we use the definition $`p(ϵ_k)=F(\alpha +\beta ϵ_k)`$, where the function $`F`$ is compute numerically as the inverse function of $`s_\chi ^{}(z)`$ (See Fig. 2).
To compute the histogram $`\mathrm{\Omega }(ϵ_k)`$, the “Histogram by Overlapping Windows (HOW)” method is used. A naive way of computing the histogram consists in generating different system configurations randomly and counting how many times a configuration with energy $`ϵ_k`$ appears. However, since the $`\mathrm{\Omega }(ϵ_k)`$ values span too many orders of magnitude it is not possible to find in this way a histogram over all the energy levels. The HOW method avoids this problem by generating system configurations only in a restricted energy interval and estimating the relative weights $`\mathrm{\Omega }(ϵ_k)/\mathrm{\Omega }(ϵ_l)`$ of these energy levels from the number of times they appear in the sample. From the overlap between energy intervals, one gets the complete $`\mathrm{\Omega }(ϵ_k)`$ function, apart from an irrelevant normalization factor. Details about the HOW method can be found in .
Finally we note a particularity of this new formalism, which is that the determination of the $`\alpha `$ constant from the normalization condition $`_i^wp_i=1`$, require to solve the following equation
$$\underset{k=1}{\overset{K}{}}\mathrm{\Omega }(ϵ_k)F(\alpha +\beta ϵ_k)=1,$$
(54)
where $`ϵ_k`$,$`\beta `$,$`\mathrm{\Omega }(ϵ_k)`$ are input data for the equation. This equation for $`\alpha `$ was solved numerically using a dicotomic searching method.
Using the described procedure we have calculated the dependence over the temperature $`T`$ of the internal energy $`E(N,T)`$, spontaneous magnetization $`M(N,T)=_{i=1}^Ns_i`$, entropy $`S(N,T)`$ and free energy $`F(N,T)=ETS`$ for the $`1`$–dimensional long–range Ising model in the non–extensive regime $`\alpha =0.8`$ and the corresponding value $`\chi =1.2`$ in the Eq. (8).
This system has been recently studied within the standard Boltzmann–Gibbs thermostatistics , as well as within Tsallis non–extensive $`q`$–formalism . In it was numerically verified that the scaling laws of the main thermodynamical quantities associated with the Gibbs canonical ensemble are
$`E(N,T)/(N\stackrel{~}{N})`$ $`=`$ $`e(T/\stackrel{~}{N}),`$ (55)
$`M(N,T)/N`$ $`=`$ $`m(T/\stackrel{~}{N}),`$ (56)
$`S(N,T)/N`$ $`=`$ $`s(T/\stackrel{~}{N}),`$ (57)
$`F(N,T)/(N\stackrel{~}{N})`$ $`=`$ $`f(T/\stackrel{~}{N}),`$ (58)
Since the energy and the entropy scale in different ways, the temperature $`T`$ has to be scaled as $`T/\stackrel{~}{N}`$. A similar situation arises within Tsallis $`q`$-generalized formalism, the concomitant scaling laws being
$`E_q(N,T)/(N\stackrel{~}{N})`$ $`=`$ $`e(TA_q^E(N)/(N\stackrel{~}{N})),`$ (59)
$`M_q(N,T)/N`$ $`=`$ $`m(TA_q^E(N)/(N\stackrel{~}{N})),`$ (60)
$`S_q(N,T)/(A_q(N))`$ $`=`$ $`s(TA_q^S(N)/(N\stackrel{~}{N})),`$ (61)
$`F_q(N,T)/(N\stackrel{~}{N})`$ $`=`$ $`f(TA_q(N)/(N\stackrel{~}{N})),`$ (62)
where
$`A_q(N)`$ $`=`$ $`(2^{N(1q)}1)/(1q),`$ (63)
$`A_q^S(N)`$ $`=`$ $`(2^{N|1q|}1)/|1q|,`$ (64)
$`A_q^E(N)`$ $`=`$ $`A_q(N)^2/A_q^S(N).`$ (65)
The scaling laws corresponding to the generalized $`\chi `$-canonical ensemble associated to the weakly nonextensive entropy $`S_\chi `$ are,
$`E(N,T)/(N\stackrel{~}{N})`$ $`=`$ $`e(TR_\chi (N)),`$ (66)
$`M(N,T)/N`$ $`=`$ $`m(TR_\chi (N)),`$ (67)
$`S_\chi (N,T)/(N^\chi +N^{1/\chi })`$ $`=`$ $`s(TR_\chi (N)),`$ (68)
$`F(N,T)/(N\stackrel{~}{N})`$ $`=`$ $`f(TR_\chi (N)),`$ (69)
where
$$R_\chi (N)=\frac{N^\chi +N^{1/\chi }}{N^\chi (2\chi )N}.$$
(70)
It is plain from the above equation that
$$\underset{N\mathrm{}}{lim}R_\chi (N)=\mathrm{\hspace{0.17em}1},$$
(71)
so that for large enough values of $`N`$ the scaling laws (66) become
$`E(N,T)/(N^\chi )`$ $`=`$ $`e(T),`$ (72)
$`M(N,T)/N`$ $`=`$ $`m(T),`$ (73)
$`S(N,T)_\chi /(N^\chi )`$ $`=`$ $`s(T),`$ (74)
$`F(N,T)/(N^\chi )`$ $`=`$ $`f(T).`$ (75)
According to the above equations, the thermodynamic curves of the Ising models (49) computed with increasing $`N`$-values must collapse without the need of a temperature rescalation. Numerical evidence of this scaling behavior is provided by Figures (6,7). It must be realized, however, that the $`N`$-values used are not large enough to see a complete collapse of the curves depicted. In order to reach complete collapse we need $`R_\chi (N)1`$ (see the insert of Figure (6b)).
We plot in Fig. 6 the scaled curves for the internal energy and magnetization using the Eqs. (66). In the insert of that plot we see the asimptotical behavior for the scaling factor $`R_\chi (N)`$, and with points we indicate the sizes plotted $`N=100,400,1000,2000,3000`$. Unfortunately we would need very large sizes $`N10^8`$ to observe an actual scaling without any scale factor, that is $`R_\chi 1`$. However from the Fig.6 one can see that there is an actual tendency to the collapse. In fact we expect that the final universal curves will resemble to those corresponding to $`N=2000,3000`$ and we see in the Fig. 6 a tendency towards a complete collapse.
Also in Fig.7 we show the collapse for the entropy in the main plot with the scaled relations Eqs. (66) and in the insert without scaling. In addition we can see in this plot the actual superadditive behavior of the new entropy in an interacting system instead of the simple $`N`$ independent two states systems of the last section.
## 6. Conclusions
We have shown that the entropy functional $`S_\chi `$ verifies the main properties usually regarded as essential requirements for physically meaningful entropy functionals . The entropy $`S_\chi `$ verifies the first three Khinchin axioms. The measure $`S_\chi `$ satisfies the requirements of positivity, equiprobability, concavity and irreversibility. The associated MaxEnt scheme also complies with Jaynes thermodynamic relations. The properties verified by $`S_\chi `$ suggest that it might be a useful measure for describing nonextensive physical phenomena as well as for other practical applications.
The entropic functional $`S_\chi `$ grows as a power $`N^\chi `$ of the size $`N`$ of the system and is thus characterized by a weak nonextensive behavior (as opposed to the strong, exponential nonextensivity exhibited by Tsallis measure). An attentive reader might think that a power law nonextensive measure can be obtained in a trivial way by just defining a new ”entropy” equal to $`(S_{Boltzmann})^\chi `$. However, this simple proposal would not lead to anything new. The probability distributions maximizing $`(S_{Boltzmann})^\chi `$ are the same exponential distributions that maximize $`S_{Boltzmann}`$. This means that the standard (extensive) thermostatistics would be obtained again. On the contrary, the entropic functional $`S_\chi `$ furnishes a nontrivial new thermostatistics that might be appropriate for dealing with some nonextensive systems. As an application of the thermostatistical formalism associated with $`S_\chi `$ we considered a ferromagnetic Ising model with long-range interactions. We studied numerically the concomitant scaling laws.
Tsallis thermostatistics, which have attracted a great deal of attention recently , arise from a nonextensive entropic measure $`S_q`$ growing exponentially with the size of the system. Now, if there are in Nature extensive systems (that is, those successfully studied within standard statistical mechanics and described by Gibbs exponential distributions) as well as Tsallis’ strong (exponentially) nonextensive systems, it is not unreasonable to expect also the existence of systems exhibiting an intermediate (weak) power-law kind of nonextensivity.
Let us consider a family of dynamical systems characterized by a set of parameters $`\lambda `$. We assume that there is a certain region $`\mathrm{\Sigma }`$ in $`\lambda `$-space such that the system’s thermodynamic behavior is extensive for $`\lambda \mathrm{\Sigma }`$ and nonextensive otherwise. It would be very surprising if, as a continuous change in the parameters $`\lambda `$ is considered, the system abruptly jumps from an extensive regime to an exponentially nonextensive one. It would be physically more sensible if there is an ”onset of nonextensivity” boundary between the extensive and nonextensive regions in $`\lambda `$-space, characterized by a weak, power-law nonextensive regime. A suggestive analogy can be made here with the onset of chaos is nonlinear dynamics. Consider, for instance, the well known Logistic map. For values of the map’s parameter corresponding to chaotic behavior the concomitant Lyapunov exponent is positive and near trajectories diverge exponentially in time. However, critical values of the map parameter associated with a vanishing Lyapunov exponent still do exhibit a form of ”weak chaos”. In those critical situations, corresponding to the onset chaos, near trajectories do not diverge exponentially in time: They diverge in a power-law way .
It would be important to gain a further clarification of the physical meaning of the parameter $`\chi `$, and to understand under what circumstances might Nature choose to maximize the functional $`S_\chi `$. The only way to attain such an understanding is by a detailed study of the dynamics of particular nonextensive systems endowed with long range interactions . We hope that our present contribution may stimulate further work within this line of inquiry.
#### Acknowledgements
A.R.P. acknowledges the AECI Scientific Cooperation Program (Spain), and the CONICET (Argentine Agency) for partial financial support. R. Toral and R. Salazar acknowledges the financial support from DGES, grants PB94-1167 and PB97-0141-C02-01.
REFERENCES
|
warning/0005/astro-ph0005385.html
|
ar5iv
|
text
|
# [OII] emission, eigenvector 1 and orientation in radio-quiet quasars
## 1 Introduction
Studies of optical emission lines in quasars have revealed some striking correlations that may well be related to the fundamental properties of the accreting black hole system. Boroson & Green (1992, hereafter BG92) performed a principal component analysis (PCA) on the BQS quasar sample (Schmidt & Green 1983) and showed that the primary eigenvector (hereafter eigenvector 1 or EV1), which was responsible for $``$ 30% of the variance in the data, was anticorrelated with various measures of FeII $`\lambda 4570`$ strength (equivalent width and FeII/H$`\beta `$ ratio), correlated with \[OIII\] $`\lambda 5007`$ strength (luminosity and peak) and H$`\beta `$ FWHM, and anticorrelated with the blue asymmetry of the H$`\beta `$ line. It was later found that these optical line properties correlate with UV properties: CIII\] width, SiIII\]/CIII\] ratio, CIV and NV strength (Wills et al. 1999; Kuraszkiewicz et al. 2000) and with soft X-ray properties: luminosity and spectral index (Boroson & Green 1992; Corbin 1993; Laor et al. 1994, 1997). Recently Brandt & Boller (1998) showed that the correlations between EV1 and the X-ray properties are stronger than those with the individual line parameters, suggesting that the EV1 has a more fundamental physical meaning. A number of physical parameters have been suggested to drive EV1 including accretion rate, orientation, and black hole spin.
BG92 and Boroson (1992) argued that EV1 is not driven by an orientation effect (i.e. some anisotropic property), despite the strong dependence on H$`\beta `$ line width, as it is strongly correlated with the \[OIII\] $`\lambda 5007`$ (hereafter \[OIII\]) luminosity, which was assumed to be isotropic. However, the isotropy of the \[OIII\] emission in other AGN has since been called into question. Jackson & Browne (1990) studied a sample of powerful narrow-line radio galaxies and radio-loud quasars, which, in the context of Unified Models (e.g. Antonucci 1993), are considered to be the same type of object viewed from different angles to the radio axis. The \[OIII\] line luminosity of the narrow-line radio galaxies (viewed edge-on) is lower by 5–10 times than that of the quasars, matched in redshift and extended radio luminosity. This result was surprising. It was expected that the \[OIII\] emission would be the same in both samples, as it was thought to originate from distances large enough to be unaffected by obscuring material from the dusty torus. Hes, Barthel & Fosbury (1996) found that radio-loud quasars and powerful narrow-line radio galaxies show no difference in \[OII\] $`\lambda 3727`$ (hereafter \[OII\]) emission, suggesting that \[OII\] emission, and not \[OIII\] emission, is isotropic. As \[OIII\] has a higher critical density and higher ionization potential, and hence lies nearer to the central engine, this difference can be explained if the \[OIII\] emission region extends to sufficiently small radii to be obscured by the dusty torus when the active nucleus is viewed “edge-on”. Support for this scenario was provided by the detection of \[OIII\] emission in polarized light in 4 out of 7 radio galaxies (one also showing \[OII\] polarization) while a sample of radio-loud quasars showed none (Di Serego Alighieri et al. 1997). Polarized \[OIII\] emission has also been observed in NGC 4258 (Wilkes et al. 1995; Barth et al. 1999) a Seyfert galaxy with an edge-on molecular disk surrounding the nucleus. Baker (1997), studying a complete sample of low frequency radio selected quasars from the Molonglo Quasar Sample, found that the \[OII\] to \[OIII\] ratio is anticorrelated with the radio-core to lobe flux density ratio $`R`$, which is generally used as an orientation indicator. This again implies that \[OIII\] is affected by dust absorption as the orientation becomes more edge-on. Similarly, the \[OIII\] luminosity versus radio luminosity correlation shows a larger scatter than the similar \[OII\] versus radio correlation (Tadhunter et al. 1998) in the 2 Jy extended radio selected sample (Wall & Peacock 1985). This additional scatter could again be explained by dust obscuration of the \[OIII\] emission although the authors prefer an interpretation in terms of the higher sensitivity of \[OIII\] to the ionization parameter (Tadhunter et al. 1998).
If the central regions of radio-loud quasars and powerful radio galaxies are basically similar to the central regions of radio-quiet quasars (with the exception of the existence of the radio jets) then by analogy we would expect the behavior of the \[OII\] and \[OIII\] lines to be similar in both classes. Indeed, Seyfert 1 galaxies, which in the Unified Model scenario correspond to the face-on Seyfert 2 galaxies, have higher \[OIII\] luminosities than Seyfert 2s with comparable radio luminosity (Lawrence 1987; but see Keel et al. 1994 who find no difference in a sample of IRAS selected Seyfert galaxies). This suggests that L(\[OIII\])/L(\[OII\]) could be an orientation indicator not only in radio-loud but also in radio-quiet quasars. Similarly FeII emission strength and the broad line widths are strongly dependent on orientation in radio-loud QSOs (e.g. Miley & Miller 1979; Wills & Browne 1986; Vestergaard, Wilkes & Barthel 2000) with stronger FeII and narrower lines in face-on sources. Again, by analogy, this suggests that the extreme EV1 objects, Narrow-Line Seyfert 1s (NLS1), which have stronger FeII emission and narrow lines, could also be face-on.
Given the strong evidence for anisotropic \[OIII\] emission in radio-loud quasars, we present an investigation of the behavior of \[OII\] emission in a radio-quiet subset of the optically-selected Palomar BQS sample to study the \[OII\] relation to EV1 in comparison with that of \[OIII\]. This allows us to revisit the question of orientation as a driver of EV1 and will lead to a better understanding of the underlying physics driving the strongest set of emission line correlations found to date for quasars, and so provide information on their central regions. We also compare the \[OII\] and \[OIII\] emission in our radio-quiet, optically selected sample with radio-loud samples to investigate whether the behavior of \[OII\] and \[OIII\] emission is similar in the two classes. Finally, we address the question of whether the \[OII\]/\[OIII\] ratio is an orientation indicator in radio-quiet quasars.
## 2 Observations and Data Reduction
In order to carry out this investigation the sample needs to cover a wide range of EV1 values. Since the \[OIII\] luminosity ($`M_{[\mathrm{OIII}]}`$) is directly measurable and strongly correlates with EV1, our objects were selected to have either high or low \[OIII\] luminosity i.e. $`M_{[\mathrm{OIII}]}<28`$ or $`M_{[\mathrm{OIII}]}>25.5`$ in the BG92 sample (see Figure 1; however a few radio-quiet quasars satisfying these selection criteria were not included in our sample due to lack of observing time). Our sample consists of 11 objects with low \[OIII\] luminosity and 9 objects with high \[OIII\] luminosity. We have obtained high signal-to-noise spectra of these quasars which include both the \[OIII\] $`\lambda `$5007Å and \[OII\] $`\lambda `$3727Å lines. The observations were made between 1997 June and December with the FAST spectrograph on the 1.5 m Tillinghast telescope on Mt. Hopkins in Arizona. A 300 gpm grating set to cover the wavelength range 3500$``$7500Å was used with a $`2^{\prime \prime }`$ aperture, yielding a resolution of $`6`$Å. Spectrophotometry was carried out, in photometric conditions, by observing each quasar twice, first through a large $`5^{\prime \prime }`$ aperture and second through the small, $`2^{\prime \prime }`$, aperture with a longer exposure time to obtain high spectral resolution and signal-to-noise. A standard star was observed through the wide aperture, at similar air mass, immediately before or after the quasar observation, to provide flux calibration. The data were reduced in the standard manner using IRAF<sup>1</sup><sup>1</sup>1IRAF (Image Reduction and Analysis Facility) is distributed by the National Optical Astronomy Observatories, which are operated by AURA, Inc., under cooperative agreement with the National Science Foundation. (see Tokarz & Roll 1997 for details). The continuum of the small aperture data was then normalized to match the shape and absolute flux level of the large aperture observation yielding a final spectrum with a (judged from our experience) photometric accuracy of $`5`$%. The observational details are given in Table 1 and the calibrated spectra are presented in Figure 2.
### 2.1 FeII subtraction
It has been shown (e.g. BG92) that FeII emission is present in the form of broad humps of blended lines at $`\lambda \lambda `$4450$``$4700 and $`\lambda \lambda `$5150$``$5350. This emission severely complicates measurements of emission line strengths and widths. Around the \[OIII\] $`\lambda \lambda `$4949,5007 lines, a strong FeII optical multiplet (42) consisting of 3 lines at $`\lambda \lambda `$4924,5018,5169 contaminates the \[OIII\] emission. This contamination is particularly strong in the weak \[OIII\] sources since, following EV1, these objects usually have strong FeII emission (see BG92). At the \[OII\] $`\lambda `$3727 wavelength a “small bump” (spanning 2000$``$4000Å, which is a blend of FeII lines with the Balmer continuum emission) resides, which has to be taken into account when making measurements of the \[OII\] line.
In our spectra we have fitted the underlying continuum with a power-law using only those regions of the spectrum uncontaminated by the FeII emission ($`\lambda \lambda `$4150$``$4270 and $`\lambda \lambda `$6160-6280). The underlying power-law continuum was then subtracted from each spectrum and the FeII emission was modeled using an optical ($`\lambda \lambda 42477000`$) FeII template, kindly provided by T. Boroson (BG92). The template FeII spectrum was broadened by convolving with Gaussian functions to multiple widths starting at 1000 km s<sup>-1</sup> and separated by steps of 250 km s<sup>-1</sup>. A two-dimensional iron emission model was constructed with line width as one dimension and rest wavelength as the other (see Vestergaard & Wilkes 2000). The iron emission in the object’s spectrum was fitted by scaling this 2-D iron model to the iron emission on either side of the H$`\beta `$ and \[OIII\] lines. As a check-up and confirmation of this primary normalization, five additional scalings from 0.6 to 1.4 in steps of 0.2 were applied to each normalized iron model and these were also compared to the object’s spectrum. These additionally scaled models were never needed, confirming that the primary normalization was satisfactory and adequate in each case. $`\chi ^2`$ statistics and residual flux measurements were used to determine the best fit iron model, but manual inspection of how each model fits the spectrum and of the residual spectra was also necessary and was of high importance due to the complexity of QSO spectra. Once the best-fitting FeII model was identified it was subtracted from the original QSO spectrum allowing for an improved underlying continuum to be determined. The iteration over the FeII model and continuum setting (Vestergaard & Wilkes 2000) was continued until little change was seen from one step to next. Usually no more than 2 iterations were needed.
The best-fitting FeII emission model was then subtracted from the QSO spectrum. The previously subtracted power-law continuum was then added back into the spectrum giving an FeII-subtracted spectrum. We present these spectra in Figure 2 along with the fitted FeII models to each spectrum and the original, uncorrected QSO spectra.
### 2.2 Line Measurements
The fluxes and equivalent widths (EW) of the \[OIII\] $`\lambda 5007`$, \[OII\] $`\lambda 3727`$, H$`\beta `$$`\lambda `$4861 and FeII $`\lambda 4570`$ lines (for comparison with the BG92 data) in the FeII subtracted spectra were measured using the splot task in IRAF. For \[OIII\] and H$`\beta `$ we took the previously fitted, underlying, power-law continuum and integrated the spectrum above this continuum and across the observed emission line (we used keystroke ‘e’ in the IRAF splot task). The flux and equivalent width of the FeII $`\lambda 4570`$ optical multiplet were measured in the $`\lambda \lambda `$4434$``$4684 range across the fitted FeII emission models. The equivalent widths and line luminosities of the emission lines are presented in Table 2. Our measurements of equivalent widths of H$`\beta `$, \[OIII\], FeII lines and FeII/H$`\beta `$ agree with BG92 to within 30%, except where noted in the table.
The flux of the \[OII\] $`\lambda 3727`$ line was measured by integrating the spectrum above a “local” continuum i.e. the “small bump” (dashed line in Figure 3; we did not subtract the “small bump” as it is not included in the template). The equivalent width was then defined as the ratio of the flux of the \[OII\] line estimated above the “small bump” to the flux over the same wavelength range in the power-law continuum (solid line in Figure 3; this took care of the contamination from the “small bump”). If instead the \[OII\] flux was divided by the local underlying continuum, as is more usual, then the equivalent width would be underestimated (by factors up to $``$ 5). A comparison of the equivalent widths obtained by the two methods (Figure 4) illustrates a significant systematic shift in equivalent widths whose magnitude varies from source to source, emphasizing the need to take into account the small bump when measuring the \[OII\] line.
Our sample consists of luminous quasars ($`M_B<23`$), which are generally not highly variable at optical wavelengths (e.g. Giveon et al. 1999). However, if any of the quasars had undergone a change of continuum from the time of BG92 observations, it would be seen in the differing EW(\[OIII\]) measurements (as the \[OIII\] emitting region lies far enough from the central engine to be unaffected by the changing continuum on time scales of a few years). This may be the case for the following objects: PG 0026+129, PG 1427+480, where our EW(\[OIII\]) measurements are lower than those of BG92 and in PG 1354+213, where our EW(\[OIII\]) is higher (see Table 2). These objects also show lower (or higher respectively) EW(H$`\beta `$) and slightly lower/higher EW(FeII) measurements (although the change is less than 30% and hence not noted by ‘\*’ in Table 2). The FeII/H$`\beta `$ ratio did not differ significantly from BG92 in these objects. For PG 1543+489 BG92 quote EW(\[OIII\])=0, while we were able to measure this line in our spectra and obtained a value of 4.8Å. A comparison of our line equivalent width measurements in PG 1612+261 and PG 2304+042 indicates that we have set the underlying continuum higher for PG 1612+261 and lower for PG 2304+042, and with a flatter slope, probably due to the wider wavelength range covered by our spectra, allowing a better continuum determination.
The line equivalent width measurements are influenced by the choice of aperture. BG92 used $`1.^{\prime \prime }5`$ aperture, while we use spectra obtained with a $`2^{\prime \prime }`$ aperture and flux calibrated using quasar spectra through a $`5^{\prime \prime }`$ aperture (which in turn reference a star through a $`5^{\prime \prime }`$ aperture). The amount of starlight for many PG quasars has been measured by McLeod & Rieke (1994a,b). We found that for most of the objects in our sample the starlight contribution is of the order of 20% of the total flux in the H (1.65$`\mu `$m) band i.e. 13% at 4000Å, assuming a starlight template from Elvis et al. (1994). However, PG 0157+001 has a 43% starlight contribution at band H (i.e. 29% at 4000Å) and is also spatially extended ($`12^{\prime \prime }\times 12^{\prime \prime }`$). Assuming a uniform distribution of starlight, and a constant AGN energy output, we can roughly estimate the starlight contribution in this (worst case scenario) object, which is 7% ($`0.43\times 5^2/12^2`$) at H band and 5% ($`0.29\times 5^2/12^2`$) at 4000Å in our spectra, and 0.7% and 0.5% respectively in BG92 spectra. Hence the level of starlight contamination in ours and the BG92’s equivalent widths is well below the typical 30% errors due to other factors such as continuum placement and line measurement.
### 2.3 Eigenvector 1
The EV1 values for our sample QSOs (kindly provided by T. Boroson) were calculated by applying the PCA analysis to the BQS QSOs sample. We quote these values (after BG92) in the last column of Table 2. EV1 was shown by BG92 to depend strongly on the peak and absolute magnitude of the \[OIII\] line ($`M_{[\mathrm{OIII}]}`$) and the FeII/H$`\beta `$ ratio. In general, our line measurements agree well (to within 30%) with those of BG92, implying that it is valid to use the EV1 values from BG92. In Figure 5 we present a comparison of the \[OIII\] luminosity measured by us and the absolute magnitude $`M_{[\mathrm{OIII}]}`$ from BG92, defined as $`M_\mathrm{V}2.5\mathrm{log}EW([OIII])`$. The general agreement is good with only one highly discrepant object: PG 1543+489 (indicated in Figure 5 by a filled circle; see Section 2.2 for further explanation of the differences between our and BG92 measurements). In four other objects (PG 0157+001, PG 0953+414, PG 1612+261 and PG 2304+042) we measured the FeII/H$`\beta `$ ratio to be significantly larger ($`>30\%`$) than in BG92 (see Table 2 and Figure 6). A comparison of our line equivalent width measurements in these objects indicate that we have set the underlying continuum lower and with a flatter slope in our objects, probably due to the fact that our spectra cover a larger wavelength range and that we iterated over the continuum setting and FeII models in the FeII subtraction process. In PG 0052+251 the FeII/H$`\beta `$ measured by us is smaller than in BG92. However in our FeII subtracted spectrum we still have some residual FeII $`\lambda `$4570 emission left (as our primary goal was to optimize the FeII fit around the \[OIII\] line and not FeII $`\lambda `$4570), so the BG92 FeII measurement, and hence the EV1, are probably more correct.
For the five, discrepant objects, discussed above, we indicate the direction in which EV1 should move in our figures based on the EV1 range of objects with similar values of FeII/H$`\beta `$, $`M_{[\mathrm{OIII}]}`$ and \[OIII\] peak measurements in BG92. The only way to improve on this would be to re-run the PCA analysis for the full BG92 sample using our new values of line measurements, which is beyond the scope of this paper.
## 3 Discussion
As outlined in the introduction the differences in \[OIII\] emission between the narrow-line radio galaxies and radio-loud quasars reported by Jackson & Browne (1990) and lack thereof in \[OII\] (Hes, Barthel & Fosbury 1996) suggest that \[OIII\] is orientation dependent, while \[OII\] is more isotropic in the radio-loud AGN. The correlation between the \[OII\]/\[OIII\] ratio and the orientation indicator $`R`$ reported by Baker (1997), furthermore suggests orientation-dependent dust obscuration of \[OIII\] emission and more isotropic \[OII\] emission. These results question the BG92 conclusion that EV1 is independent of orientation based on the assumption of \[OIII\] isotropy and allow us to readdress the question of orientation as the driver of EV1 by studying the dependence of BG92 EV1 on isotropic \[OII\] emission. To ensure a wide range in EV1 values we selected radio-quiet quasars with either high or low \[OIII\] luminosities (see Section 2). Under the assumption that our radio-quiet quasars, which are a subset of the BQS sample, have similar narrow line emitting regions to the radio-loud quasars and powerful radio-loud galaxies we presume that \[OII\] emission is independent of orientation in radio-quiet quasars. As a result finding a strong relation between EV1 and \[OII\] luminosity would imply that EV1 is independent of orientation (furthermore suggesting isotropic \[OIII\] emission in radio-quiet quasars, as \[OIII\] correlates with EV1) while the lack of such a relation would suggest that orientation is a factor.
The relations between \[OIII\] and \[OII\] luminosities (L(\[OIII\]), L(\[OII\])) and the Boroson & Green EV1 are presented in Figure 7.
We find a significant correlation between L(\[OIII\]) and EV1 consistent with the $`M_{[\mathrm{OIII}]}`$ versus EV1 correlation found by BG92. The Spearman rank test shows a 0.09% probability of this correlation occurring by chance (hereafter we will use $`P_S`$ to indicate the chance probability in the Spearman rank test<sup>2</sup><sup>2</sup>2We use the ASURV statistical package (Isobe, Feigelson & Nelson 1986), which includes allowance for the presence of upper limits in \[OII\] measurements.). We also find a significant correlation between L(\[OII\]) and EV1 with $`P_S=`$ 0.23%, which becomes stronger with $`P_S=`$ 0.08% if the values of EV1 were updated to allow for the differences between our measurements and those of BG92 (i.e. values in the range shown by the arrows in Figure 7). These results imply that EV1 is independent of orientation and suggest that an intrinsic property, such as the accretion rate onto a black hole (as suggested by BG92; Pounds, Done & Osborne 1995; Boller, Brandt & Fink 1996; Laor et al. 1997) or the black hole spin (BG92) may be driving EV1. In Figure 8 we present the spectra around the \[OII\] wavelength for the most positive and the most negative EV1 objects to show in detail the dependence of \[OII\] on EV1.
### 3.1 Radio-quiet versus Radio-loud Quasars
The presence of the correlations between L(\[OII\]), L(\[OIII\]) and EV1 found above suggest that in our radio-quiet quasars from the BQS sample the \[OIII\] emission is independent of orientation in contrast to the case of radio-loud quasars. In order to understand this apparent dichotomy we study in detail the L(\[OII\]) versus L\[OIII\] and EW(\[OIII\]) versus EW(\[OII\]) relations in our radio-quiet sample and compare it with the radio-loud samples of Baker (1997; hereafter JB97), and Tadhunter et al. (1998) where orientation combined with dust or ionization effects (respectively) were found to be present.
The \[OII\] and \[OIII\] luminosities and equivalent widths in our radio-quiet sample correlate significantly with one another ($`P_S`$ = 0.04% for L(\[OIII\]) versus L(\[OII\]) correlation and $`P_S`$ = 0.31% for EW(\[OIII\]) versus EW(\[OII\]) correlation). Additionally the range in L(\[OII\]) and L(\[OIII\]) is similar ($``$ 2 dex) and the best-fitted linear regression slope is consistent with 1 within the errors (0.84$`\pm `$0.11 for L(\[OII\]) versus L(\[OIII\]) and 1.27$`\pm `$0.37 for L(\[OIII\]) versus L(\[OII\]), see Figure 9). The range in equivalent widths is also similar (1.6 dex for EW(\[OIII\]) and 1.7 dex for EW(\[OII\])).
If our sample was affected by orientation dependent dust obscuration (where, as in JB97, substantial numbers of dust clouds lie within the torus opening angle, and their number increases towards the plane of the torus) a larger range in L(\[OIII\]) than in L(\[OII\]) would be observed, due to the obscuration of \[OIII\] emission at large inclination angles. Additionally a smaller range in EW(\[OIII\]) than EW(\[OII\]) would be expected as the result of the orientation dependent dust reddening of the continuum and the \[OIII\] emission. If, on the other hand, only ionization effects were present in our sample, we would observe a larger range in L(\[OIII\]) than in L(\[OII\]), and a larger range in EW(\[OIII)) than EW(\[OII\]), as the \[OIII\] line is much more dependent on the ionization parameter $`U`$ than \[OII\] (see for example Simpson 1998 Figure 5). Neither effect is present in our sample, suggesting that the BQS quasars (at least our radio-quiet sample) is remarkably free of orientation dependent dust effects or ionization effects in the narrow-line region (NLR). As both the \[OII\] and \[OIII\] emission are independent of orientation effects, the \[OII\]/\[OIII\] ratio in these optically selected radio-quiet quasars is not an orientation indicator, contrary to results for radio-loud AGN.
We compare the equivalent widths and luminosities of \[OIII\] and \[OII\] lines of objects in our optically selected radio-quiet sample with the JB97 low frequency radio selected quasar sample (Figure 10a,11a) and the complete sample of southern 2 Jy radio sources presented by Tadhunter et al. (1998; Figure 10b,11b). A number of JB97 quasars and almost all broad- and narrow-line radio galaxies of Tadhunter et al. are found to occupy a region of higher EW(\[OII\]) and EW(\[OIII\]) (see Figure 10a, 10b) than our radio-quiet quasars. We found that these comparison objects cover the whole range of \[OIII\] and \[OII\] luminosities, indicating that the high equivalent widths of the radio-loud objects are due either to higher \[OIII\] and \[OII\] luminosities or to lower observed continuum. In the latter case the continuum could be obscured by dust in the radio-selected AGN. This would confirm previous suggestions (based on the comparison of the BQS quasars optical slopes \[Francis et al. 1991\] with the X-ray selected RIXOS sample of Puchnarewicz et al. 1996 and a heterogeneous sample of Elvis et al. 1994) that the blue color selection of the BQS QSOs biases against dust obscured objects while radio-selection is uneffected.
Figures 10a, 11a also show that our sample extends to lower EW(\[OIII\]) and L(\[OIII\]) than JB97 while having similar cut-off minimum values of EW(\[OII\]) and L(\[OII\]). The \[OII\] was measured with respect to the underlying continuum in both samples and the lowest values are at the detection limit. It is possible that the \[OIII\] emission may be overestimated in the lowest equivalent width/luminosity JB97 objects due to the lack of FeII subtraction. For the extremely strong FeII objects in our radio-quiet sample, the equivalent width and luminosity of \[OIII\] would be overestimated by a factor of up to 10 if the FeII emission were not subtracted (e.g. for PG 1402+261 EW(\[OIII\])=36 with FeII included and EW(\[OIII\])=3 after FeII subtraction). Correction for FeII contamination in JB97 sample could potentially result in an intrinsic range of EW(\[OIII\]) and L(\[OIII\]) larger (by a factor of 10) than shown in Figures 10a,11a and comparable to the range of EW(\[OII\]) and L(\[OII\]) respectively. This would suggest (contrary to the conclusion reached by the author), that in the JB97 sample there is no dust obscuring the inner region of \[OIII\] emission. Confirmation of this suggestion would require a re-analysis of the JB97 sample. However to account for the broad line and continuum reddening observed by JB97, dust between the broad-line and narrow-line region is still needed.
### 3.2 The \[OII\]/\[OIII\] ratio as an orientation indicator
In the previous section we concluded that \[OII\]/\[OIII\] ratio is not an orientation indicator in our radio-quiet BQS sample. In this section we address the issue of the \[OII\]/\[OIII\] ratio as an orientation measure in radio-loud quasars.
The comparison of \[OII\]/\[OIII\] ratios with JB97 (Figure 12a) shows a lack of objects in our sample with the lowest values of \[OII\]/\[OIII\] ratio i.e. surprisingly the most core-dominated objects in JB97. One possibility is an overestimation of the \[OIII\] emission in JB97 data resulting from FeII contamination, as discussed above, which may be the case for four quasars with the smallest EW(\[OIII\]) (see also Baker et al. 1999 for spectra). Core-dominated radio-loud quasars have stronger FeII emission than lobe-dominated quasars (e.g. Miley & Miller 1979) so the FeII contamination would be larger in core-dominated objects leading to higher apparent \[OIII\] luminosity (as observed by Jackson & Browne 1990) and lower \[OII\]/\[OIII\] ratios in JB97. In this case, the \[OII\]/\[OIII\] versus $`R`$ relation of JB97 could be caused by the orientation dependence of FeII rather than of \[OIII\].
Another possible cause of low \[OII\]/\[OIII\] ratios in the JB97 sample (which could be the case for two quasars with extremely large EW(\[OIII\])) is a higher ionization parameter in radio-loud quasars. A comparison of our subset of the BQS sample and the JB97 low frequency radio selected quasar sample, with the complete sample of 2 Jy radio sources presented by Tadhunter et al. (1998) shows that both ours and JB97 samples lack objects with the highest \[OII\]/\[OIII\] ratios ($`\mathrm{log}L([OII])/L([OIII])>0`$ see Figure 12b). These high \[OII\]/\[OIII\] objects in Tadhunter et al. (1998) are mostly narrow-line radio galaxies (see Figure 12b) believed to be edge-on AGN. These objects are expected to have a large fraction of the \[OIII\] nuclear emission obscured by the dusty torus resulting in a higher \[OII\]/\[OIII\] ratio. However, there are also narrow-line radio galaxies (in Figure 12b) which show values of L(\[OII\])/L(\[OIII\]) $`<0`$, within the range of ours and the JB97 quasars as well as the broad-line radio galaxies from Tadhunter et al. This seems to be inconsistent with the orientation dependent \[OIII\] scenario in powerful radio-loud galaxies, and suggests that the \[OII\]/\[OIII\] ratio instead depends on the ionization parameter $`U`$, as suggested by Tadhunter et al. (1998).
Based on our comparisons, we conclude that the \[OII\]/\[OIII\] ratio is not a reliable orientation indicator either in the radio-quiet sample of the BQS quasars or in the radio-loud quasars.
## 4 Conclusions
Until recently it was generally accepted that eigenvector 1 does not depend on orientation as it is strongly correlated with \[OIII\] emission, originally thought to be an isotropic property in quasars. As recent studies of radio selected AGN samples have questioned the isotropy of \[OIII\] emission, we have investigated the relation between \[OII\] emission, which appears to be more isotropic, and eigenvector 1 and once again addressed the question of orientation as a driver of eigenvector 1.
We chose radio-quiet quasars from the optically selected Bright Quasar Survey which showed either high or low \[OIII\] luminosity, spanning a wide range of EV1 values in BG92. We subtracted FeII emission, which contaminates the \[OIII\] emission, from our spectra (using the BG92 iron template). We also demonstrated the significant effect of the presence of the small blue bump (Balmer continuum and FeII emission) on accurate measurements of the \[OII\] emission line, emphasizing the need for spectra covering a wide ($``$ 1000Å) wavelength range in order to determine the underlying continuum.
We found:
1. strong correlations between L(\[OII\]), L(\[OIII\]) and EV1 implying that EV1 does not depend on orientation, confirming earlier conclusions of BG92 and Boroson (1992), based on \[OIII\] alone. EV1 is likely driven by an intrinsic property (e.g. accretion rate or black hole spin).
2. significant EW(\[OIII\])– EW(\[OII\]) and L(\[OIII\])– L(\[OII\]) correlations
3. similar ranges in EW(\[OIII\]) and EW(\[OII\]) and in L(\[OIII\]) and L(\[OII\]) respectively.
These results lead us to conclude that the optically selected BQS sample (at least our radio-quiet sample) is free from orientation dependent dust effects and ionization dependent effects in the narrow-line region. Assuming our sample is representative of bright, optically selected radio-quiet quasars, this implies that their \[OIII\] emission is isotropic and the \[OII\]/\[OIII\] ratio is not an orientation indicator. This is in contrast with earlier results for the radio selected AGN (Baker 1997; Jackson & Browne 1990). We suggest that this discrepancy may be due to, contamination of the \[OIII\] emission by orientation dependent FeII emission in the latter samples.
Acknowledgements - We are grateful to Perry Berlind for observing the spectra of our sample quasars, Martin Elvis and Joanne Baker for helpful discussions, and Todd Boroson for providing the FeII optical template and the eigenvector 1 values. We gratefully acknowledges the support: of the Smithsonian pre-doctoral fellowship at the Harvard-Smithsonian Center for Astrophysics and grant no. 2P03D00410 of the Polish State Committee for Scientific Research (JK), NASA contract NAS8-39073(CXC) (BJW), NASA LTSA grant NAG5-8107 and the Alfred P. Sloan Foundation (WNB), and a Research Assistantship at SAO made possible through NASA grants: NAGW-4266, NAGW-3134, NAG5-4089 to BJW and the Columbus Fellowship at The Ohio State University (MV).
|
warning/0005/hep-ph0005177.html
|
ar5iv
|
text
|
# References
ADP-00-25/T408
ELECTROMAGNETIC RADIATION OF BARYONS
CONTAINING TWO HEAVY QUARKS
Wu-Sheng Dai<sup>1,2,3</sup>, Xin-Heng Guo<sup>4</sup>, Hong-Ying Jin<sup>5</sup> and Xue-Qian Li<sup>1,2</sup>
1. CCAST (World Laboratory), P.O.8730, Beijing 100080, China
2. Department of Physics, Nankai University, Tianjin 300071, China
3. Department of Applied Physics, Tianjin University, Tianjin 300072, China
4. Department of Physics and Mathematical Physics, and Special Research Center for the Subatomic Structure of Matter, University of Adelaide, SA 5005, Australia
5. Institute of High Energy Physics, Beijing 100039, China
Abstract
The two heavy quarks in a baryon which contains two heavy quarks and a light one, can constitute a scalar or axial vector diquark. We study electromagnetic radiations of such baryons, (i) $`\mathrm{\Xi }_{(bc)_1}\mathrm{\Xi }_{(bc)_0}+\gamma `$, (ii) $`\mathrm{\Xi }_{(bc)_1}^{}\mathrm{\Xi }_{(bc)_0}+\gamma `$, (iii) $`\mathrm{\Xi }_{(bc)_0}^{}(1/2,l=1)\mathrm{\Xi }_{(bc)_0}+\gamma `$, (iv) $`\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=1)\mathrm{\Xi }_{(bc)_0}+\gamma `$ and (v) $`\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=2)\mathrm{\Xi }_{(bc)_0}+\gamma `$, where $`\mathrm{\Xi }_{(bc)_{0(1)}},\mathrm{\Xi }_{(bc)_1}^{}`$ are S-wave bound states of a heavy scalar or axial vector diquark and a light quark, and $`\mathrm{\Xi }_{(bc)_0}^{}(l1)`$ are P- or D-wave bound states of a heavy scalar diquark and a light quark. Analysis indicates that these processes can be attributed into two categories and the physical mechanisms which are responsible for them are completely distinct. Measurements can provide a good judgment for the diquark structure and better understanding of the physical picture.
PACS numbers: 12.39.Hg, 11.10.St, 13.40.Hq, 13.30.-a
I. Introduction
Lack of an effective way to properly handle non-perturbative QCD effects becomes a more and more intriguing problem when one needs to extract information from data. In other words, the hadronic matrix elements cannot be reliably estimated in the present theoretical framework. Thanks to the heavy quark effective theory (HQET) , an extra symmetry $`SU(2)_sSU(2)_f`$ greatly simplifies the picture in the heavy flavor involved processes. Developments in this field enable us to more accurately evaluate hadronic transition matrix elements since the number of form factors is reduced in the heavy quark limit .
As many authors suggested, there may exist the diquark structure in baryons . If it is the real physics, or at least a good approximation, we only need to deal with two-body problems instead of three-body one. Cosequently, the number of independent form factors can be remarkably reduced. Especially, when the baryons contain two heavy quarks, it is reasonable to assume that the two heavy quarks constitute a color-anti-triplet boson-like diquark of spin 0 or 1 . Based on this picture Savage and Wise studied the spectrum of baryons with two heavy quarks and in the potential model, the spectra have been evaluated .
Althoug the diquark structure is very likely, the small color-anti-triplet system is not point-like in general. Consequently, we should replace the vertex gained from any fundamental theory such as the Standard Model by an effective vertex. A (or a few) reasonable form factor(s) will be involved in the effective vertex for compensating the non-point-like spatial dispersion of the diquark. The form factor(s) can be derived in many ways, and one of them is the Bethe-Salpeter (B-S) equation. With the effective vertex, we estimated the production and weak decay rates of such baryons in our previous work based on the superflavor symmetry .
To further investigate the diquark structure and the governing mechanisms inside the diquark, we will study electromagnetic radiations of baryons with two heavy quarks in the present work. Since such processes are cleaner, we may expect to gain more exact knowledge from the data. In fact, similar electromagnetic radiation processes for baryons containing only one heavy quark have been discussed in literature recently .
At the tree level, the $`\gamma `$emission is a pure electromagnetic process. In this work we study two cases which in fact are determined by completely different mechanisms. First, we consider (i) $`\mathrm{\Xi }_{(bc)_1}\mathrm{\Xi }_{(bc)_0}+\gamma `$ and (ii) $`\mathrm{\Xi }_{(bc)_1}^{}\mathrm{\Xi }_{(bc)_0}+\gamma `$, where $`\mathrm{\Xi }_{(bc)_1}`$ and $`\mathrm{\Xi }_{(bc)_1}^{}`$ are spin 1/2 and 3/2 baryons (respectively) which consist of a heavy axial vector diquark and a light quark in S-wave, and $`\mathrm{\Xi }_{(bc)_0}`$ is a spin-1/2 baryon which consists of a heavy scalar diquark and a light quark. Then we study (iii) $`\mathrm{\Xi }_{(bc)_0}^{}(1/2,l=1)\mathrm{\Xi }_{(bc)_0}+\gamma `$, (iv) $`\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=1)\mathrm{\Xi }_{(bc)_0}+\gamma `$ and (v) $`\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=2)\mathrm{\Xi }_{(bc)_0}+\gamma `$ where $`\mathrm{\Xi }_{(bc)_0}^{}(s,l1)`$ are spin 1/2 ($`s=1/2`$) and 3/2 ($`s=3/2`$) baryons (respectively) composed of a heavy scalar diquark and a light quark in higher angular momentum states. It is noted that we study the $`(bc)_{1(0)}`$ diquark because only $`(bc)`$ can constitute either spin 1 or 0 states with even parity (i.e., the orbital angular momentum between $`Q`$ and $`Q^{}`$ is set to be 0 in our discussion).
In the reactions (i) $`\mathrm{\Xi }_{(bc)_1}\mathrm{\Xi }_{(bc)_0}+\gamma `$ and (ii) $`\mathrm{\Xi }_{(bc)_1}^{}\mathrm{\Xi }_{(bc)_0}+\gamma `$, the axial vector $`(bc)_1`$ transits into a scalar $`(bc)_0`$ by emitting a photon; whereas in the radiations (iii) $`\mathrm{\Xi }_{(bc)_0}^{}(1/2,l=1)\mathrm{\Xi }_{(bc)_0}+\gamma `$, (iv) $`\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=1)\mathrm{\Xi }_{(bc)_0}+\gamma `$, and (v) $`\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=2)\mathrm{\Xi }_{(bc)_0}+\gamma `$, the diquark $`(bc)_0`$ remains in the spin-0 state, and the photon is radiated from the light quark hand. The later three reactions are in analog to the radiation of atom where electron transits from a higher (angular and/or radial) exited state into a lower one and emits a photon. In our case, the light quark of $`\mathrm{\Xi }_{(bc)_0}^{}(s,l1)`$ in an angular-momentum excited state transits into the ground state ($`l=0`$) $`\mathrm{\Xi }_{(bc)_0}`$ and emits a photon. Analysis indicates that the possibility of radiating a photon from the spin-0 heavy diquark is very small, exactly as in the case of atoms.
Of course, in general, there may be processes like $`\mathrm{\Xi }_{(bc)_1}^{}(3/2)\mathrm{\Xi }_{(bc)_1}(1/2)+\gamma `$. However, since the spin interaction between gluon and heavy diquarks decouples in the heavy quark limit, the mass splitting between $`\mathrm{\Xi }_{(bc)_1}^{}`$ and $`\mathrm{\Xi }_{(bc)_1}`$ is 0. Consequently, in the heavy quark limit, the radiative transition between these two states is forbidden by the null phase space. So we do not discuss such processes in this work.
In the next section, we present our formulation for the two different radiation mechanisms and in the third section, we give the numerical results. The last section is devoted to discussion and conclusion and finally in the appendix, we give all the concerned expressions which are omitted in the context.
II. Formulation
In this section, we discuss the two different mechanisms respectively.
(a) Radiation from the heavy diquark hand.
As discussed in the introduction, for the radiation processes $`\mathrm{\Xi }_{(bc)_1}\mathrm{\Xi }_{(bc)_0}+\gamma `$ and $`\mathrm{\Xi }_{(bc)_1}^{}\mathrm{\Xi }_{(bc)_0}+\gamma `$, the axial vector diquark transits into a scalar diquark by emitting a photon and the light quark remains as a spectator. In this case, all the non-perturbative effects can be attributed into a form factor at the leading order of expansion with respect to the heavy quark mass. To evaluate the transition matrix elements, we employ the superflavor symmetry , which is applicable to this situation.
At the effective vertex $`AS\gamma `$, where $`A`$ and $`S`$ denote axial vector and scalar diquarks (respectively) and $`\gamma `$ is the emitted photon, a form factor can be derived in terms of the B-S equation . The transition amplitude can be written as
$$T=ϵ_\alpha ^{}J^\alpha ,$$
(1)
where $`ϵ_\alpha ^{}`$ is the polarization vector of the axial vector diquark, and $`J^\alpha `$ is the effective current at the quark level and $`J^\alpha `$ is the corresponding transition amplitude.
For $`\mathrm{\Xi }_{(bc)_1}\mathrm{\Xi }_{(bc)_0}+\gamma `$,
$$J^\alpha =\mathrm{\Xi }_{(bc)_0}(v^{})|J^\alpha |\mathrm{\Xi }_{(bc)_1}(v)=\xi (v^{}v)ifϵ^{\alpha \delta \rho \sigma }v_\rho v_{}_{}{}^{}\sigma \overline{u}^{}(v^{})\gamma _5\gamma _\delta u(v),$$
(2)
and for $`\mathrm{\Xi }_{(bc)_1}^{}\mathrm{\Xi }_{(bc)_0}+\gamma `$,
$$J^\alpha =\mathrm{\Xi }_{(bc)_0}(v^{})|J^\alpha |\mathrm{\Xi }_{(bc)_1}^{}(v)=\xi (v^{}v)ifϵ^{\alpha \delta \rho \sigma }v_\rho v_{}_{}{}^{}\sigma \overline{u}^{}(v^{})u_\delta (v),$$
(3)
where $`\xi (vv^{})`$ is the Isgur-Wise function, $`v`$ and $`v^{}`$ are the four-velocities of the parent and daughter baryons, respectively, $`u`$ is the four-component spinor for the parent or produced baryon $`\mathrm{\Xi }_{(bc)_{1(0)}}`$, and $`u_\delta `$ is the Rarita-Schwinger spinor-vector corresponding to $`\mathrm{\Xi }_{(bc)_1}^{}`$ with spin 3/2. The form factor is evaluated in the B-S equation approach and all the details were given in our previous work .
Taking the amplitude square, we have: for $`\mathrm{\Xi }_{(bc)_1}\mathrm{\Xi }_{(bc)_0}+\gamma `$,
$$\frac{1}{2}\underset{allspins}{}|T|^2=\frac{e^2}{36}|\xi (vv^{})|^2Tr[C^{\rho \sigma }C^{\rho ^{}\sigma ^{}}ϵ_{\alpha \delta \rho \sigma }ϵ_{\alpha ^{}\delta ^{}\rho ^{}\sigma ^{}}\overline{u}^{}\gamma _5\gamma ^\delta u\overline{u}\gamma ^\delta ^{}\gamma _5u^{}]\underset{\lambda }{}ϵ_{(\lambda )}^\alpha ϵ_{(\lambda )}^\alpha ^{},$$
(4)
and for $`\mathrm{\Xi }_{(bc)_1}^{}\mathrm{\Xi }_{(bc)_0}+\gamma `$,
$$\frac{1}{4}\underset{allspins}{}|T|^2=\frac{e^2}{36}|\xi (vv^{})|^2Tr[C^{\rho \sigma }C^{\rho ^{}\sigma ^{}}ϵ_{\alpha \delta \rho \sigma }ϵ_{\alpha ^{}\delta ^{}\rho ^{}\sigma ^{}}\overline{u}^{}u^\delta \overline{u}^\delta ^{}u^{}]\underset{\lambda }{}ϵ_{(\lambda )}^\alpha ϵ_{(\lambda )}^\alpha ^{},$$
(5)
where
$$C^{\rho \sigma }=fv^\rho v^{}_{}{}^{}\sigma ,$$
(6)
and numerically
$$f1.$$
In our case, the photon emitted from the heavy diquark only carries very small momentum and energy, thus $`vv^{}`$ would be very close to unity, so
$$\xi (vv^{})1.$$
Then we can easily obtain the widths of these radiative decay processes as
$$\mathrm{\Gamma }=\frac{1}{2M}\frac{d^3p}{(2\pi )^3}\frac{1}{2E}\frac{d^3k}{(2\pi )^3}\frac{1}{2\omega }(2\pi )^4\delta ^4(Ppk)\frac{1}{2s+1}\underset{allspins}{}|T|^2,$$
(7)
where $`P,p,k`$ are the four-momenta of the initial, final baryons and emitted photon, respectively, and $`M`$ is mass of the initial baryon. Because it is a two-body final state case, the integration is very easy to be carried out.
(b) Radiation from the light quark hand.
In this case, $`\mathrm{\Xi }_{(bc)_0}^{}(s,l1)`$ are composed of a scalar diquark and a light quark in a higher angular momentum state ($`l1`$), thus the radiation is realized via a process that the light quark transits from a higher angular momentum state into the ground state ($`l=0`$) via emitting a photon. This process is in analog to the photon radiation of atoms where the electron jumps from an excited state (radial or $`l1`$) into the ground state via emitting a photon.
In these processes, the heavy diquark acts as a spectator. Since the reaction happens on the light flavor side, HQET is not applicable in this case. Instead, we use the B-S equation to calculate the transition matrix elements. For consistency, the wavefunctions of $`\mathrm{\Xi }_{(bc)_0}^{}(1/2(3/2),l=1,2)`$ and $`\mathrm{\Xi }_{(bc)_0}(1/2,l=0)`$ are also obtained in terms of the B-S equation. The wavefunctions are given in the following:
$`\kappa _P^{(1/2,0)}(p)`$ $`=`$ $`(\varphi _1^{(10)}(p)+\varphi _2^{(10)}(p)\text{/}p_t)u(P),(s={\displaystyle \frac{1}{2}},l=0)`$ (8)
$`\kappa _P^{(1/2,1)}(p)`$ $`=`$ $`(\varphi _1^{(11)}(p)+\varphi _2^{(11)}(p)\text{/}p_t)\gamma _5u(P),(s={\displaystyle \frac{1}{2}},l=1)`$ (9)
$`\kappa _P^{(3/2,1)}(p)`$ $`=`$ $`(\varphi _1^{(31)}(p)+\varphi _2^{(31)}(p)\text{/}p_t)p_{t\mu }u^\mu (P),(s=3/2,l=1)`$ (10)
$`\kappa _P^{(3/2,2)}(p)`$ $`=`$ $`[\varphi _1^{(32)}(p)+\varphi _2^{(32)}(p)\text{/}p_t]\gamma _5p_{t\mu }u^\mu (P),(s={\displaystyle \frac{3}{2}},l=2),`$ (11)
where $`u(P)`$ is the spinor for the baryon of spin-1/2 and $`u^\mu (P)`$ is the Rarita-Schwinger spinor-vector. Here we use the transverse momentum $`p_t`$ which is defined as
$$p_t^\mu =p^\mu p_lv^\mu ,$$
(12)
and $`v^\mu `$ is the four-velocity of the concerned baryon, $`p_lpv`$ is the longitudinal momentum.
The vertex $`\overline{q}q\gamma `$ is the typical QED coupling. Taking the loop integration with the obtained B-S wavefunctions we can have the transition amplitude square as the following.
For $`\mathrm{\Xi }_{(bc)_0}^{}(1/2,l=1)\mathrm{\Xi }_{(bc)_0}+\gamma `$,
$$\frac{1}{2}\underset{allspins}{}|T|^2=\frac{e^2}{2}\underset{allspins}{}|\overline{u}(v^{})G^\mu u(v)ϵ_\mu ^{(\lambda )}|^2,$$
(13)
where
$$G^\mu G_1\gamma ^\mu \gamma _5+G_2\gamma ^\mu \text{/}v_t^{}\gamma _5+G_3\text{/}v_t^{}\gamma ^\mu \gamma _5+G_4(2\gamma ^\mu \gamma _5+\text{/}v\gamma ^\mu \gamma _5)+G_5\text{/}v_t^{}\gamma ^\mu \text{/}v_t^{}\gamma _5.$$
(14)
For $`\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=1)\mathrm{\Xi }_{(bc)_0}+\gamma `$,
$$\frac{1}{4}\underset{allspins}{}|T|^2=\frac{e^2}{4}\underset{allspins}{}|\overline{u}(v^{})H_\nu ^\mu u^\nu (v)ϵ_\mu ^{(\lambda )}|^2,$$
(15)
where
$$H_\nu ^\mu H_1\gamma ^\mu v_\nu ^{}+H_3\gamma ^\mu \text{/}v_t^{}v_\nu ^{}+2H_4g_\nu ^\mu +H_5\text{/}v_t^{}\gamma ^\mu v_\nu ^{}+H_7\text{/}v_t\gamma ^\mu \text{/}v_t^{}v_\nu ^{}.$$
(16)
For $`\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=2)\mathrm{\Xi }_{(bc)_0}+\gamma `$,
$$\frac{1}{4}\underset{allspins}{}|T|^2=\frac{e^2}{4}\underset{allspins}{}|\overline{u}(v^{})F_\nu ^\mu \gamma _5u^\nu (v)ϵ_\mu ^{(\lambda )}|^2,$$
(17)
where
$$F_\nu ^\mu =F_1\gamma ^\mu v_\nu ^{}+F_3\gamma ^\mu \text{/}v_t^{}v_\nu ^{}+2F_4g_\nu ^\mu +F_5\text{/}v_t^{}\gamma ^\mu v_\nu ^{}+F_7\text{/}v_t\gamma ^\mu \text{/}v_t^{}v_\nu ^{}.$$
(18)
All the coefficients $`G_i`$, $`H_i`$ and $`F_i`$ in eqs.(14,16,18) are related to the B-S integrals and we give their explicit expressions in the appendix. The derivation in terms of the B-S equation is very tedious but standard.
The partial width is obtained in the same way as in II(a).
III. The numerical results
(a) Radiation from the heavy diquark hand.
Since there are no data for the masses of baryons containing two heavy quarks yet, we have to take the theoretically estimated values which are given in literatures. Here we use the results given by Ebert et al. as $`M_{\mathrm{\Xi }_{(bc)}^{}}=7.02`$ GeV and $`M_{\mathrm{\Xi }_{(bc)}}=6.95`$ GeV. We have
$$\mathrm{\Gamma }(\mathrm{\Xi }_{(bc)_1}\mathrm{\Xi }_{(bc)_0}+\gamma )2.75\times 10^9\mathrm{GeV},$$
$$\mathrm{\Gamma }(\mathrm{\Xi }_{(bc)_1}^{}\mathrm{\Xi }_{(bc)_0}+\gamma )7.48\times 10^9\mathrm{GeV}.$$
Namely, the widths are in order of eV’s.
(b) Radiation from the light quark hand.
For consistency, we have also obtained the binding energies of the concerned baryons in terms of the B-S equation. We have
$$E_{\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=2)}=1.39\mathrm{GeV},E_{\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=1)}=0.69\mathrm{GeV},$$
$$E_{\mathrm{\Xi }_{(bc)_0}^{}(1/2,l=1)}=0.66\mathrm{GeV},E_{\mathrm{\Xi }_{(bc)_0}(1/2,l=0)}=0.026\mathrm{GeV}.$$
In this framework, we have
$$M_{\mathrm{\Xi }_{(bc)_0}^{}(s,l)}=m_1+m_2+E_{\mathrm{\Xi }_{(bc)_0}^{}(s,l)},$$
$$M_{\mathrm{\Xi }_{(bc)_0}}=m_1+m_2+E_{\mathrm{\Xi }_{(bc)_0}},$$
where $`m_1`$ and $`m_2`$ are the masses of the light quark and the heavy scalar diquark, respectively, $`E`$ is the binding energy. To evaluate the binding energies, we take the simplest potential form which contains only the Coulomb and linear confinement pieces as the B-S kernel .
Numerically, we take
$$m_1=0.33\mathrm{GeV}(\mathrm{for}\mathrm{u}\mathrm{and}\mathrm{d}\mathrm{quark}),\mathrm{\hspace{0.33em}0.5}\mathrm{GeV}(\mathrm{for}\mathrm{s}\mathrm{quark});m_2=6.52\mathrm{GeV},$$
as inputs .
We use these values in the numerical evaluations and obtain:
$$\mathrm{\Gamma }(\mathrm{\Xi }_{(bc)_0}^{}(1/2,l=1)\mathrm{\Xi }_{(bc)_0}(1/2)+\gamma )1.5\times 10^4\mathrm{GeV},$$
$$\mathrm{\Gamma }(\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=1)\mathrm{\Xi }_{(bc)_0}(1/2)+\gamma )3.7\times 10^5\mathrm{GeV},$$
$$\mathrm{\Gamma }(\mathrm{\Xi }_{(bc)_0}^{}(3/2,l=2)\mathrm{\Xi }_{(bc)_0}(1/2)+\gamma )6.2\times 10^4\mathrm{GeV}.$$
As discussed above, these partial widths are evaluated in terms of the B-S equation. Indeed these reactions are governed by a mechanism different from that in (a), and the methods we use for evaluating the widths are distinct.
In this subsection, we obtain the masses of $`\mathrm{\Xi }_{(bc)_0}^{}(1/2(3/2),l1)`$ and $`\mathrm{\Xi }_{(bc)_0}(1/2)`$ and the transition matrix element $`\mathrm{\Xi }_{(bc)_0}(1/2)|J_\mu |\mathrm{\Xi }_{(bc)_0}^{}(1/2(3/2),l1)`$ in the same framework, i.e. the B-S equation. In fact, there is no any substantial difference from the values we take in subsection (a) for $`\mathrm{\Xi }_{(bc)_1}\mathrm{\Xi }_{(bc)_0}+\gamma `$ and $`\mathrm{\Xi }_{(bc)_1}^{}\mathrm{\Xi }_{(bc)_0}+\gamma `$.
It is noted that the mass difference between the angular-momentum excited state $`\mathrm{\Xi }_{(bc)_0}^{}(3/2,l1)`$ and the ground S-state $`\mathrm{\Xi }_{(bc)_0}`$ is about 0.6$``$1.4 GeV. It is much larger than that between $`\mathrm{\Xi }_{(bc)_1}^{}`$ and $`\mathrm{\Xi }_{(bc)_0}`$ (0.07 GeV). It is easy to understand: the former one is due to the orbital angular momentum excitation and the later one is due to an energy splitting between axial vector and scalar diquarks, which is caused by the spin-spin interaction. Therefore for $`\mathrm{\Xi }_{(bc)_0}^{}(s,l1)\mathrm{\Xi }_{(bc)_0}+\gamma `$ the threshold effects are not obvious and the widths are about 4 orders of magnitude larger than $`\mathrm{\Gamma }(\mathrm{\Xi }_{(bc)_1}(\mathrm{\Xi }_{(bc)_1}^{})\mathrm{\Xi }_{(bc)_0}(1/2)+\gamma )`$. In other words, the remarkable width difference for the two processes is due to the threshold effects while the matrix elements for both reactions are of the same order of magnitude.
IV. Conclusion and discussion
HQET is proved to be effective in many processes where heavy flavors are involved. In most cases, the light flavors in the hadrons just behave as spectators for the reactions and these degrees of freedom manifest in the hadronization processes, and therefore determine the form factors such as the Isgur-Wise function. However, in some cases, the light flavors may participate in reactions and sometimes can play a crucial role. As we know, when the quark level final state interaction is involved, the W-annihilation and especially the Pauli Interference can be very important in the inclusive B meson decays , then the contribution from the light flavor could be as important as that from the heavy one.
In this work, we choose two different kinds of processes where the heavy and light flavors are active respectively. $`\mathrm{\Xi }_{(bc)_1}`$ and $`\mathrm{\Xi }_{(bc)_1}^{}`$ consist of an axial vector diquark and a light quark. When they transit into $`\mathrm{\Xi }_{(bc)_0}`$ by radiating a photon, the axial vector diquark turns into a scalar one, and the light quark serves as a spectator in this process. On the contrary, $`\mathrm{\Xi }_{(bc)_0}^{}(1/2(3/2),l1)`$ consists of a scalar heavy diquark, and a light quark at angular-momentum excited states ($`l=1,2`$ in this work). Thus when it transits into $`\mathrm{\Xi }_{(bc)_0}`$, the heavy diquark stands as a spectator and the light quark jumps from a higher-excited state into the ground state while radiating a photon. For the former one, HQET definitely applies and by the superflavor symmetry, we can expect to obtain a more accurate result of the decay width. Once the doubly heavy baryon masses are measured, we can immediately have the final numbers with our formula for the partial width. As long as HQET works, the result should be close to data. Of course, there is also an uncertain factor, it is the form factor at the effective vertex of $`SA\gamma `$. We obtain it in terms of the B-S equation, where the potential kernel would bring up some uncertainty. However, in this case, the diquark is composed of two heavy quarks, so the non-relativistic Cornell potential works well as understood. Moreover, careful studies indicate that for so small recoil situation, $`(vv^{})1`$, the form factor $`f`$ is close to 1. Therefore, we can expect that the relative errors for the partial widths of $`\mathrm{\Xi }_{(bc)_1}\mathrm{\Xi }_{(bc)_0}+\gamma `$ and $`\mathrm{\Xi }_{(bc)_1}^{}\mathrm{\Xi }_{(bc)_0}+\gamma `$ are quite small. The widths are of order of eV’s and similar to that for atomic radiation. The smallness is easy to understand. Let us use $`\mathrm{\Xi }_{(bc)_1}^{}\mathrm{\Xi }_{(bc)_0}+\gamma `$ as an example. From eq.(5), we have
$$\frac{1}{4}\underset{allspins}{}|T|^2=\frac{4e^2}{27}f^2M_{1/2}M_{3/2}^{}(vv^{}1)(1+v^{}v)^2,$$
(19)
where $`M_{1/2}`$ and $`M_{3/2}^{}`$ are the masses of $`\mathrm{\Xi }_{(bc)}(1/2)`$ and $`\mathrm{\Xi }_{(bc)}^{}(3/2)`$ respectively. In this case, $`vv^{}1`$ is close to zero and it is nothing but the threshold effect. With this expression, we can easily obtain the partial width of this radiative decay as
$$\mathrm{\Gamma }=\frac{\alpha }{216}f^2\frac{(M_{3/2}^{}_{}{}^{}2M_{1/2}^2)^3}{M_{3/2}^{}_{}{}^{}5M_{1/2}^2}(M_{3/2}^{}+M_{1/2})^2.$$
(20)
It is noted that the width is proportional to $`(M_{3/2}^2M_{1/2})^3/M_{3/2}^5`$, hence for small difference between the masses of the parent and daughter baryons, the threshold effects are very obvious. One can expect this threshold effects to strongly suppress the width.
As a matter of fact, these radiative decay processes where the heavy axial vector diquark emits a photon and transits into a scalar one, are in analog to the radiative decay $`J/\psi \eta _c+\gamma `$ whose partial width is about 1.13 KeV . But there are several suppression factors in the doubly heavy baryon case. First in $`J/\psi `$ $`c`$ and $`\overline{c}`$ reside in a color singlet, but in the diquark, $`b`$ and $`c`$ quarks are in a color $`\overline{3}`$ state, there should be a factor 1/8 suppression for the diquark transition. From the formula (20), one has a factor $`(M_{3/2}^2M_{1/2}^2)/M_{3/2}^5`$, so totally there could be a suppression of about $`5\times 10^3`$ compared to the $`J/\psi `$ radiative decay. The net result is of eV order.
For $`\mathrm{\Xi }_{(bc)_0}^{}(1/2(3/2),l1)\mathrm{\Xi }_{(bc)_0}+\gamma `$, HQET does not apply and we need to employ the B-S equation method to evaluate the transition matrix elements. In the calculations, the B-S wavefunctions of the initial and final states are needed. Since in such radiative decays, the recoil energy-momentum of the final baryon is very small compared to the involved energy scales, we expect the theoretical predictions are quite reliable.
The numerical results show that for $`\mathrm{\Xi }_{(bc)_1}\mathrm{\Xi }_{(bc)_0}+\gamma `$ and $`\mathrm{\Xi }_{(bc)_1}^{}\mathrm{\Xi }_{(bc)_0}+\gamma `$, the partial width is in order of eV, and for $`\mathrm{\Xi }_{(bc)_0}^{}(3/2(1/2),l=1(2))\mathrm{\Xi }_{(bc)_0}(1/2)+\gamma `$, it is of 10$``$100 KeV. The difference is due to the threshold effects.
Besides the study on the reaction mechanisms, this work also concerns testifying the diquark structure in baryons. It is believed that the two heavy quarks inside a baryon can constitute a diquark of scalar or axial vector which is a relatively stable physical subject . Our calculations are based on such a physical picture and the future experiments should test it. Lack of data on baryons which consist of two heavy quarks so far makes drawing a definite conclusion difficult. But it is possible that the data can be accumulated in near future experiments. Once we have the data on the masses, we can re-evaluate the numbers of decay widths easily. Then comparing the calculated results with data, we can determine the validity of the diquark structure and the reaction mechanisms. No doubt, the experiments for the electromagnetic radiation are difficult, but as suggested , the radiative decay may be measurable soon, and the background in this case is clean. We believe that the results can enrich our knowledge on baryons, so is worth of careful investigations.
Acknowledgment:
This work is supported in part by the National Natural Science Foundation of China and the Australian Research Council.
Appendix
Here we present the explicit expressions of the form factors $`G_i`$, $`H_i`$ and $`F_i`$ in eqs.(14,16,18).
$$G_i\frac{dp_l}{2\pi }g_i;H_i\frac{dp_l}{2\pi }h_i;F_i\frac{dp_l}{2\pi }f_i.$$
(21)
$`g_1`$ $`=`$ $`{\displaystyle \frac{d^3p_t}{(2\pi )^3}ac^{}};`$
$`g_2`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{(vv^{})^21}}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}ad^{}|\stackrel{}{p}_t|\mathrm{cos}\theta };`$
$`g_3`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{(vv^{})^21}}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}bc^{}|\stackrel{}{p}_t|\mathrm{cos}\theta };`$
$`g_4`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}bd^{}|\stackrel{}{p}_t|^2(1\mathrm{cos}^2\theta )};`$
$`g_5`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{(vv^{})^21}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}bd^{}|\stackrel{}{p}_t|^2(13\mathrm{cos}^2\theta )};`$
$`h_1`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{(vv^{})^21}}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}ac^{\prime \prime }|\stackrel{}{p}_t|\mathrm{cos}\theta };`$
$`h_2`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}ad^{\prime \prime }|\stackrel{}{p}_t|^2(1\mathrm{cos}^2\theta )};`$
$`h_3`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{(vv^{})^21}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}ad^{\prime \prime }|\stackrel{}{p}_t|^2(13\mathrm{cos}^2\theta )};`$
$`h_4`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}bc^{\prime \prime }|\stackrel{}{p}_t|^2(1\mathrm{cos}^2\theta )};`$
$`h_5`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{(vv^{})^21}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}bc^{\prime \prime }|\stackrel{}{p}_t|^2(13\mathrm{cos}^2\theta )};`$
$`h_6`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}bd^{\prime \prime }(\lambda _2M_{(3/2,l=1)}+p_l)|\stackrel{}{p}_t|^2(1\mathrm{cos}^2\theta )};`$
$`h_7`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{(vv^{})^21}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}bd^{\prime \prime }(\lambda _2M_{(3/2,l=1)}+p_l)|\stackrel{}{p}_t|^2(13\mathrm{cos}^2\theta )};`$
$`f_1`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{(vv^{})^21}}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}ac^{\prime \prime }|\stackrel{}{p}_t|\mathrm{cos}\theta };`$
$`f_2`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}ad|\stackrel{}{p}_t|^2(1cos^2\theta )};`$
$`f_3`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{(vv^{})^21}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}ad|\stackrel{}{p}_t|^2(13\mathrm{cos}^2\theta )};`$
$`f_4`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}bc|\stackrel{}{p}_t|^2(1\mathrm{cos}^2\theta )};`$
$`f_5`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{(vv^{})^21}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}bc|\stackrel{}{p}_t|^2(13\mathrm{cos}^2\theta )};`$
$`f_7`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{(vv^{})^21}}{\displaystyle \frac{d^3p_t}{(2\pi )^3}bd(\lambda _2M_{(3/2,l=2)}+p_l)|\stackrel{}{p}_t|^2(13\mathrm{cos}^2\theta )},`$ (22)
where $`\theta `$ is the angle between $`p_t`$ and $`v_t`$,
$`a`$ $`=`$ $`\left[{\displaystyle \frac{2\omega _{p_t}^{}(\omega _{p_t}^{}m_1E_{(1/2,l=0)})}{E_{(1/2,l=0)}p_l^{})+iϵ}}{\displaystyle \frac{2m_1+p_l^{}}{(p_l^{}+m_1)^2\omega _{p_t}^{}_{}{}^{}2+iϵ}}\right]\stackrel{~}{\mathrm{\Phi }}_2^{(10)};`$
$`b`$ $`=`$ $`\left[{\displaystyle \frac{2\omega _{p_t}^{}(\omega _{p_t}^{}m_1E_{(1/2,l=0)})}{E_{(1/2,l=0)}p_l^{})+iϵ}}{\displaystyle \frac{1}{(p_l^{}+m_1)^2\omega _{p_t}^{}_{}{}^{}2+iϵ}}\right]\stackrel{~}{\mathrm{\Phi }}_2^{(10)};`$
$`c`$ $`=`$ $`i\left[{\displaystyle \frac{2\omega _{p_t}(\omega _{p_t}m_1E_{(3/2,l=2)})}{E_{(3/2,l=2)}p_l)+iϵ}}{\displaystyle \frac{p_l}{(p_l+m_1)^2\omega _{p_t}^2+iϵ}}\right]`$
$`\times (2m_2(p_lE_{(3/2,l=2)}))\stackrel{~}{\mathrm{\Phi }}_2^{(32)};`$
$`d`$ $`=`$ $`i\left[{\displaystyle \frac{2\omega _{p_t}(\omega _{p_t}m_1E_{(3/2,l=2)})}{(E_{(3/2,l=2)}p_l)+iϵ}}{\displaystyle \frac{1}{(p_l+m_1)^2\omega _{p_t}^2+iϵ}}\right]`$
$`\times (2m_2(p_lE_{(3/2,l=2)}))\stackrel{~}{\mathrm{\Phi }}_2^{(32)};`$
$`c^{}`$ $`=`$ $`i\left[{\displaystyle \frac{2\omega _{p_t}(\omega _{p_t}m_1E_{(1/2,l=1)})}{E_{(1/2,l=1)}p_l)+iϵ}}{\displaystyle \frac{p_l}{(p_l+m_1)^2\omega _{p_t}^2+iϵ}}\right]`$
$`\times (2m_2(p_lE_{(1/2,l=1)}))\stackrel{~}{\mathrm{\Phi }}_2^{(11)};`$
$`d^{}`$ $`=`$ $`i\left[{\displaystyle \frac{2\omega _{p_t}(\omega _{p_t}m_1E_{(1/2,l=1)})}{E_{(1/2,l=1)}p_l)+iϵ}}{\displaystyle \frac{1}{(p_l+m_1)^2\omega _{p_t}^2+iϵ}}\right]`$
$`\times (2m_2(p_lE_{(1/2,l=1)}))\stackrel{~}{\mathrm{\Phi }}_2^{(11)};`$
$`c^{\prime \prime }`$ $`=`$ $`i\left[{\displaystyle \frac{2\omega _{p_t}(\omega _{p_t}m_1E_{(3/2,l=1)})}{E_{(3/2,l=1)}p_l)+iϵ}}{\displaystyle \frac{2m+p_l}{(p_l+m_1)^2\omega _{p_t}^2+iϵ}}\right]`$
$`\times (2m_2(p_lE_{(3/2,l=1)})\stackrel{~}{\mathrm{\Phi }}_2^{(31)};`$
$`d^{\prime \prime }`$ $`=`$ $`i\left[{\displaystyle \frac{2\omega _{p_t}(\omega _{p_t}m_1E_{(3/2,l=1)})}{E_{(3/2,l=1)}p_l)+iϵ}}{\displaystyle \frac{1}{(p_l+m_1)^2\omega _{p_t}^2+iϵ}}\right]`$ (23)
$`\times (2m_2(p_lE_{(3/2,l=1)}))\stackrel{~}{\mathrm{\Phi }}_2^{(31)},`$
where $`\stackrel{~}{\mathrm{\Phi }}_i^{(s,l)}`$ are the B-S wavefunctions after integrated over $`p_l`$,
$$\stackrel{~}{\mathrm{\Phi }}_i^{(s,l)}\frac{dp_l}{2\pi }\varphi _i^{(s,l)}(p_l,p_t^2),$$
$`\omega _{p_t}=\sqrt{|p_t|^2+m_1^2}`$, and we have defined
$$\lambda _2=\frac{m_2}{m_1+m_2},$$
with $`m_1`$ being the light quark mass and $`m_2`$ the heavy diquark mass ($`m_1m_2`$). $`E_{(1/2,l)}`$ and $`E_{(3/2,l)}`$ are binding energies in the corresponding baryons.
All the functions are obtained by carrying out the B-S integrations which are very tedious, but straightforward (see ref.).
|
warning/0005/hep-th0005122.html
|
ar5iv
|
text
|
# USACH-FM-00/05 On the nature of fermion-monopole supersymmetry
## Abstract
It is shown that the generator of the nonstandard fermion-monopole supersymmetry uncovered by De Jonghe, Macfarlane, Peeters and van Holten, and the generator of its standard $`N=1/2`$ supersymmetry have to be supplemented by their product operator to be treated as independent supercharge. As a result, the fermion-monopole system possesses the nonlinear $`N=3/2`$ supersymmetry having the nature of the 3D spin-1/2 free particle’s supersymmetry generated by the supercharges represented in a scalar form. Analyzing the supercharges’ structure, we trace how under reduction of the fermion-monopole system to the spherical geometry the nonlinear $`N=3/2`$ superalgebra comprising the Hamiltonian and the total angular momentum as even generators is transformed into the standard linear $`N=1`$ superalgebra with the Hamiltonian to be the unique even generator.
1. Some time ago, De Jonghe, Macfarlane, Peeters and van Holten uncovered that in addition to the standard $`N=1/2`$ supersymmetry , the fermion-monopole system has the hidden supersymmetry of the nonstandard form. The latter is characterized by the supercharge anticommuting for the nonlinear operator different from the Hamiltonian and equal to the (shifted for the constant) square of the total angular momentum operator. The nonlinear supersymmetry of the similar form was observed also under investigation of the space-time symmetries in terms of the motion of pseudoclassical spinning point particles , and was revealed in the 3D $`P,T`$-invariant systems of relativistic fermions and Chern-Simons fields . In this letter we show that the generators of the nonstandard and standard supersymmetries have to be supplemented by their product operator to be treated as independent supercharge. As a result, the fermion-monopole system possesses the nonlinear $`N=3/2`$ supersymmetry having the nature of the 3D spin-1/2 free particle’s supersymmetry generated by the supercharges represented in a scalar form. So, the recent observation of the essentially free nature of the dynamics of the scalar charged particle in a monopole field is extended for the fermion-monopole system in the context of supersymmetry. Analyzing the supercharges’ structure, we trace how under reduction of the fermion-monopole system to the spherical geometry the nonlinear $`N=3/2`$ superalgebra comprising the Hamiltonian and the total angular momentum as even generators is transformed into the standard linear $`N=1`$ superalgebra with the Hamiltonian to be the unique even generator.
The paper is organized as follows. We start with the analysis of the supersymmetry of the 3D spin-1/2 free particle, and then investigate the fermion-monopole supersymmetry and discuss the reduction of the latter system to the spherical geometry. In conclusion we compare the structure of supersymmetries of the free fermion and fermion-monopole systems with the structure of the $`N=1`$ supersymmetric quantum mechanical systems .
2. Let us consider a 3D free spin-1/2 nonrelativistic particle given by the classical Lagrangian
$$L=\frac{1}{2}\dot{𝒓}{}_{}{}^{2}+\frac{i}{2}𝝍\dot{𝝍}.$$
(1)
The corresponding Hamiltonian is $`H=\frac{1}{2}𝒑^2`$, and nontrivial Poisson-Dirac brackets are $`\{r_i,p_j\}=\delta _{ij}`$ and
$$\{\psi _i,\psi _j\}=i\delta _{ij}.$$
(2)
The set of even integrals of motion is given by the vectors $`𝒑`$, $`𝑳=𝒓\times 𝒑`$ and $`𝑲=𝒑\times 𝑳`$, which have a nonzero projection on a unit of Grassmann algebra, and by the nilpotent spin vector $`𝑺=\frac{i}{2}𝝍\times 𝝍`$ generating rotations of odd Grassmann variables $`\psi _i`$. The vector $`𝑲`$ is the analog of the Laplace-Runge-Lentz vector of the Kepler system, which together with $`𝒑`$ and $`𝑳`$ constitute a non-normalized basis of orthogonal vectors forming a nonlinear algebra with nontrivial part given by the relations
$$\{L_i,V_j\}=ϵ_{ijk}V_k,V_i=p_i,L_i,K_i,\{K_i,p_j\}=𝒑^2\delta _{ij}𝒑_i𝒑_j,\{K_i,K_j\}=𝒑^2ϵ_{ijk}L_k.$$
(3)
The variables $`\psi _i`$ form the odd vector integral of motion, and their algebra (2) is the classical analog of Clifford algebra with three generators. Projecting the odd vector $`𝝍`$ onto the even vector integrals of motion, we get three odd scalar integrals of motion (supercharges)
$$Q_1=𝒑𝝍,Q_2^{}=𝑳𝝍,Q_3=𝑲𝝍.$$
(4)
The supercharge $`Q_1`$ is a “square root from the Hamiltonian”, $`\{Q_1,Q_1\}=2iH`$. It has zero bracket with the supercharge $`Q_3`$, $`\{Q_1,Q_3\}=0`$, but $`Q_1`$ and $`Q_3`$ have nontrivial brackets with $`Q_2^{}`$, in particular, $`\{Q_1,Q_2^{}\}=2i𝑳𝑺`$. One can find the linear combination of the odd scalar integrals $`Q_2^{}`$ and $`iϵ_{ijk}\psi _i\psi _j\psi _k`$ having zero brackets with other two supercharges,
$$Q_2=𝑳𝝍\frac{i}{3}(𝝍\times 𝝍)𝝍.$$
(5)
Finally, we get the set of three scalar supercharges, $`\{J_i,Q_a\}=0,`$ $`a=1,2,3`$, forming together with $`H`$ and $`𝑱`$ the nonlinear superalgebra
$`\{Q_1,Q_1\}=2iH,\{Q_2,Q_2\}=i𝑱^2,\{Q_3,Q_3\}=2iH𝑱^2,`$ (6)
$`\{Q_a,Q_b\}=0,ab,`$ (7)
where $`𝑱=𝑳+𝑺`$ is the total angular momentum vector. The scalar supercharges satisfy also the algebraic relations:
$$Q_1Q_2=i\stackrel{~}{Q}_3,Q_1Q_3=2iH\stackrel{~}{Q}_2,Q_2Q_3=i𝑱^2\stackrel{~}{Q}_1.$$
(8)
Here $`\stackrel{~}{Q}_a`$ means $`Q_a`$ with odd vector $`𝝍`$ changed for even $`𝑺`$. Since $`𝑱^2\stackrel{~}{Q}_1=𝑳^2\stackrel{~}{Q}_1`$, relations (8) reflect the analogous relations between the even vector integrals $`𝒑`$, $`𝑳`$ and $`𝑲`$:
$$𝒑\times 𝑳=𝑲,𝒑\times 𝑲=2H𝑳,𝑳\times 𝑲=𝑳^2𝒑.$$
(9)
Taking into account the equalities $`𝑱^2=2H\mathrm{\Delta }`$ and
$$\{\mathrm{\Delta },H\}=\{\mathrm{\Delta },Q_a\}=0,$$
(10)
where $`\mathrm{\Delta }=𝒓_{}^2+(𝑳𝑺)H^1`$, $`𝒓_{}=𝒓𝒑(𝒑𝒓)𝒑^2=𝑲𝒑^2`$, one can transform (at $`𝒑^20`$, $`𝑲^20`$) the set of scalar supercharges $`Q_a`$ into the set
$$\overline{Q}_1=Q_1,\overline{Q}_2=Q_2\mathrm{\Delta }^{1/2},\overline{Q}_3=Q_3(2H\mathrm{\Delta })^{1/2}.$$
(11)
This set of (nonlinearly) transformed scalar supercharges gives rise to the $`N=3/2`$ linear superalgebra of the standard form,
$$\{\overline{Q}_a,\overline{Q}_b\}=2iH\delta _{ab}.$$
(12)
The quantum analogs of the odd variables can be realized via the Pauli matrices (we use the system of units $`\mathrm{}=c=1`$), $`\widehat{\psi }_i=\frac{1}{\sqrt{2}}\sigma _i`$, and the quantum spin vector $`\widehat{𝐒}`$ is proportional (“parallel”) to $`\widehat{𝝍}`$: $`\widehat{𝐒}=\frac{1}{2}𝝈=\frac{1}{\sqrt{2}}\widehat{𝝍}`$. Note that classically this property is reflected in the relation $`𝝍\times 𝐒=0`$. To construct the quantum analogs of the supercharges $`Q_a`$ in the form of Hermitian operators, we choose the natural prescription $`𝐩\times 𝐋\frac{1}{2}(\widehat{𝐩}\times \widehat{𝐋}+(\widehat{𝐩}\times \widehat{𝐋})^{})`$, and get
$$\widehat{Q}_1=\widehat{𝐩}\widehat{𝝍},\widehat{Q}_2=\widehat{𝐋}\widehat{𝝍}+\frac{1}{\sqrt{2}},\widehat{Q}_3=(\widehat{𝐩}\times \widehat{𝐋})\widehat{𝝍}i\widehat{Q}_1.$$
(13)
These operators are Hermitian and satisfy the quantum relations (there is no summation in repeated indexes)
$`\widehat{Q}_a\widehat{Q}_b={\displaystyle \frac{1}{2}}\widehat{A}_a\delta _{ab}+{\displaystyle \frac{i}{\sqrt{2}}}\widehat{B}_cϵ_{abc}\widehat{Q}_c,`$ (14)
$`\widehat{A}_1=2\widehat{H},\widehat{A}_2=\widehat{𝐉}{}_{}{}^{2}+{\displaystyle \frac{1}{4}},\widehat{A}_3=\widehat{A}_1\widehat{A}_2,`$ (15)
$`\widehat{B}_1=\widehat{A}_2,\widehat{B}_2=\widehat{A}_1,\widehat{B}_3=1.`$ (16)
The symmetric part of relations (14) is the nonlinear $`N=3/2`$ superalgebra being the exact quantum analog of classical relations (6), (7), whereas, with taking into account the above mentioned relation between $`\widehat{S}_i`$ and $`\widehat{\psi }_i`$, we find that the antisymmetric part of (14) corresponds to the classical relations (8).
Due to the relation $`\widehat{Q}_3=i\sqrt{2}\widehat{Q}_1\widehat{Q}_2`$, it seems that one could interpret $`\widehat{Q}_1`$ and $`\widehat{Q}_2`$ as the “primary” supercharge operators and $`\widehat{Q}_3`$ as the “secondary” operator. But such an interpretation is not correct. Indeed, on the one hand, the relations (14) can be treated as a quantum generalization of the symmetric in indexes relations for the three Clifford algebra generators $`\sigma _i=\sqrt{2}\widehat{\psi }_i`$, $`\sigma _i\sigma _j=\delta _{ij}+iϵ_{ijk}\sigma _k`$, and we remember that the operators $`\widehat{\psi }_i`$ are the quantum analogs of the set of the classical odd integrals of motion $`\psi _i`$ forming a 3D vector. On the other hand, the antisymmetric part of (14) is a reflection of the cyclic classical relations (9). It is worth noting also that we can pass from the set of odd integrals (11) (constructed at $`𝒑^20`$, $`𝑲^20`$) to the integrals
$$\xi _a=\overline{Q}_a(2H)^1.$$
(17)
Then, due to the relations $`\{\xi _a,\xi _b\}=i\delta _{ab}`$, one can treat the transition from the integrals $`\psi _i`$ to the integrals $`\xi _a`$ as a simple canonical transformation for the odd sector of the phase space (note, however, that $`\xi _a`$ have nontrivial brackets with even variables). Formally, the same unitary in the odd sector transformations can be realized at the quantum level. So, it is natural to treat all the three supercharge operators $`\widehat{Q}_a`$ on the equal footing.
3. Let us consider the case of fermion particle with charge $`e`$ in arbitrary time-independent magnetic field $`𝑩(𝒓)=\mathbf{}\times 𝑨(𝒓)`$. At the Hamiltonian level this corresponds to the change of the canonical momentum vector $`𝒑`$ for the vector $`𝑷=𝒑e𝑨`$, $`\{P_i,P_j\}=eϵ_{ijk}B_k`$. By projecting the odd vector variable $`𝝍`$ onto $`𝑷`$, we get the scalar $`Q_1=𝑷𝝍`$. Identifying the bracket $`\frac{i}{2}\{Q_1,Q_1\}`$ as the Hamiltonian,
$$H=\frac{1}{2}𝑷^2e𝑩𝑺,$$
(18)
$`Q_1`$ is automatically the odd integral of motion (supercharge). However, now, unlike the case of a free particle, either even vectors $`𝑷`$, $`𝑳=𝒓\times 𝑷`$, $`𝑷\times 𝑳`$, $`𝑺`$ or odd vector $`𝝍`$ are not integrals of motion, whereas the odd scalar $`(𝝍\times 𝝍)𝝍`$ is conserved. Having in mind the analogy with the free particle case, let us check other odd scalars for their possible conservation. First, it is worth noting that the quantum relation $`\widehat{𝑺}=\frac{1}{\sqrt{2}}\widehat{𝝍}`$ is reflected classically also in the form of the identical evolution, $`\dot{𝝍}=e𝝍\times 𝑩`$, $`\dot{𝑺}=e𝑺\times 𝑩`$. Like the projection of the odd vector $`𝝍`$, the projection of $`𝑺`$ on $`𝑷`$ is conserved, i.e. as in a free case, $`\stackrel{~}{Q}_1=𝑷𝑺`$ is the integral of motion, but generally $`\frac{d}{dt}(𝑺𝑳)=e(𝑺\times 𝑷)(𝒓\times 𝑩)0`$. The scalar $`\stackrel{~}{Q}{}_{2}{}^{}=𝑺𝑳`$ is the integral of motion only if $`𝑩=f(r)𝒓`$, $`r=\sqrt{𝒓^2}`$. This corresponds to the case of the monopole field, for which $`f(r)=gr^3`$ ($`𝒓0`$, $`g=const`$) is fixed by the condition $`\mathbf{}𝑩=0`$, and from now on, we restrict the analysis by the fermion-monopole system. Though in this case the brackets $`\{L_i,L_j\}=ϵ_{ijk}(L_k+\alpha n_k)`$, $`\alpha =eg`$, are different from the corresponding brackets for the free particle, nevertheless the odd scalar $`Q_2^{}=𝑳𝝍`$ is the integral of motion. One can check that the direct analog of the free particle’s supercharge (5) has zero bracket with $`Q_1`$ and that
$$\{Q_2,Q_2\}=i(𝑱^2\alpha ^2).$$
(19)
Here $`𝑱=𝑳+𝑺\alpha 𝐫r^1`$ is the conserved angular momentum vector of the fermion-monopole system, whose components form $`su(2)`$ algebra, $`\{J_i,J_j\}=ϵ_{ijk}J_k`$, and generate rotations. The scalar $`Q_3=(𝑷\times 𝑳)𝝍`$ is also the integral of motion and in the fermion-monopole case the classical relations of the form (6), (7), (8) take place with the change $`𝑱^2𝑱^2\alpha ^2`$ and with the Hamiltonian given by Eq. (18). Like for the free particle, the superalgebra can be reduced to the standard linear form (12) via the nonlinear transformation (11), for which in the present case we proceed from the relation $`𝑱^2\alpha ^2=2H\mathrm{\Delta }`$ with $`\mathrm{\Delta }=\stackrel{~}{𝒓}{}_{}{}^{2}+𝑳𝑺H^1`$, $`\stackrel{~}{𝒓}{}_{}{}^{2}=𝒓^2(𝑷𝒓)^2(2H)^1`$. The quanitity $`\stackrel{~}{𝒓}^2`$ is the integral of motion which in the case of the scalar charged particle ($`\psi _i=0`$) in the field of monopole gives a minimal charge-monopole distance in the point of perihelion: $`r_{min}=\sqrt{\stackrel{~}{𝒓}^2}`$ .
Constructing the quantum analogs of the supercharges in the same way as in the free particle case, we get the supercharge operators of the form (13) with $`\widehat{𝒑}`$ changed for $`\widehat{𝑷}`$. They satisfy the set of (anti-)commutation relations of the same form (14)–(16) with $`\widehat{A}_2=\widehat{𝑱}{}_{}{}^{2}+\frac{1}{4}`$ changed for $`\widehat{𝑱}{}_{}{}^{2}\alpha ^2+\frac{1}{4}`$, where $`\alpha `$ is subject to the Dirac quantization condition, $`2\alpha \text{ZZ}`$.
The fermion-monopole integral $`\widehat{Q}_2`$ was observed for the first time on algebraic grounds by d’Hoker and Vinet in the context of generalization of the so called dynamical symmetries of the charge-monopole system for the supersymmetric case. The supercharge nature of $`\widehat{Q}_2`$ and the associated nonstandard nonlinear superalgebra of the operators $`\widehat{Q}_2`$ and $`\widehat{Q}_1`$, was uncovered by De Jonghe, Macfarlane, Peeters and van Holten . The present analysis shows that the set of supercharge operators $`\widehat{Q}_1`$ and $`\widehat{Q}_2`$ has to be extended by the scalar integral $`\widehat{Q}_3`$, and these three odd operators together with even operators $`\widehat{H}`$ and $`\widehat{𝑱}`$ form the described nonlinear $`N=3/2`$ superalgebra. As we have seen, this nonlinear supersymmetry of the fermion-monopole system has the nature of the free fermion particle’s supersymmetry generated by the supercharges represented in a scalar form.
Comparing the fermion-monopole Hamiltonian (18) (with $`𝑩=\alpha 𝒓r^3`$) with the free fermion particle Hamiltonian $`H=\frac{1}{2}𝒑^2`$, it seems that they have rather different structure, but this is not so, and their similarity can be revealed, like in the case of the scalar particle , by separating the even phase space coordinates into the radial and angular ones. The radial coordinates for the fermion-monopole system are $`r`$ and $`P_r=𝑷𝒏`$, and the angular phase space variables are $`𝒏=𝒓r^1`$, $`𝓙=𝑳\alpha 𝒏`$. These coordinates have the nontrivial brackets $`\{r,P_r\}=1`$, $`\{𝒥_i,𝒥_j\}=ϵ_{ijk}𝒥_k`$, $`\{𝒥_i,n_j\}=ϵ_{ijk}n_k`$, and satisfy the relations $`𝓙𝒏=\alpha `$, $`𝒏^2=1`$. In terms of these variables, the Hamiltonian of the fermion-monopole system is
$$H=\frac{1}{2}P_r^2+\frac{𝑱^2\alpha ^2}{2r^2}\frac{𝑳𝑺}{r^2},$$
with $`𝑱=𝓙+𝑺`$. The case of the free fermion corresponds to $`\alpha =0`$, and its Hamiltonian takes the similar form $`H=\frac{1}{2}p_r^2+\frac{1}{2r^2}𝑱^2\frac{1}{r^2}𝑳𝑺,`$ where $`𝑱=𝑳+𝑺`$. The difference between the two systems is encoded now in the topology of even angular phase space variables .
4. Let us look at the fermion-monopole supersymmetry from the point of view of the reduction of the system to the spherical geometry. To this end we first note that the bracket of the supercharge $`Q_2`$ with itself can be represented in the form
$$i\{Q_2,Q_2\}=2r^2\left(H\frac{1}{2}P_r^2+\frac{i}{r}Q_1(𝝍𝒏)\right),$$
(20)
and the supercharge $`Q_3`$ can be reduced to the equivalent form
$$Q_3=2Hr\left(𝝍𝒏Q_1\frac{P_r}{2H}\right).$$
(21)
As it was shown in ref. , the reduction of the fermion-monopole system to the spherical geometry can be realized by introducing into the system the classical relations
$`𝒓^2\rho ^2=0,P_r=0,`$ (22)
$`𝝍𝒏=0,`$ (23)
which have to be treated as the set of second class constraints with $`\rho 0`$ being a constant, and for simplicity we fix it in the form $`\rho =1`$. The relations (20) and (21) allow us to observe directly that the described $`N=3/2`$ nonlinear fermion-monopole supersymmetry is transformed into the $`N=1`$ supersymmetry of the standard linear form in the case of reduction (22), (23). Indeed, after reducing the fermion-monopole system onto the surface of even second class constraints (22), we find that the structure of the supercharge $`Q_3`$ is trivialized and takes the form of the odd scalar $`𝝍𝒏`$ multiplied by $`2H`$. Two other supercharges $`Q_1`$ and $`Q_2`$ after such a reduction take the form of linear combinations of the odd vector $`𝝍`$ projected on the vectors $`𝓙+\alpha 𝒏`$ and $`𝓙\times 𝒏`$ orthogonal to $`𝒏`$. Then taking into account the odd second class constraint (23) results in eliminating the supercharge $`Q_3`$ and in reducing the bracket (and corresponding anticommutator at the quantum level) of the supercharge $`Q_2`$ to $`2h`$, where $`h`$ is the reduced Hamiltonian,
$$h=\frac{1}{2}(𝑱^2\alpha ^2).$$
(24)
In other words, the supersymmetry of the fermion-monopole system in spherical geometry is reduced to the standard linear $`N=1`$ supersymmetry characterized by two supercharges anticommuting for the Hamiltonian . More explicitly, after reduction to the surface of the second class constraints, the radial variables $`r`$ and $`P_r`$ are eliminated from the theory. The even variables can be represented by the total angular momentum $`𝑱`$ and by the unit vector $`𝒏`$ having the nontrivial Dirac brackets coinciding with corresponding initial Poisson brackets, $`\{J_i,J_j\}^{}=ϵ_{ijk}J_k`$, $`\{J_i,n_j\}^{}=ϵ_{ijk}n_k`$. The odd variables $`\psi _i`$ satisfy the relation (23) which has to be treated as a strong equality, and their nontrivial Dirac brackets are $`\{\psi _i,\psi _j\}^{}=i(\delta _{ij}n_in_j)`$, $`\{J_i,\psi _j\}^{}=ϵ_{ijk}\psi _k`$. The even and odd variables are subject also to the relation $`𝑱𝒏=\alpha +iq_1q_2(2h)^1`$, where $`h`$ is given by Eq. (24) and
$$q_1=(𝑱\times 𝒏)𝝍,q_2=𝑱𝝍$$
(25)
are the supercharges $`Q_1`$ and $`Q_2`$ reduced to the surface (22), (23). With the listed Dirac brackets, one can easily check that now the reduced supercharges (25) satisfy the superalgebra of the standard $`N=1`$ supersymmetry: $`\{q_\mu ,q_\nu \}^{}=2i\delta _{\mu \nu }h`$, $`\{q_\mu ,h\}=0`$, $`\mu ,\nu =1,2`$.
5. To conclude, let us compare the structure of supersymmetric quantum mechanics with the structure of the free fermion and fermion-monopole systems. A supersymmetric quantum mechanical system is characterized by the Hamiltonian $`\widehat{H}=\frac{1}{2}(\widehat{p}{}_{}{}^{2}+W^2(x)+\sigma _3W^{}(x))`$ with $`\widehat{p}=id/dx`$, and by the supercharges $`\widehat{Q}_1=\widehat{\theta }_1W(x)\widehat{\theta }_2\widehat{p}`$, $`\widehat{Q}_2=\widehat{\theta }_2W(x)+\widehat{\theta }_1\widehat{p}`$ with $`\widehat{\theta }_\mu =\frac{1}{\sqrt{2}}\sigma _\mu `$, $`\mu =1,2`$, which form the $`N=1`$ superalgebra $`[\widehat{Q}_\mu ,\widehat{Q}_\nu ]__+=2\delta _{\mu \nu }\widehat{H}`$, $`[\widehat{H},\widehat{Q}_\mu ]=0`$. The operator $`\widehat{Q}_3=\frac{1}{\sqrt{2}}\sigma _3`$ is the trivial integral of motion, $`[\widehat{Q}_3,\widehat{H}]=0`$, and the set of supercharges $`\widehat{Q}_{1,2}`$ together with $`\widehat{Q}_3`$ satisfy the relations of the form (14) with $`\widehat{A}_1=\widehat{A}_2=2\widehat{B}_3=2\widehat{H}`$, $`\widehat{A}_3=\widehat{B}_1=\widehat{B}_2=1`$. In this case the operator $`\sigma _3=2i\widehat{\theta }_1\widehat{\theta }_2=\sqrt{2}\widehat{Q}_3`$ (being analogous to any component of the odd vector integral $`\sqrt{2}\widehat{𝝍}`$ of the free fermion system) plays the role of the grading operator commuting (anticommuting) with operators $`\widehat{x}`$ and $`\widehat{p}`$ ($`\widehat{\theta }_{1,2}`$), and, as a consequence, commuting (anticommuting) with the Hamiltonian $`\widehat{H}`$ (supercharges $`\widehat{Q}_{1,2}`$). On the other hand, for the fermion-monopole and the free fermion systems, one can construct the operator $`\sqrt{2}\widehat{\xi }_3`$ proceeding from the classical relation (17). It seems that such operator could be treated as the grading operator due to the relation $`2\widehat{\xi }_3^2=1`$ and its anticommutation with the supercharges $`\widehat{Q}_{1,2}`$. But such interpretation is not correct since unlike the case of supersymmetric quantum mechanics, this operator has a nontrivial dependence on even operators, and as a consequence, does not commute with them.
The ordinary form of the classical Lagrangian for the supersymmetric quantum mechanics,
$$L=\frac{1}{2}(\dot{x}^2W^2(x)2iW^{}(x)\theta _1\theta _2+i\theta _\mu \dot{\theta }_\mu ),$$
(26)
contains only two Grassmann variables. At the classical level the even quantity $`2i\theta _1\theta _2`$ corresponds to the odd operator $`\sigma _3`$ (the latter being one of three generators of the corresponding Clifford algebra), i.e. here we have some sort of classical anomaly . However, the symmetry between quantum and classical pictures can easily be restored (“the anomaly can be canceled”) extending the set $`\theta _\mu `$, $`\mu =1,2`$, by the independent Grassmann variable $`\theta _3`$ and changing Lagrangian (26) for $`=L+\frac{i}{2}\theta _3\dot{\theta }_3.`$ Such a classical system has two nontrivial supercharges $`Q_1`$, $`Q_2`$, and the third odd integral of motion given by $`\theta _3`$ (like $`\psi _i`$ for the free fermion) is trivial and completely decoupled from other variables. Therefore, the difference of the superalgebraic structures of the supersymmetric quantum mechanics on the one hand and fermion-monopole system on the other hand is also reflected in the absence in the latter case of the grading operator commuting with the Hamiltonian and anticommuting with any two of three scalar supercharges and which simultaneously would commute with the initial coordinate and momenta operators.
Acknowledgements
The work has been supported in part by the grant 1980619 from FONDECYT (Chile) and by DICYT (USACH).
|
warning/0005/cond-mat0005031.html
|
ar5iv
|
text
|
# Crossover from percolation to diffusion
## Abstract
A problem of the crossover from percolation to diffusion transport is considered. A general scaling theory is proposed. It introduces phenomenologically four critical exponents which are connected by two equations. One exponent is completely new. It describes the increase of the diffusion below percolation threshold. As an example, an exact solution of one dimensional lattice problem is given. In this case the new exponent $`q=2`$.
Percolation theory is often used to describe transport properties of disordered systems with a large disorder. The typical problems are random mixture of conducting and non-conducting elements, or hopping conduction. The regular percolation theory assumes that the random elements do not change their positions with time so that the percolation paths do not change their spatial locations. The low-temperature electron transport is one of examples where this assumption is not valid. Due to electron-electron interaction the random potential persistently and substantially changes with time, which may affect conductivity near the metal-insulator transition. Such problems appear outside solid state physics as well. The class of this problem is known as dynamical percolation. They have been studied theoretically using effective medium approximation (See e.g. Ref. and references therein). One dimensional problems of this type have been considered for some models without any approximations.
We concentrate here on the case when the diffusion through conducting media is drastically faster then the fluctuations of the positions of the conducting and non-conducting elements. In this case the transition from diffusion to percolation mechanism has all features of the phase transition and it can be characterized by critical exponents. In this paper we introduce a set of these exponents and establish relation between them. For illustration we present an exact solution of one dimensional lattice problem.
Thus, we consider here a random mixture of the conducting and non-conducting elements which can change there positions but very slowly. In this situation the resulting diffusion (or conductivity) of the particles is non-zero at any small fraction $`p`$ of the conducting elements. The mechanism of this diffusion is as follows. A particle can diffuse only in the conducting medium. To move from one conducting element to the other it is waiting until another conducting element comes to the element where it resides. At this moment of time a particle is able to do a next move. This is the simple diffusion which is characterized by some slow waiting time $`\tau _s`$. The diffusion inside conducting medium is characterized by much faster time $`\tau _f`$. When the fraction of conducting elements $`p`$ becomes close to the percolation threshold $`p_c`$, but it is less than $`p_c`$ the resulting diffusion increases as $`1/(p_cp)^q`$, where $`q`$ is a novel critical exponent. This happens because the conducting clusters become large. But since they are disconnected the particle should wait to jump from one of them to the other. Finally, when $`pp_c`$ is positive and it becomes larger than the width of some critical region, the diffusion is described by a regular percolation law $`D(pp_c)^t`$ and slow motion of the conducting elements is not important. To describe this physics one can use the same scaling arguments as for the problem of frozen mixture of elements with large and small conductivity. Note that the new problem is not equivalent to the old one, so that the exponent $`q`$ might be different. The scaling hypothesis is valid in the proximity of the percolation threshold so that $`|X|/p_c1`$, where $`X=pp_c`$. In this region the scaling hypothesis can be written in the form
$$d=h^s\psi (\frac{X}{h^m}).$$
(1)
Here $`d=D/D_f`$, $`h=D_s/D_f`$, where $`D_f=a^2/2\tau _f`$ is the diffusion coefficient in the conducting medium,$`a`$ is a characteristic length which depends on the model, $`D_w=a^2/2\tau _s`$, and $`\psi (Z)`$ is some analytical function at all real values of its argument, $`\mathrm{}<Z<\mathrm{}`$ . We assume that $`h1`$ and that $`\psi (0)=1`$.
Eq. (1) contains two independent critical exponents. The meaning of the exponent $`s`$ is that $`d=h^s`$ at $`p=p_c`$. The exponent $`m`$ describes the width of the critical region $`|pp_c|=h^m`$ between the percolation and diffusion.
All the critical exponents which describe $`d(X)`$ can be related to $`s`$ and $`m`$. At $`X>0`$ and $`Xh^m`$ the slow changes of the percolation paths are not important. Thus, we have the diffusion on the percolation cluster. If the system is larger than the correlation radius of the percolation cluster, the diffusion coefficient is related to the conductivity by the Einstein relation. Then one has $`dX^t`$, where the exponent $`t`$ is a usual percolation exponent which describes the conductivity above the percolation threshold. To get this result from Eq. (1) one should assume that $`m=s/t`$. At $`X<0`$ and $`|X|h^m`$ we expect that $`dh/(|X|)^q`$. This gives $`s+mq=1`$. Finally we get two connections between four exponents $`s,q,t,m`$, namely $`q=t(s^11)`$ and $`m=s/t`$.
As an example, we consider below an exactly soluble one dimensional site or bond model on a lattice. At the beginning we consider the site model. The sites may be white and black with the fractions $`p`$ and $`1p`$ respectively. The particle may occupy the white sites only and it is able to jump at the nearest site, if this site is white, during the time $`\tau _f`$. The configuration of the white and black sites slowly changes with time. This change can be introduced by many ways. We consider the simplest one. After each time interval $`\tau _s`$, which is called the renewal time, the configuration of all sites completely changes preserving the same $`p`$, while the particle remains at the same white site. We assume that $`\tau _s\tau _f`$. Since $`p_c=1`$ in the one-dimensional case, our solution may only illustrate the increase of the diffusion coefficient $`D(1p)^q`$ and the width of the critical region $`|X|`$, where it deviates from this law and tends to the value $`D_f=a^2/2\tau _f`$ which is reached at $`p=1`$. Since the diffusion of a particle within the time interval $`\tau _s`$ is completely independent of the diffusion during other time intervals, one has
$$\overline{r^2(t)}=\overline{(r_1(t_1)+r_2(t_2)+\mathrm{}+r_N(t_N))^2}=\overline{r_1^2(t_1)}+\overline{r_2^2(t_2)}+\mathrm{}+\overline{r_N^2(t_N)}$$
(2)
It follows that
$$D=\underset{t\mathrm{}}{lim}\frac{r^2(t)}{2t}=\frac{\overline{r^2(\tau _s)}}{2\tau _s},$$
(3)
where $`\overline{r^2(\tau _s)}`$ is averaged over all possible initial position of a particle.
Now our strategy is as follows. We assume first that the clusters of the white sites are not very large so that the particle, which change position during the time $`\tau _f`$ crosses them to and fro many times during the time $`\tau _s`$. In this approximation we find diffusion coefficient $`D_1`$. For the larger clusters we use the continuum approximation, which is valid at $`1p1`$ only, and find $`D_2`$. This coefficient gives the correct value at $`p=1`$. These two approximations have a large region of overlap if
$$\frac{\tau _f}{\tau _s}1.$$
(4)
Matching $`D_1`$ and $`D_2`$ in the overlap region we find the result which is exact if Eq. (4) is fulfilled.
To calculate $`D_1`$ we assume that the particle crosses each cluster many times during the time $`\tau _s`$ .Then the average quadratic displacement $`R^2(S)`$ is independent of $`\tau _s`$ and it is given by the equation
$$R^2(S)=\frac{a^2}{S^2}\underset{n=1}{\overset{S}{}}\underset{k=1}{\overset{S}{}}(nk)^2=\frac{a^2}{6}(S^21).$$
(5)
The probability that a particle is within a cluster of $`S`$ white sites is
$$w_s=(1p)^2p^{S1}S.$$
(6)
By averaging $`R^2(S)`$ over all the clusters one gets
$$\overline{r^2(\tau _s)}=\frac{a^2}{6}((1p)^2\underset{S=1}{\overset{\mathrm{}}{}}p^{S1}S(S^21)=\frac{a^2p}{(1p)^2}.$$
(7)
Thus,
$$D_1=\frac{\overline{r^2(\tau _s)}}{2\tau _s}=\frac{D_sp}{(1p)^2},$$
(8)
where $`D_s=a^2/2\tau _s`$. Note that Eq. (8) has been obtained by Druger et al.. The Eq.(8) is valid if $`\overline{r^2(\tau _s)}<<D_f\tau _s`$. It is fulfilled if
$$\frac{p}{(1p)^2}\frac{\tau _s}{\tau _f}.$$
(9)
At $`p1`$ one gets
$$D_1=\frac{a^2p}{\tau _s}.$$
(10)
To calculate $`D_2`$ we use the continuous approximation which is valid when the size of white clusters $`S1`$. This is true if $`1p1`$. In this approximation one should solve the diffusion equation
$$\frac{u(x,t)}{t}=D_f\frac{^2u(x,t)}{x^2}$$
(11)
assuming zero current at the beginning and at the end of the cluster ($`x=0,L`$)
$$\frac{u(x,t)}{x}|_{x=0,L}=0.$$
(12)
The initial condition has a form
$$u(x,t)|_{t=0}=\delta (xx_0),$$
(13)
where $`x_0`$ is a random point within the interval $`(0,L)`$.
The solution has a form
$$u(x,t)=\frac{1}{L}+\underset{n=1}{\overset{\mathrm{}}{}}\mathrm{cos}\left(\frac{\pi nx_0}{L}\right)\mathrm{cos}\left(\frac{\pi nx}{L}\right)\mathrm{exp}\left(\frac{\pi ^2n^2}{L^2}D_ft\right).$$
(14)
The mean squared displacement with respect to the initial position $`x_0`$ is
$`\overline{r_L^2(x_0,t)}={\displaystyle _0^L}(xx_0)^2u(x,t)𝑑x=`$ (15)
$`={\displaystyle \frac{L^2}{3}}(Lx_0)x_0+4L^2{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\mathrm{cos}\left(\frac{\pi nx_0}{L}\right)}{(\pi n)^2}}\left[{\displaystyle \frac{x_0}{L}}+(1)^n\left(1{\displaystyle \frac{x_0}{L}}\right)\right]\mathrm{exp}\left({\displaystyle \frac{\pi ^2n^2}{L^2}}D_ft\right)`$ (16)
Finally, taking the average over the initial positions $`x_0`$ on the cluster, we get the time dependence of the mean square displacement of the particle on the cluster of the size $`L`$.
$`\overline{r_L^2(t)}={\displaystyle \frac{1}{L}}{\displaystyle _0^L}\overline{r^2(x_0,t)}𝑑x_0=`$ (17)
$`={\displaystyle \frac{L^2}{6}}L^2{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{\left(\frac{\pi }{2}(2n+1)\right)^4}}\mathrm{exp}\left({\displaystyle \frac{\pi ^2(2n+1)^2}{L^2}}D_ft\right)=`$ (18)
$`={\displaystyle \frac{16}{\pi ^4}}L^2{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{(2n+1)^4}}\left[1\mathrm{exp}\left({\displaystyle \frac{\pi ^2n^2}{L^2}}D_ft\right)\right].`$ (19)
One can see from Eq. (19) that for relatively small times, $`t<<L^2/D_f`$, the displacement $`\overline{r_L^2(t)}`$ grows as $`2D_ft`$, just as for the normal diffusion, while at large times the value $`\overline{r_L^2(t)}`$ tends to the asymptotic value $`L^2/6`$ indicating that the particle crosses the cluster many times.
To find the diffusion coefficient $`D_2`$ one has to average $`\overline{r_L^2(\tau _s)}`$ over all clusters with the distribution function given by Eq. (6).
In the region $`(1p)<<1`$ one can substitute summation over the cluster sizes by integration to get
$`\overline{r^2(\tau _s)}={\displaystyle \frac{(1p)^2}{a^2}}{\displaystyle _0^{\mathrm{}}}\overline{r_L^2(\tau _s)}L\mathrm{exp}({\displaystyle \frac{L}{a}}\mathrm{ln}p)𝑑L=`$ (20)
$`={\displaystyle \frac{a^2}{(1p)^2}}{\displaystyle \frac{16}{\pi ^4a^2}}(1p)^2{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{(2n+1)^4}}{\displaystyle _0^{\mathrm{}}}L^3\mathrm{exp}\left({\displaystyle \frac{L}{a}}\mathrm{ln}p{\displaystyle \frac{\pi ^2n^2a^2}{2L^2}}{\displaystyle \frac{\tau _s}{\tau _f}}\right)𝑑L`$ (21)
The diffusion coefficient $`D_2`$ can be represented in a form
$`D_2={\displaystyle \frac{\overline{r^2(\tau _s)}}{2\tau _s}}={\displaystyle \frac{16D_s}{\pi ^4a^4}}(1p)^2{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{(2n+1)^4}}{\displaystyle _0^{\mathrm{}}}L^3\mathrm{exp}\left({\displaystyle \frac{L}{a}}\mathrm{ln}p\right)\left[1\mathrm{exp}\left({\displaystyle \frac{\pi ^2n^2a^2}{2L^2}}{\displaystyle \frac{\tau _s}{\tau _f}}\right)\right]𝑑L`$ (22)
or
$`D_2={\displaystyle \frac{16D_s}{\pi ^4(1p)^2}}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{(2n+1)^4}}{\displaystyle _0^{fty}}x^3\mathrm{exp}(x)\left[1\mathrm{exp}\left({\displaystyle \frac{\pi ^2(2n+1)^2}{2x^2}}{\displaystyle \frac{\tau _s}{\tau _f}}(\mathrm{ln}p)^2\right)\right]𝑑x`$ (23)
where $`x=\frac{L}{a}\mathrm{ln}p`$.
Introducing the dimensionless parameter $`A=\pi ^2(\mathrm{ln}p)^2\tau _s/2\tau _f`$ and changing the integration variable $`x=b_n\sqrt{z}`$, where $`b_n=\sqrt{A}(2n+1)`$ one can perform the summation over $`n`$ to get
$$D_2=\frac{a^2A}{\pi ^2\tau _f}_0^{\mathrm{}}\frac{z(1\mathrm{exp}(1/z))dz}{\mathrm{sinh}(\sqrt{Az})}.$$
(24)
At $`A1`$ one can neglect the exponent in the nominator of Eq.(24). Then
$$D_2=D_s/(1p)^2$$
(25)
At $`A1`$ one can expand this exponent to get $`D_2=D_f`$. One can also get the next term at small $`A`$. Thus,
$$D_2=D_f\left[1\frac{2\sqrt{A}}{pi^2}_0^{\mathrm{}}\sqrt{z}\left(\frac{1}{z}1+\mathrm{exp}\left(\frac{1}{z}\right)\right)𝑑z\right].$$
(26)
Calculating the integral in Eq. (26 ) one gets
$$D_2=D_f(10.479\sqrt{A})=D_f(11.06\sqrt{\tau _s/\tau _f}(1p))$$
(27)
at $`A1`$.
Note that the linear dependence in $`1p`$ means that the correction is proportional to the ratio of the mean displacement of a particle $`\sqrt{D_f\tau _s}`$ during the time $`\tau _s`$ to the typical size of a white cluster $`(1p)^1`$.
The scaling arguments in some different form than Eq. (1) can be applied to the one-dimensional case as well. It can be written in a form
$$D_2=D_fF(\frac{1p}{h^m}),$$
(28)
where $`h=\tau _f/\tau _s`$. It has the same form as Eq. (1) except that $`s=0`$. This relation is defined at $`p1`$ and $`1p1`$. In the same way as before, we put $`F(0)=1`$. At large values of $`(1p)/h^m`$ one has $`Dh/(1p)^q`$. It follows that $`mq=1`$ which is analogous to the relation $`s+mq=1`$ in the many-dimensional case. It follows from Eq.(25) that $`q=2`$. At $`1p1`$ one can write $`A=\pi ^2(1p)^2\tau _s/2\tau _f`$. Then it follows that $`m=1/2`$ in agreement with scaling relation. The exponent $`t`$ is not defined in the one-dimensional case.
Finally we present the solution for the effective diffusion coefficient $`D`$ which is valid at all values of $`p`$ within the interval $`0p1`$. It has been shown that $`D=D_1`$ if $`p/(1p)^2\tau _s/\tau _f`$. On the other hand $`D=D_2`$ if $`(1p)1`$. Thus, the two approximation have a wide region of overlap $`(p\tau _f/\tau _s)^{0.5}(1p)1`$. In this region $`D_1=D_sp/(1p)^2`$ and $`D_2=D_s/(1p)^2`$. Thus one can get a result which is exact everywhere if $`\tau _s\tau _f`$. This result is $`pD_2`$. Finally,
$$D=\frac{a^2Ap}{\pi ^2\tau _f}_0^{\mathrm{}}\frac{z(1\mathrm{exp}(1/z))dz}{\mathrm{sinh}(\sqrt{Az})}.$$
(29)
This result has been derived above for the site problem. One can see, however, that it remains unchanged for the bond problem as well.
The diffusion coefficient $`D/D_f`$ as given by Eq.(29) is plotted in Fig. 1 versus $`(1p)`$ in the double logarithmic scale at different values of $`\tau _s/\tau _f`$. The function $`(1p)^2`$ is also shown there as a dot-dashed line. One can see that at large values of $`\tau _s/\tau _f`$ the curves have wide regions which are parallel to this line. In these regions the diffusion coefficient increases as $`(1p)^2`$.
Finally, we have presented a novel problem which fills a gap between diffusion and percolation in case when the motion of the random media is very slow. We have considered the dc transport only. It would be interesting to study a frequency dependent transport in the same conditions.
We thank B. D. Laikhtman for a fruitful discussion. This work was supported by the US-Israel Binational Science Foundation, grant 9800097.
|
warning/0005/gr-qc0005042.html
|
ar5iv
|
text
|
# Self-force on a scalar particle in spherically-symmetric spacetime via mode-sum regularization: radial trajectories
## I introduction
The motion of a test pointlike mass (a “particle”) in orbit outside a black hole is commonly studied to model, and gain understanding of, realistic astrophysical scenarios involving highly relativistic two-body systems — particularly, the capture of a small compact object by a supermassive black hole . To describe the orbital evolution of such a particle on a strongly curved background, one must take into account non-geodesic effects caused by the interaction of the particle with its own gravitational field. This problem of deducing the self force (or “radiation reaction” force) exerted on the particle is often treated via perturbation theory: one assumes that the particle is endowed with a charge $`q`$ much smaller than the mass of the black hole (this charge may represent the particle’s mass, electric charge or—as in the toy model studied in the current paper—its scalar charge) and looks for the $`O(q^2)`$ correction to the equation of motion. The basic challenging task involved in this calculation, already in flat space, is, of course, correctly handling the divergence of the self field at the very location of the particle; namely, the introduction and justification of an appropriate regularization method. When considering the case of curved spacetime, additional difficulty arises due to the nonlocal nature of the self-force effect: waves emitted by the particle at some moment may backscatter off spacetime curvature and interact back with the particle at later stages of its motion. The occurrence of this so-called “tail” contribution to the self force results in that the calculation of the self force at a given moment requires, in principle, knowledge of the entire causal history of the particle. A number of methods (briefly surveyed below) have been proposed over the years for calculating the self force in curved backgrounds. The interest in this problem has greatly risen lately by virtue of recent years’ developments towards experimental gravitational waves detection, and the consequent need for accurate predictions for the orbital evolution of strongly gravitating two-body systems. Yet, actual calculations of the self force have been restricted thus far only to a few very simple cases (see below).
The standard technique for calculating the radiative evolution of orbits around black holes is the one based on energy-momentum balance considerations . In this approach one computes the flux to infinity (and across the horizon) of quantities associated with the constants of motion in the lack of self-force effects (specifically, in the Schwarzschild and Kerr backgrounds, the particle’s energy $`E`$ and azimuthal angular momentum $`L_z`$), thus deducing the temporal rate of change of these “constants”. Such balance calculations, though developed to a great extent, present two basic drawbacks: First, in the important Kerr case they are inapplicable for calculating the rate of change of the third constant of motion necessary for a full specification of the orbital evolution, i.e., the Carter constant $`Q`$, as this quantity is not additive. Second, these calculations do not account for the non-dissipative, yet important, part of the self force .
For the above reasons, a method based on direct calculation of the momentary force along the worldline seems more adequate. In the context of electromagnetism in flat space, such a method is familiar from the classic work by Dirac , concerning the electromagnetic self force on a (classical) electron. Dirac avoided the singularity of the self field at the particle’s location by introducing the “radiative potential”, constructed by taking the difference between the retarded and advanced electromagnetic potentials (what results in the cancellation of the problematic singular part). This procedure gave rise to what is now called the Abraham-Lorentz-Dirac (ALD) self force in flat space \[see Eq. (10) below for the analogous scalar particle case\]. The concept of “radiative potential” was much later employed by Gal’tsov for calculating the temporal rate of change of the energy and azimuthal angular momentum parameters for electrically charged particles orbiting a Kerr black hole. Though Gal’tsov analysis yielded correct results, it seems conceptually difficult, in general cases, to justify the use of such a non-causal approach. The problem becomes obvious when considering curved spacetime, where the self force exhibits nonlocal contributions: according to this approach, the force acting on a particle at a given moment turns out to be affected by the entire future evolution of the particle.
A formal method for calculating the momentary self force in curved spacetime, which employs the purely retarded Green’s function and is thus inherently causal, was developed long ago by DeWitt and Brehme (for the electromagnetic case). Dewitt and Brehme first carried out a local analysis of the retarded Green’s function near the particle’s worldline, based on the Hadamard expansion (which is, basically, an expansion of the Green’s function in powers of the geodesic distance between the source and evaluation points). Then, the particle’s equation of motion was deduced by imposing local energy-momentum conservation on a thin world-tube surrounding the particle’s worldline. Following the same approach (though using a different renormalization scheme), Mino, Sasaki, and Tanaka studied the gravitational self force by analyzing the local metric perturbations near a particle. They concluded that the regularized gravitational self force in vacuum is due solely to a nonlocal tail contribution. It remained unclear, though, how to practically evaluate the formal expression derived for this tail (see, however, recent attempts to tackle this problem ).
Recently, Quinn and Wald developed a different, axiomatic, approach to the regularization problem of the self force in curved spacetimes. Their approach relies on a “comparison axiom”, which allows the calculation of the self force by comparing the given problem with a (properly chosen) analogous problem in flat space (see for details). The implementation of this approach for both the electromagnetic and the gravitational cases yielded formal results in agreement with those obtained by DeWitt and Brehme (as corrected by Hobbs ; the main result by DeWitt and Brehme contained a trivial error) and by Mino et al., with the advantage of involving much simpler calculations. More recently, the same approach was applied by Quinn for a scalar particle \[the main results of this work, also quoted in Ref. , are summarized below; see Eqs. (9)–(13)\]. Again, however, despite the availability of a simple formal framework for obtaining equations of motion for a test particle in curved spacetime, the practical implementation of the formalism in actual calculations (particularly, the evaluation of the nonlocal tail contribution) remained a challenging task.
So far, the study of the self force effect in concrete situations have been restricted to very few simple cases. DeWitt and DeWitt employed the above mentioned formalism by DeWitt and Brehme to study the self force correction to the geodesic equation for an electrically charged particle freely falling in a static weak gravitational field, in the limit of small velocity. They concluded that a repulsive force of $`q^2r^3`$ magnitude (where $`r`$ is the Schwarzschild radial coordinate) would be experienced by such a particle, in addition to the usual attractive inverse-square force. Later, Smith and Will (and, independently, Frolov and Zel’nikov ) were able to derive an exact analytic expression for the $`O(q^2)`$ self force acting on an electrically charged particle held static in the Schwarzschild exterior. They found a repulsive self force of exact magnitude $`Mq^2r^3`$ \[measured by a momentarily static freely falling observer, and expressed in relativistic (geometrized) units\]. This result was later extended by Frolov and Zel’nikov to scalar and electrically-charged particles held static outside a charged, Reissner-Nordström type, black hole. It was concluded that, unlike in the electromagnetic case, no self force is experienced by a static scalar particle. This last result has been reproduced very recently by Wiseman in a thorough analysis of the self force acting on a static scalar particle in Schwarzschild spacetime.
The exact calculation in the static electric charge case was made possible owing to the existence of an exact analytic solution, discovered long ago by Copson (and later corrected by Linet ), for the electrostatic potential of a static charge in the Schwarzschild geometry. (The analogous closed-form solution for a scalar particle was constructed by Wiseman in .) In more general cases one cannot benefit from the existence of such exact solutions. The usual approach for treating this problem, of obtaining solutions to the field equation in black hole backgrounds, is through the Fourier-multipole decomposition of the field. In the context of the radiation reaction problem this approach seems to offer two obvious advantages: First, it allows, in the usual manner, reduction of the field equation \[originally a partial differential equation (DE) in 1+3 dimensions\] to an ordinary DE, thus making it accessible to simple numerical treatment. Second, each individual mode of the field turns out to be continuous (and the corresponding self force to be bounded) even at the particle’s location. Having this in mind, Ori previously proposed that a practical calculation of the self force effect may be carried out by first evaluating the effect of each Fourier-multipole mode of a particle’s self retarded field on its radiative evolution (through the local self force experienced by the particle), and then summing over all modes.
In Ref. the above sum-over-modes approach has been successfully applied for the calculation of the adiabatic, orbit-integrated, evolution rate of the three constants of motion in Kerr spacetime, including the Carter constant.<sup>*</sup><sup>*</sup>*However, whereas the mode sum for the evolution rate of the energy and azimuthal angular momentum parameters was shown to converge , it is not clear yet whether the corresponding mode sum for the Carter constant converges or not. However, it appears that a naive application of this method for the calculation of the momentary self force would not be useful, in general. The reason is that, although each mode yields a finite contribution to the self force, the sum over all modes is found, in general, to diverge. This situation manifests itself already in the most simple case, that of a static scalar charge in flat space: in this basic example, the contribution of each multipole mode to the radial component of the self force is the same, $`q^2/(2r_0^2)`$ (where $`r_0`$ is the distance of the particle from the origin of coordinates, with respect to which the spherical harmonic functions are defined), with an obvious divergence of the sum over modes. This, however, does not mean that one has to abandon the whole sum-over-modes approach; one may still be able to benefit from its advantages, by introducing a suitable regularization procedure into the calculation, properly designed to overcome the above kind of divergence.
In a previous paper we introduced the basic elements of a method for the calculation of the self force in curved spacetime through regularization of the mode sum. The implementation of the proposed calculation method for a specific trajectory in a given spacetime consists of two stages. First, one solves (numerically, in most cases) the appropriate ordinary DE for each Fourier-harmonic mode $`ł,m,\omega `$ of the retarded field, and evaluates the (finite) contribution of each of these modes to the self force. (Alternatively, one may numerically solve the 1+1 partial DE in the time domain, for each multipole mode $`l,m`$.) Then, the sum over all modes is made subject to a certain regularization procedure, which requires the knowledge of several regularization parameters. These parameters are derived analytically, for any given trajectory, through local perturbative analysis of the (retarded) Green’s function. In Ref. we outlined this regularization method as applied to a scalar particle moving on a Schwarzschild background, and presented final results (i.e., the values of all necessary regularization parameters) for the case of radial motion. The target of the present paper is three-fold: (i) providing a systematic presentation of the regularization scheme (including a discussion of some mathematical subtleties left untreated in ); (ii) providing full details of the calculations involved in deriving the regularization parameters for radial trajectories; and (iii) extending the analysis to a wider class of static spherically-symmetric black hole spacetimes.
This paper (as well as Ref. ) is concerned with the analytic part of the regularization scheme; namely, it sets the mathematical foundation for the scheme, and demonstrates the calculation of the regularization parameters involved in its implementation (in the example of radial motion). As explained above, full calculation of the self force requires the supplementary numerical determination of the various modes’ “bare” contributions to the self force. This was recently done for various trajectories of a scalar particle outside a Schwarzschild black hole: Burko first analyzed the case of a static particle and the one of a particle in circular motion (see also ). More recently, Barack and Burko applied the regularization scheme for studying radial trajectories in Schwarzschild . These numerical works confirm the applicability of the regularization scheme, and provide support for the values of the analytically deduced regularization parameters. Of course, they also yield significant physical information. In the static scalar particle case, Burko recovered the familiar result, of a zero self force. Calculations of the self force on a scalar particle in circular and radial trajectories were carried out for the first time (see Refs. for details).
The current paper is organized as follows: In Sec. II we give some preliminary relations involving the self field, the Green’s function, and the self force for a scalar particle. In Sec. III we decompose the Green’s function into its spherical harmonic components, and discuss the applicability of this expansion. In Sec. IV we decompose the (tail part of the) self force into its spherical harmonic contributions, discuss the need for regularization of the mode sum, and present the regularization scheme. The implementation of this scheme involves local analysis of the Green’s function modes for large multipole numbers, which is carried out in Sec. V. The particular case of radial motion is then considered in Secs. VI and VII, where the regularization parameters for this case are explicitly calculated. In Sec. VIII we summarize, discuss possible extensions of the analysis, and briefly survey some related work.
## II Self force on a scalar charge: preliminaries
We consider a class of static spherically-symmetric (not necessarily vacuum) black hole geometries, having a line element of the form Throughout this paper we use relativistic units (with $`G=c=1`$), and metric signature $`+++`$.
$$ds^2=f(r)dt^2+f^1(r)dr^2+r^2(d\theta ^2+\mathrm{sin}^2\theta d\phi ^2),$$
(1)
where $`t`$, $`r`$, $`\theta `$, and $`\phi `$ are the Schwarzschild coordinates, and $`f`$ is a function of $`r`$ only, positive outside the event horizon. Important members of this class include the Schwarzschild solution, with $`f=12M/r`$; the Reissner-Nordström solution, with $`f=12M/r+Q/r^2`$; and the Schwarzschild-de Sitter solution, with $`f=12M/r+\mathrm{\Lambda }r^2/3`$. Here, $`M`$ stands for the black hole’s mass, $`Q`$ represents its net electric charge, and $`\mathrm{\Lambda }`$ is the cosmological constant.
We next consider a point-like particle of scalar charge $`q`$, with $`|q|M`$, moving in a spacetime of the above type. Let $`x^\mu =x_p^\mu (\tau )`$ represent the particle’s worldline (not necessarily a geodesic), with $`\tau `$ being its proper time. The scalar particle exhibits a Klein-Gordon field $`\mathrm{\Phi }`$, satisfying
$$\mathrm{}\mathrm{\Phi }\mathrm{\Phi }_{;\alpha }{}_{}{}^{;\alpha }=\frac{1}{\sqrt{g}}\left(\sqrt{g}g^{\alpha \beta }\mathrm{\Phi }_{,\alpha }\right)_{,\beta }=4\pi \rho (x^\mu ),$$
(2)
where $`\mathrm{}`$ represents the covariant D’Alembertian operator, $`g`$ is the metric determinant, and $`\rho (x^\mu )`$ is the scalar charge density, given by
$$\rho (x^\mu )=q_{\mathrm{}}^{\mathrm{}}\frac{1}{\sqrt{g}}\delta ^4(x^\mu x_p^\mu (\tau ))𝑑\tau .$$
(3)
The solution for the scalar field can be written as
$$\mathrm{\Phi }(x^\mu )=G(x^\mu ;x_{}^{}{}_{}{}^{\mu })\rho (x_{}^{}{}_{}{}^{\mu })\sqrt{g}d^4x^{},$$
(4)
where $`G(x^\mu ;x_{}^{}{}_{}{}^{\mu })`$ is the retarded Green’s function, satisfying
$$\mathrm{}G(x^\mu ;x_{}^{}{}_{}{}^{\mu })=\frac{4\pi }{\sqrt{g}}\delta ^4(x^\mu x_{}^{}{}_{}{}^{\mu }),$$
(5)
and subject to the causality condition, $`G=0`$ whenever $`x^\mu `$ lies outside the future light cone of $`x_{}^{}{}_{}{}^{\mu }`$. Combining Eqs. (3) and (4) we obtain for the scalar field
$$\varphi (x^\mu )=q_{\mathrm{}}^{\mathrm{}}G[x^\mu ;x_p^\mu (\tau )]𝑑\tau .$$
(6)
The “scalar force” experienced by the particle due to its own field shall be taken, following , to be
$$F_\alpha =q\mathrm{\Phi }_{;\alpha }=q\mathrm{\Phi }_{,\alpha },$$
(7)
evaluated at the particle’s location. We comment that the so-defined force is not perpendicular to the four-velocity of the particle, $`u^\alpha dx_p^\alpha (\tau )/d\tau `$, resulting in that the mass parameter of the particle is not conserved along the worldline. Indeed, the force on a scalar particle can be calculated otherwise (as in , e.g.), such as to make the mass parameter conserved: $`F_\alpha ^{}=q(\mathrm{\Phi }_{,\alpha }+u_\alpha u^\beta \mathrm{\Phi }_{,\beta })`$. Although we shall adopt here the simpler definition, Eq. (7), the results of our analysis could then easily be applied for the force $`F_\alpha ^{}`$ as well. (Given all vectorial components of $`F_\alpha `$, one can easily construct both the force component perpendicular to the worldline, and the component tangent to it.) With Eq. (6) we now have for the self force acting on the particle at a point $`x_0^\mu x_p^\mu (\tau =0)`$ along its worldline,
$$F_\alpha =q^2_{\mathrm{}}^{\mathrm{}}G_{,\alpha }[x^\mu ;x_p^\mu (\tau )]𝑑\tau ,$$
(8)
where the gradient (taken with respect to $`x^\mu `$) is to be evaluated at $`x^\mu =x_0^\mu `$.
The “bare” self force given in Eq. (8) needs to be regularized to avoid divergences associated with the behavior of the scalar field at the very location of the particle. For that goal, Quinn applied the “comparison axiom” approach by Quinn and Wald for the scalar particle case. The total self force acting on the scalar particle was found to be composed of three parts:
$$F_\alpha ^{(\mathrm{total})}=F_\alpha ^{(\mathrm{ALD})}+F_\alpha ^{(\mathrm{Ricci})}+F_\alpha ^{(\mathrm{tail})}.$$
(9)
The first term here is a local ALD-like term, reading
$$F_\alpha ^{(\mathrm{ALD})}=\frac{1}{3}q^2(\dot{a}_\alpha a^2u_\alpha ),$$
(10)
where $`a^\alpha `$ is the four-acceleration of the particle, $`a^2a_\beta a^\beta `$, and an overdot represents covariant differentiation with respect to the particle’s proper time $`\tau `$. The second term in Eq. (9) is related to the local Ricci curvature at the particle location. It is given by
$$F_\alpha ^{(\mathrm{Ricci})}=\frac{1}{6}q^2(R_{\alpha \beta }u^\beta +u_\alpha R_{\beta \gamma }u^\beta u^\gamma Ru_\alpha /2),$$
(11)
where $`R_{\alpha \beta }`$ is the Ricci tensor and $`R`$ is the curvature scalar. The third term in the expression for the total self force represents the non-local “tail” contribution. It may be expressed as
$$F_\alpha ^{(\mathrm{tail})}\underset{ϵ0^+}{lim}F_\alpha ^{(ϵ)},$$
(12)
where
$$F_\alpha ^{(ϵ)}q^2_{\mathrm{}}^ϵG_{,\alpha }[x_0^\mu ;x_p^\mu (\tau )]𝑑\tau .$$
(13)
As we mentioned above, the occurrence of a tail term — a prominent feature of the self force in curved spacetimes — is due to the Green’s function having its support also inside the source’s future light cone. From the physical point of view, this is associated with the fact that waves are scattered off spacetime curvature while propagating on a curved background.
The task of implementing the formal expression, Eq. (9), in practical calculations of the self force is a challenging one. The difficulty stems, of course, from the need to evaluate the tail part, which requires knowledge of the Green’s function everywhere along the particle’s past worldline. Below we therefore focus on the tail term contribution to the self force, presenting a practical method for its calculation.
## III Multipole decomposition of the Green’s function
The regularization scheme to be introduced below is based on evaluating the contribution to the (tail part of the) self force due to each multipole mode of the (retarded) Green’s function. To that end we first consider the multipole decomposition of the Green’s function.
To begin, one may be tempted to decompose $`G`$ into its multipole modes $`G^l`$ in the usual manner, as $`G=_{l=0}^{\mathrm{}}G^l`$ (where $`G^l`$ represents the quantity resulting from summing over azimuthal numbers $`m`$). Although this may look as standard procedure, caution is necessary here: in general, such a decomposition turns out ill-defined, as the sum over $`l`$ is found to diverge. This can be illustrated already in flat space. In this case, the modes $`G^l`$ admit a closed-form expression, which, for evaluation point $`x^\mu `$ lying inside the future light cone of the source point $`x_{}^{}{}_{}{}^{\mu }`$, is given by
$$G_{\mathrm{Flat}}^l(x^\mu ,x_{}^{}{}_{}{}^{\mu })=\frac{(2l+1)P_l(\mathrm{cos}\chi )P_l[1\sigma /(rr^{})]}{2rr^{}}.$$
(14)
Here, $`P_l`$ is the Legendre polynomial, $`\sigma \frac{1}{2}\left[(tt^{})^2(rr^{})^2\right]`$, and
$$\mathrm{cos}\chi \mathrm{cos}\theta \mathrm{cos}\theta ^{}+\mathrm{sin}\theta \mathrm{sin}\theta ^{}\mathrm{cos}(\phi \phi ^{}).$$
(15)
\[Eq. (14) can be verified by direct substitution, using Eqs. (27), (28), and (30), to be given below.\] Consider, for example, the case $`\chi =0`$, with $`P_l(\mathrm{cos}\chi )1`$ for all $`l`$, corresponding to both the source and evaluation points lying in the same radial direction. At large $`l`$ values, the Legendre polynomial $`P_l(\sigma )`$ admits the asymptotic form $`l^{1/2}\times `$ oscillations with respect to $`l`$ \[the exact asymptotic form is given in App. A below; see Eq. (A16)\]. Thus, at large $`l`$ one finds $`G_{\mathrm{Flat}}^ll^{1/2}\times `$ oscillations, implying that the infinite sum over all modes fails to converge.In the more general case, with $`\chi 0`$, there is also a $`l^{1/2}\times \mathrm{oscillations}`$ factor coming at large $`l`$ from $`P_l(\mathrm{cos}\chi )`$, yielding the asymptotic form $`G_{\mathrm{Flat}}^l\mathrm{const}\times `$ oscillations. Hence, clearly, the sum over modes $`G^l`$ diverges in the general case as well. Mathematically speaking, this failure of the naive multipole decomposition may be associated with the fact that the Green’s function of our problem exhibits a strong irregularity along the intersection of the future light cone of the source with the sphere of constant $`r`$ and $`t`$. In Appendix A we discuss this irregularity in more detail, referring to the analysis by DeWitt and Brehme . We then suggest a way to overcome the difficulty caused by the presence of the irregularity, and explain how a well defined mode decomposition can still be accomplished. Although the detailed discussion of this issue is left to the Appendix, we outline here the basic argument, and present some definitions and notation needed in the sequel.
In Appendix A we construct a “modified” Green’s function $`G_{\mathrm{mod}}=G\delta G`$, where the function $`\delta G`$ is chosen such that $`G_{\mathrm{mod}}`$ has the following properties: (i) it is a continuous function of $`\theta `$ and $`\phi `$ across the sphere of constant $`r`$ and $`t`$; and (ii) it yields the same self force, through Eq. (13), as the original Green’s function $`G`$ (this is guaranteed by taking the function $`\delta G`$ to have no support inside the future light cone of the source). It is then argued that the modified Green’s function admits an (absolutely) convergent multipole expansion, $`G_{\mathrm{mod}}=_l(G^l\delta G^l)`$. Next, we define the new operation $`\stackrel{~}{lim}_l\mathrm{}`$ (“tilde-limit”) of a series of numbers $`A_l`$, as the standard limit $`lim_l\mathrm{}`$ (when existing and finite) of the series $`A_lB_l^{(1)}B_l^{(2)}\mathrm{}B_l^{(k)}`$, where the $`B_l^{(j)}`$’s are any finite number of terms having the form $`B_l^{(j)}=a_jl^{b_j}\mathrm{cos}(\alpha _jl+\beta _j)`$, with $`a_j`$, $`b_j`$, $`\alpha _j`$, and $`\beta _j`$ being some $`l`$-independent real numbers, and with none of the numbers $`\alpha _j`$ vanishing. Namely, if there exist $`k`$ quantities $`B_l^{(j)}`$ of the above form, such that subtracting them from the original series $`A_l`$ would yield a well-defined finite limit as $`l\mathrm{}`$, then we define
$$\stackrel{~}{lim}_l\mathrm{}A_l\underset{l\mathrm{}}{lim}\left[A_l\underset{j=1}{\overset{k}{}}B_l^{(j)}\right].$$
(16)
We also define the “tilde-sum” of a series $`A_l`$ by
$$\underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}A_l\stackrel{~}{lim}_{\overline{l}\mathrm{}}\underset{l=0}{\overset{\overline{l}}{}}A_l,$$
(17)
where $`_l^{\overline{l}}`$ is the standard summation operation. It can be easily verified (see App. A) that when the “tilde-limit” (or the “tilde-sum”) of a series exists, then it is unique. In particular, if a standard infinite sum $`_l^{\mathrm{}}A_l`$ converges, then one may replace it with a “tilde-sum” operation. Thus, we may replace the convergent standard sum $`G_{\mathrm{mod}}=_{l=0}^{\mathrm{}}(G^l\delta G^l)`$ with a tilde-summation. In Appendix A we show that $`_{l=0}^\stackrel{~}{\mathrm{}}(\delta G^l)`$=0. Consequently, we conclude
$$G_{\mathrm{mod}}=\underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}G^l.$$
(18)
We emphasize once more that the modified Green’s function $`G_{\mathrm{mod}}`$ can serve instead of the original function $`G`$ for the calculation of self force, as both functions yield the same force. Eq. (18) thus implies that the calculation of the self force can be carried out through analysis of the original Green’s function’s modes $`G^l`$, by applying the tilde-summation instead of the (ill-defined) standard summation. (A more thorough discussion of the arguments leading to Eq. (18) will be given in Appendix A.)
Let us now turn to study the form of the multipole modes $`G^l`$ in greater detail. These modes can be written more explicitly as
$$G^l(x^\mu ,x_{}^{}{}_{}{}^{\mu })=\underset{m=l}{\overset{l}{}}Y_{lm}(\theta ,\phi )\widehat{g}^{lm}(t,r;x_{}^{}{}_{}{}^{\mu }),$$
(19)
where $`Y^{lm}(\theta ,\phi )`$ are the standard spherical harmonic functions on the sphere of constant $`r`$ and $`t`$. Substituting Eq. (19) and the relation $`\delta (\theta \theta ^{})\delta (\phi \phi ^{})/\mathrm{sin}\theta =_{l,m}Y_{lm}(\theta ,\phi )Y_{lm}^{}(\theta ^{},\phi ^{}),`$ (where an asterisk denotes complex conjugation) in Eq. (5), we obtain from the orthogonality of the spherical harmonics,
$$r^2f^1(r)\widehat{g}_{,tt}^{lm}\left[r^2f(r)\widehat{g}_{,r}^{lm}\right]_{,r}+l(l+1)\widehat{g}^{lm}=4\pi \delta (tt^{})\delta (rr^{})Y_{lm}^{}(\theta ^{},\phi ^{}).$$
(20)
In terms of the $`m`$-independent variable $`\stackrel{~}{g}^l(t,r;t^{},r^{})`$, defined through
$$\widehat{g}^{lm}=2\pi \stackrel{~}{g}^lY_{lm}^{}(\theta ^{},\phi ^{})/(rr^{}),$$
(21)
Eq. (20) becomes
$$\stackrel{~}{g}_{,tt}^l\stackrel{~}{g}_{,r_{}r_{}}^l+4V^l(r)\stackrel{~}{g}^l=2f(r)\delta (tt^{})\delta (rr^{}),$$
(22)
where the radial coordinate $`r_{}(r)`$ admits $`dr_{}/dr=f^1(r)`$, and the effective potential $`V^l(r)`$ is given by
$$V^l(r)=\frac{1}{4}f(r)\left(\frac{l(l+1)}{r^2}+\frac{f^{}(r)}{r}\right)$$
(23)
(with a prime denoting $`d/dr`$).
To account for the causality condition, it is convenient to introduce the (Eddington-Finkelstein-like) null coordinates
$$vt+r_{}\text{and}utr_{}.$$
(24)
The relation $`\delta (tt^{})\delta (rr^{})=2f^1(r^{})\delta (vv^{})\delta (uu^{})`$ can then be used to write Eq. (22) in the simple form
$$\stackrel{~}{g}_{,vu}^l+V^l(r)\stackrel{~}{g}^l=\delta (vv^{})\delta (uu^{}).$$
(25)
We now impose causality by writing
$$\stackrel{~}{g}^l=g^l(v,u;v^{},u^{})\mathrm{\Theta }(vv^{})\mathrm{\Theta }(uu^{}),$$
(26)
where $`\mathrm{\Theta }`$ is the standard step function. The “reduced Green’s function” $`g^l(v,u;v^{},u^{})`$ obeys the homogeneous equation
$$g_{,vu}^l+V^l(r)g^l=0$$
(27)
for all $`u>u^{}`$ and $`v>v^{}`$. Substituting Eq. (26) into Eq. (25) and examining the behavior along the null rays $`v=v^{}`$ and $`u=u^{}`$, one finds that $`g^l`$ must admit
$$g^l(v=v^{})=g^l(u=u^{})=1.$$
(28)
For any fixed source point $`v^{},u^{}`$, the homogeneous equation (27), supplemented by the initial conditions (28), constitutes a characteristic initial-value problem for the function $`g^l`$ anywhere at $`u>u^{}`$ and $`v>v^{}`$.
Finally, to express $`G^l`$ in terms of the reduced Green’s function $`g^l`$, we substitute Eq. (21) \[with Eq. (26)\] into Eq. (19). In the resulting expression we can explicitly sum over $`m`$ by making use of the relation
$$\underset{m=l}{\overset{l}{}}Y_{lm}(\theta ,\phi )Y_{lm}^{}(\theta ^{},\phi ^{})=(4\pi )^1(2l+1)P_l(\mathrm{cos}\chi ),$$
(29)
where $`\mathrm{cos}\chi `$ is the quantity given in Eq. (15). We then find for the $`l`$-mode of the Green’s function,
$$G^l=LP_l(\mathrm{cos}\chi )\frac{g^l(v,u;v^{},u^{})}{rr^{}}\mathrm{\Theta }(vv^{})\mathrm{\Theta }(uu^{}),$$
(30)
where we have set
$$Ll+1/2.$$
(31)
## IV Mode sum regularization scheme
### A The need for a mode sum regularization
Following the discussion of the preceding section, we now replace $`G`$ in Eq. (13) by $`G_{\mathrm{mod}}`$, and then substitute for $`G_{\mathrm{mod}}`$ from Eq. (18). We find<sup>§</sup><sup>§</sup>§ It is assumed here that both the differentiation and the integration involved in constructing $`F_\alpha ^{(ϵ)}`$ out of $`G`$ can be performed term-by-term with respect to the tilde-summation. This assumption should be verified by more closely inspecting the convergence properties of the tilde-sum over $`G^l`$ in Eq. (18), which, however, would be beyond the scope of the current paper.
$`F_\alpha ^{(ϵ)}={\displaystyle \underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}}F_\alpha ^{l(ϵ)},`$ (32)
where $`F_\alpha ^{l(ϵ)}`$ represents the contribution to $`F_\alpha ^{(ϵ)}`$ associated with the $`l`$-mode Green’s function:
$$F_\alpha ^{l(ϵ)}=q^2_{\mathrm{}}^ϵG_{,\alpha }^l[x_0^\mu ;x_p^\mu (\tau )]𝑑\tau .$$
(33)
For practical reasons which become clear below, let us now write
$$F_\alpha ^{l(ϵ)}=F_\alpha ^l\delta F_\alpha ^{l(ϵ)},$$
(34)
in which Strictly speaking, the two quantities $`F_\alpha ^l`$ and $`\delta F_\alpha ^{l(ϵ)}`$ are not well defined without specifying the direction through which the gradient of $`G^l`$ is calculated. This issue is discussed in length later in this section.
$$F_\alpha ^l=q^2_{\mathrm{}}^{\mathrm{}}G_{,\alpha }^l[x_0^\mu ;x_p^\mu (\tau )]𝑑\tau \text{and}\delta F_\alpha ^{l(ϵ)}=q^2_ϵ^{\mathrm{}}G_{,\alpha }^l[x_0^\mu ;x_p^\mu (\tau )]𝑑\tau .$$
(35)
Here, $`F_\alpha ^l`$ is the $`l`$-mode of $`F_\alpha =q\mathrm{\Phi }_{,\alpha }`$—the quantity given in Eq. (8), which is sourced by the entire worldline. This quantity can be obtained from the $`l`$-mode of the self field, which, in turn, can be calculated essentially with no difficulty (using numerical methods, in most cases ). The other quantity appearing in Eq. (34), $`\delta F_\alpha ^{l(ϵ)}`$, is local in nature, and thus may be treated, in principle, by means of local analytic methods (as we, indeed, demonstrate in this paper).
In terms of $`F_\alpha ^l`$ and $`\delta F_\alpha ^{l(ϵ)}`$, the tail part of the self force is calculated through
$$F_\alpha ^{(\mathrm{tail})}=\underset{ϵ0^+}{lim}\underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}\left(F_\alpha ^l\delta F_\alpha ^{l(ϵ)}\right).$$
(36)
To carry out this calculation, one may tempt to first calculate the sum over $`F_\alpha ^l`$ (which is $`ϵ`$-independent), and then evaluate the local contribution $`lim_{ϵ0^+}\stackrel{~}{}_l\delta F_\alpha ^{l(ϵ)}`$. However, here one comes across a problem: Although each of the modes $`F_\alpha ^l`$ yields a finite contribution at the particle’s location, in general the sum over all modes $`F_\alpha ^l`$ diverges. As we mentioned in the Introduction, this can be demonstrated even in the simple case of a static scalar charge in flat space. To overcome this type of divergence, the introduction of a certain regularization procedure for the mode sum is required. Such a procedure is described (and later implemented) in what follows.
### B The regularization scheme
To regularize the modes sum, one seeks a (simple as possible) $`ϵ`$-independent function $`h_\alpha ^l`$, such that the series $`_l(F_\alpha ^lh_\alpha ^l)`$ would converge. Once such a function is found, Eq. (36) can be written as
$$F_\alpha ^{(\mathrm{tail})}=\underset{l=0}{\overset{\mathrm{}}{}}\left(F_\alpha ^lh_\alpha ^l\right)D_\alpha ,$$
(37)
where
$$D_\alpha \underset{ϵ0^+}{lim}\underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}\left(\delta F_\alpha ^{l(ϵ)}h_\alpha ^l\right).$$
(38)
In principle, a regularization function $`h_\alpha ^l`$ can be constructed by investigating the asymptotic behavior of $`F_\alpha ^l`$ as $`l\mathrm{}`$. It is also possible, however, to derive $`h_\alpha ^l`$ from the large-$`l`$ asymptotic behavior of $`\delta F_\alpha ^{l(ϵ)}`$: The latter and $`F_\alpha ^l`$ must have the same singular behavior at the tilde-limit $`l\mathrm{}`$ (for fixed $`ϵ`$), as their difference yields a convergent tilde-sum over $`l`$. Obviously, in order to determine $`h_\alpha ^l`$ (and $`D_\alpha `$) from $`\delta F_\alpha ^{l(ϵ)}`$, one merely needs the asymptotic behavior of $`\delta F_\alpha ^{l(ϵ)}`$ in the immediate neighborhood of $`ϵ=0`$. This allows one to derive $`h_\alpha ^l`$ (and $`D_\alpha `$) using local analytic methods, as shall be demonstrated in the next section.
First, however, it would be necessary to comment here about a certain indefiniteness involved in the above definitions of the quantities $`F_\alpha ^l`$ and $`\delta F_\alpha ^{l(ϵ)}`$.
### Discontinuity of $`F_\alpha ^l`$ and $`\delta F_\alpha ^{l(ϵ)}`$
Whereas the quantity $`F_\alpha ^{l(ϵ)}`$ of Eq. (33) is well defined, the values of the two quantities $`F_\alpha ^l`$ and $`\delta F_\alpha ^{l(ϵ)}`$ depend on how exactly one evaluates the gradient $`G_{,\alpha }^l`$ at the particle’s location. To make this point clear, consider first the $`r`$-components $`F_r^l`$ and $`\delta F_r^{l(ϵ)}`$. These are calculated according to Eq. (35) from the $`r`$-derivative of $`G^l`$, reading
$`G_{,r}^l={\displaystyle \frac{LP_l(\mathrm{cos}\chi )}{rr^{}}}\left\{\left[g_{,r}^lg^l/r\right]\mathrm{\Theta }(vv^{})\mathrm{\Theta }(uu^{})+f^1g^l\left[\delta (vv^{})\mathrm{\Theta }(uu^{})\mathrm{\Theta }(vv^{})\delta (uu^{})\right]\right\}.`$ (39)
To calculate $`F_r^l`$ and $`\delta F_r^{l(ϵ)}`$ one needs to evaluate this derivative at the self force’s evaluation point, $`x^\mu =x_0^\mu `$, with a source point $`x_{}^{}{}_{}{}^{\mu }=x_p^\mu (\tau )`$. Now, if the derivative at $`x_0^\mu `$ is calculated from $`r_0^+`$ \[namely, by taking the limit $`rr_0^+`$ of $`\frac{G^l(r)G^l(r_0)}{rr_0}`$\], then the $`\delta (uu^{})`$ term in Eq. (39) will have a nonvanishing contribution to $`F_r^l`$ and to $`\delta F_r^{l(ϵ)}`$ \[through the integrals in Eq. (35)\], whereas the $`\delta (vv^{})`$ term will have no contribution — see figure 1. On the other hand, if the derivative is taken from $`r^{}`$, it will be the $`\delta (vv^{})`$ term to contribute, and the $`\delta (uu^{})`$ term to have no contribution. One can easily verify (as we explicitly do in the following section) that these two different $`\delta `$ terms yield different contributions to the integrals in Eq. (35). Thus, although each of the quantities $`F_r^l`$ and $`\delta F_r^{l(ϵ)}`$ has well defined values when calculated from either the limit $`rr_0^{}`$ or the limit $`rr_0^+`$, these two one-sided values do not coincide. \[Note that the quantity $`F_r^{l(ϵ)}`$ defined in Eq. (33) does not exhibit this kind of discontinuity, as for any finite $`ϵ`$ neither of the two $`\delta `$ terms contribute to this quantity.\]
One can similarly show that the $`t`$-components $`F_t^l`$ and $`\delta F_t^{l(ϵ)}`$ also exhibit this kind of discontinuity through the particle’s location (see the explicit calculation carried out in section V below). On the other hand, the $`\theta `$ and $`\phi `$ components are obviously continuous through the particle’s location, as $`G^l`$ depends on the angular coordinates only through the regular function $`P_l(\mathrm{cos}\chi )`$.
For the sake of definiteness, we shall denote by $`F_r^{l+}`$ and $`\delta F_r^{l(ϵ)+}`$ the one-sided values arising from the $`rr_0^+`$ limit, and by $`F_r^l`$ and $`\delta F_r^{l(ϵ)}`$ the ones arising from the $`rr_0^{}`$ limit. In addition, the symbols $`F_t^{l\pm }`$ and $`\delta F_t^{l(ϵ)\pm }`$ will stand for the values derived from the limit $`tt_0^{}`$ if $`dr/dt>0`$ at the force’s evaluation point, or from the limit $`tt_0^\pm `$ if $`dr/dt<0`$ there. (In case $`dr/dr=0`$ at the force’s evaluation point, the two one-sided values of the $`t`$-component turn out to coincide, as we obtain below.) With this notation we find, for $`\alpha =r`$ or $`t`$,
$$\delta F_\alpha ^{l(ϵ)\pm }=q^2_ϵ^{0^+}G_{,\alpha }^{l\pm }[x_0^\mu ;x_p^\mu (\tau )]𝑑\tau $$
(40)
(and similarly for $`F_\alpha ^{l\pm }`$), where
$$G_{,r}^{l\pm }(x^\mu ;x_{}^{}{}_{}{}^{\mu })\frac{LP_l(\mathrm{cos}\chi )}{rr^{}}\left[g_{,r}^lg^l/rf^1g^l\delta (w_\pm w_\pm ^{})\right],$$
(41)
and
$$G_{,t}^{l\pm }(x^\mu ;x_{}^{}{}_{}{}^{\mu })\frac{LP_l(\mathrm{cos}\chi )}{rr^{}}\left[g_{,t}^l+g^l\delta (w_\pm w_\pm ^{})\right].$$
(42)
Here, we have introduced the notation
$$w_+u\text{and}w_{}v,$$
(43)
and, likewise, $`w_+^{}u^{}`$ and $`w_{}^{}v^{}`$.
## V Local analysis of $`G^l`$ for large $`l`$
The execution of the regularization procedure introduced above involves the construction of the quantities $`h_\alpha ^l`$ and $`D_\alpha ^l`$. As we pointed out earlier, this can be done by analyzing $`\delta F_\alpha ^{l(ϵ)}`$, or, more accurately, the quantities $`\delta F_\alpha ^{l(ϵ)\pm }`$ given in Eq. (40). For that goal, we must have sufficient information about the Green’s function’s $`l`$-mode $`G^l`$, for large values of $`l`$, at the immediate vicinity of the self force evaluation point. In this section we use local analysis to obtain analytic approximation for $`G^l`$, up to the accuracy needed for the derivation of $`h_\alpha ^l`$ and $`D_\alpha ^l`$.
### A perturbation analysis
In Sec. III above we have reduced the problem of calculating $`G^l`$ to that of solving a $`(1+1)`$-dimensional homogeneous partial DE for the function $`g^l`$, Eq. (27), with the characteristic initial data specified in Eq. (28). Given the function $`g^l`$, the “four-dimensional” Green’s function $`l`$-mode, $`G^l`$, is then constructed from Eq. (30).
To explore the behavior of the function $`g^l`$ for small spacetime intervals and large $`l`$, we apply the following perturbation analysis. Let us separate the effective potential given in Eq. (23) into two pieces, in the form
$$V^l(r)=L^2V_0(r)+V_1(r),$$
(44)
where
$$V_0(r)=\frac{f(r)}{4r^2},\text{and}V_1(r)=\frac{f(r)}{16r^2}\left[4rf^{}(r)1\right].$$
(45)
Let us next expand $`V^l(r)`$ in a Taylor series in the small deviation $`rr_0`$ about the particle’s location $`r=r_0`$. It is convenient to take the small expansion parameter to be $`r_{}r_0`$, yielding
$$V^l(r)=V^l(r_0)+\overline{V^l}(r_0)(r_{}r_0)+\frac{1}{2}\overline{\overline{V^l}}(r_0)(r_{}r_0)^2+\mathrm{},$$
(46)
where an overbar denotes $`d/dr^{}`$. Let us also define
$$\mathrm{\Delta }_r2V_{00}^{1/2}L(r_{}r_0),$$
(47)
where $`V_{00}V_0(r_0)`$. We shall refer to a variable of this kind, having the form $`L\times \text{(small spacetime deviation)}`$, as a “neutral” variable. Such “neutral” variables shall play an important role in our analysis, allowing one to properly take into account the delicate interplay between large $`l`$ and small spacetime deviations. Expressing $`r_{}r_0`$ in terms of $`\mathrm{\Delta }_r/L`$, substituting in the above Taylor expansion, and collecting terms of the same powers in $`L`$ (with fixed $`\mathrm{\Delta }_r`$), Eq. (46) takes the form
$$V^l(r)=V_{00}\left[L^2+L\left(f_1\mathrm{\Delta }_r\right)+\left(f_2+f_3\mathrm{\Delta }_r^2\right)\right]+O(1/L),$$
(48)
where $`f_1`$, $`f_2`$, and $`f_3`$ are coefficients given by
$`f_1`$ $``$ $`{\displaystyle \frac{1}{2}}V_{00}^{3/2}\overline{V_0}=f^{1/2}(rf^{}2f),`$ (49)
$`f_2`$ $``$ $`V_{00}^1V_1=rf^{}1/4,`$ (50)
$`f_3`$ $``$ $`{\displaystyle \frac{1}{8}}V_{00}^2\overline{\overline{V_0}}={\displaystyle \frac{r^2}{2}}\left[(f^{})^2/f+f^{\prime \prime }\right]+3(frf^{}).`$ (51)
(Here, all quantities are evaluated at $`r=r_0`$.)
Defining now the dimensionless “neutral” coordinates
$$y=V_{00}^{1/2}L(vv^{})\text{and}x=V_{00}^{1/2}L(uu^{}),$$
(52)
Eq. (27) becomes
$$g_{,yx}^l+\left[1+\frac{f_1\mathrm{\Delta }_r}{L}+\frac{f_2+f_3\mathrm{\Delta }_r^2}{L^2}+O\left(1/L^3\right)\right]g^l=0.$$
(53)
We next expand $`g^l`$ in the form
$$g^l=\underset{k=0}{\overset{\mathrm{}}{}}L^kg_k(\mathrm{\Delta }_r,\mathrm{\Delta }_r^{},z),$$
(54)
where the expansion coefficients $`g_k`$ are considered as being dependent of only the “neutral” variables $`\mathrm{\Delta }_r`$,
$$\mathrm{\Delta }_r^{}2V_{00}^{1/2}L(r_{}^{}r_0),$$
(55)
and
$$z2\sqrt{xy}=(L/r_0)s.$$
(56)
Here, $`s`$ is the geodesic distance, to leading order in $`rr^{}`$, between the Green’s function evaluation and source points (when these two points have the same $`\theta `$ and $`\phi `$ values):
$$s=\left[f(r_0)(vv^{})(uu^{})\right]^{1/2}.$$
(57)
Substituting now the expansion (54) into Eq. (53) and comparing powers of $`L`$, we obtain a hierarchy of equations for the various functions $`g_k`$, having the form
$$g_{k,yx}+g_k=S_k.$$
(58)
Here, the source $`S_k`$ is determined for each $`k>0`$ by the functions $`g_{k^{}<k}`$ preceding $`g_k`$ in the hierarchy. In the analysis below we shall only need the terms with $`k=0`$, $`1`$, and $`2`$. For these values of $`k`$, the source terms are given by
$$S_0=0,$$
(60)
$$S_1=f_1\mathrm{\Delta }_rg_0,$$
(61)
$$S_2=f_1\mathrm{\Delta }_rg_1(f_2+f_3\mathrm{\Delta }_r^2)g_0.$$
(62)
Finally, to complete the formulation of a characteristic initial data problem for each of the functions $`g_k`$, we supplement Eq. (58) with the initial conditions
$$g_k(v=v^{})=g_k(u=u^{})=\delta _{k0},$$
(63)
which conform with the original initial conditions for $`g^l`$, Eq. (28).
### B Analytic solutions for $`k=0`$, $`1`$, and $`2`$
The solution to Eq. (58) for $`k=0`$, subject to the initial conditions, Eq. (63), is given by
$$g_0=J_0(z),$$
(64)
where $`J_n`$ are the Bessel functions of the first kind, of order $`n`$.
To solve for $`g_1`$, we first express the source $`S_1`$ explicitly as a function of $`yx`$ and $`z`$, using Eq. (64) and the relation $`\mathrm{\Delta }_r=yx+\mathrm{\Delta }_r^{}`$. We find
$$S_1=f_1J_0(z)(yx)f_1\mathrm{\Delta }_r^{}J_0(z).$$
(65)
Then, with the help of Table I, we find the solution for $`k=1`$ \[satisfying Eq. (63)\] to read
$$g_1=\frac{1}{4}f_1zJ_1(z)(\mathrm{\Delta }_r+\mathrm{\Delta }_r^{}).$$
(66)
We now use the above solutions for $`g_0`$ and $`g_1`$ to express $`S_2`$ as
$$S_2=\frac{1}{4}f_1^2zJ_1(z)\left[(yx)^2+3\mathrm{\Delta }_r^{}(yx)+2\mathrm{\Delta }_r^{}^2\right]f_3J_0(z)\left[(yx)^2+2\mathrm{\Delta }_r^{}(yx)+\mathrm{\Delta }_r^{}^2\right]f_2J_0(z).$$
(67)
With the help of Table I, we then construct the following solution for $`g_2`$, satisfying Eq. (63):
$$g_2=\frac{1}{6}zJ_1(z)\left[f_3(\mathrm{\Delta }_r^2+\mathrm{\Delta }_r\mathrm{\Delta }_r^{}+\mathrm{\Delta }_r^{}^2)+3f_2\right]+\frac{1}{96}z^2J_2(z)\left[3f_1^2(\mathrm{\Delta }_r+\mathrm{\Delta }_r^{})^28f_3\right]+\frac{1}{96}f_1^2z^3J_3(z).$$
(68)
### C $`G_{,\alpha }^{l\pm }`$ expanded in powers of $`1/L`$
We are now in position to write the three leading-order terms in the $`1/L`$ expansion of th egradient $`G_{,\alpha }^{l\pm }[x_0^\mu ;x_p^\mu ]`$. To that end we shall need, in view of Eqs. (41) and (42), to evaluate the functions $`g_k`$ derived above, along with their gradients $`g_{k,\alpha }`$, for $`x^\mu =x_0^\mu `$ and $`x_{}^{}{}_{}{}^{\mu }=x_p^\mu `$. For the calculation of $`g_{k,\alpha }`$ it is convenient to use the auxiliary relations
$$d[z^nJ_n(z)]/dz=z^nJ_{n1}(z)$$
(69)
\[for $`n=0`$ recall $`J_1(z)=J_1(z)`$\], along with
$$dz/dr=f_0L(yx)/z\text{and}dz/dt=ff_0L(y+x)/z,$$
(70)
where
$$f_0[r_0f^{1/2}(r_0)]^1.$$
(71)
With the help of these relations we derive from Eqs. (64), (66), and (68) expressions for $`g_{k,r}`$ and $`g_{k,t}`$ (where $`k=0,1,2`$). We then set $`x^\mu =x_0^\mu `$ and $`x_{}^{}{}_{}{}^{\mu }=x_p^\mu `$ in these expressions, and also in the expressions for the functions $`g_k`$ themselves (noticing the vanishing of $`\mathrm{\Delta }_r`$). All resulting expressions are then substituted in the formulas for $`G_{,r}^{l\pm }`$ and $`G_{,t}^{l\pm }`$, Eqs. (41) and (42). In these equations we also make the substitution $`\delta (w_\pm w_\pm ^{})=2LV_{00}^{1/2}\delta (\widehat{w}_\pm )`$ where $`\widehat{w}_\pm `$ are ‘neutral’ variables defined by
$$\widehat{w}_+2x\text{and}\widehat{w}_{}2y.$$
(72)
Finally, collecting common powers of $`L`$ we obtain, for $`\alpha =r`$ or $`t`$, an expression of the form
$$G_{,\alpha }^{l\pm }[x_0^\mu ;x_p^\mu ]=\frac{P_l(\mathrm{cos}\chi )}{rr^{}}\left(\widehat{H}_\alpha ^{(0)\pm }L^2+\widehat{H}_\alpha ^{(1)}L+\widehat{H}_\alpha ^{(2)}+\mathrm{}\right),$$
(73)
where the various coefficients $`\widehat{H}_\alpha `$ are functions of only the ‘neutral’ spacetime-interval variablesWe hereafter use the symbols $`\mathrm{\Delta }_r^{}`$, $`z`$, and $`\widehat{w}_\pm `$ to represent the values of these variables for $`x^\mu =x_0^\mu `$ and $`x_{}^{}{}_{}{}^{\mu }=x_p^\mu `$. $`\mathrm{\Delta }_r^{}`$, $`z`$, $`\widehat{w}_\pm `$, and
$$\mathrm{\Delta }_t2V_{00}^{1/2}L(t_0t_p).$$
(74)
These coefficient functions are given by
$$\widehat{H}_r^{(0)\pm }=f_0\left[\mathrm{\Delta }_r^{}J_1(z)/zJ_0(z)\delta (\widehat{w}_\pm )\right],$$
(76)
$$\widehat{H}_r^{(1)}=\frac{1}{4}f_0f_1\left[zJ_1(z)+\mathrm{\Delta }_r^{}^2J_0(z)\right]J_0(z)/r_0,$$
(77)
$$\widehat{H}_r^{(2)}=\frac{1}{96}f_0\mathrm{\Delta }_r^{}\left[7f_1^2z^2J_2(z)+3zJ_1(z)(f_1^2\mathrm{\Delta }_r^{}^28f_3)16J_0(z)(f_3\mathrm{\Delta }_r^{}^2+3f_2)\right]+\frac{1}{4}f_1\mathrm{\Delta }_r^{}zJ_1(z)/r_0,$$
(78)
and
$$\widehat{H}_t^{(0)\pm }=ff_0\left[\mathrm{\Delta }_tJ_1(z)/z+J_0(z)\delta (\widehat{w}_\pm )\right],$$
(80)
$$\widehat{H}_t^{(1)}=\frac{1}{4}ff_0f_1\mathrm{\Delta }_r^{}\mathrm{\Delta }_tJ_0(z),$$
(81)
$$\widehat{H}_t^{(2)}=\frac{1}{96}ff_0\mathrm{\Delta }_t\left[f_1^2z^2J_2(z)+zJ_1(z)(3f_1^2\mathrm{\Delta }_r^{}^28f_3)16J_0(z)(f_3\mathrm{\Delta }_r^{}^2+3f_2)\right]$$
(82)
(where the function $`f`$ is to be evaluated at $`r=r_0`$). In the above expressions for $`\widehat{H}_r^{(1),(2)}`$ and $`\widehat{H}_t^{(1),(2)}`$ we have omitted terms of the form $`z^kJ_n(z)\delta (\widehat{w}_\pm )`$, with $`k`$ being a positive integer, as such terms would yield vanishing contributions to $`\delta F_\alpha ^{l(ϵ)\pm }`$ when integrated over $`\tau `$ in Eq. (40).
The quantities $`\delta F_\alpha ^{l(ϵ)\pm }`$ can now be constructed, in principle, for any given worldline, by inserting Eq. (73) into Eq. (40) and carrying out the integration over $`\tau `$. In practice, to integrate over $`\tau `$, one should proceed as follows: Recalling that $`\tau `$ is a small quantity (we have $`|\tau |ϵ`$), one first expands in powers of $`\tau `$ all $`\tau `$-dependent quantities in the integrand of Eq. (40). (At that point, the details of the specific trajectory under consideration enter the calculation in an explicit manner; specifically, the power expansion coefficients turn out to depend on the values of $`u^\alpha `$, $`\dot{u}^\alpha `$, and $`\ddot{u}^\alpha `$ at the force’s evaluation point.) Then, since we are interested in extracting the large $`l`$ (large $`L`$) behavior of $`\delta F_\alpha ^{l(ϵ)\pm }`$, we introduce the “neutral” dimensionless proper time variable, defined by
$$\lambda (L/r_0)\tau ,$$
(83)
and replace $`\tau `$ by $`r_0(\lambda /L)`$. The integrand then takes the form of a power series in $`1/L`$, with $`\lambda `$-dependent coefficients. Transforming finally from integration over $`\tau `$ to integration over $`\lambda `$, one obtains an expression for $`\delta F_\alpha ^{l(ϵ)\pm }`$ in the form of a power series in $`1/L`$, as desired. In the rest of this paper we carry out the above calculation in full detail (and derive $`h_\alpha ^l`$ and $`D_\alpha )`$ for the case of a purely radial trajectory.
## VI Form of the regularization function $`h^l`$: the case of radial motion
When considering radial trajectories (namely, ones along which $`d\theta =d\phi =0`$) one has $`P_l(\mathrm{cos}\chi )1`$. Consequently, the Green’s function given in Eq. (30) becomes $`\theta ,\phi `$-independent, resulting in the vanishing of both angular components of the self force (as should be expected, of course, by virtue of the background being spherically-symmetric). In the following we discuss the $`r`$ and $`t`$ components of the self-force.
To carry out the integration in Eq. (40) we first expand each $`\tau `$-dependent quantity in the integrand in powers of $`1/L`$, with $`\lambda `$ held fixed. The $`\tau `$-dependent quantities to be expanded are $`\mathrm{\Delta }_r^{}`$, $`\mathrm{\Delta }_t`$, $`1/r^{}`$, $`z`$, and the various Bessel functions appearing in Eqs. (74) and (78).
By expanding $`\mathrm{\Delta }_r^{}`$ in a Taylor series in $`\tau `$ about $`r=r_0`$, and transforming to the variable $`\lambda `$, we obtain the expansion
$`\mathrm{\Delta }_r^{}`$ $`=`$ $`\dot{\mathrm{\Delta }}_r^{}\tau +{\displaystyle \frac{1}{2}}\ddot{\mathrm{\Delta }}_r^{}\tau ^2+{\displaystyle \frac{1}{6}}\stackrel{\dot{}\dot{}\dot{}}{\mathrm{\Delta }}_r^{}\tau ^3+\mathrm{}=`$ (85)
$`f^{1/2}\left[\dot{r}_{}\lambda +{\displaystyle \frac{1}{2}}\ddot{r}_{}\lambda ^2(r_0/L){\displaystyle \frac{1}{6}}\stackrel{\dot{}\dot{}\dot{}}{r}_{}\lambda ^3(r_0/L)^2\right]+O(1/L)^3,`$
where
$`\dot{r}_{}`$ $`=`$ $`f^1\dot{r},`$ (86)
$`\ddot{r}_{}`$ $`=`$ $`f^2(f\ddot{r}f^{}\dot{r}^2),`$ (87)
$`\stackrel{\dot{}\dot{}\dot{}}{r}_{}`$ $`=`$ $`f^3\left[(2f^2f^{\prime \prime }f)\dot{r}^33f^{}f\dot{r}\ddot{r}+f^2\stackrel{\dot{}\dot{}\dot{}}{r}\right],`$ (88)
and where all quantities (except $`\lambda `$) are evaluated at $`r=r_0`$ ($`\tau =0`$). In a similar manner we obtain for $`\mathrm{\Delta }_t`$,
$$\mathrm{\Delta }_t=f^{1/2}\left[\dot{t}\lambda \frac{1}{2}\ddot{t}\lambda ^2(r_0/L)+\frac{1}{6}\stackrel{\dot{}\dot{}\dot{}}{t}\lambda ^3(r_0/L)^2\right]+O(1/L)^3,$$
(89)
and for $`1/r^{}`$,
$$\frac{1}{r^{}}=\frac{1}{r_0}[1+\dot{r}(\lambda /L)+\frac{1}{2}(2\dot{r}^2r_0\ddot{r})(\lambda /L)^2)]+O(1/L)^3.$$
(90)
Next, recalling $`z=(L/r_0)s`$, we obtain
$$z=\dot{s}\lambda +\frac{1}{2}\ddot{s}\lambda ^2(r_0/L)\frac{1}{6}\stackrel{\dot{}\dot{}\dot{}}{s}\lambda ^3(r_0/L)^2+O(1/L)^3.$$
(91)
To calculate the $`\tau `$-derivatives of $`s`$ (which are understood here to be evaluated at $`\tau =0`$), we make use of the normalization relation $`\dot{v}\dot{u}=1/f`$, and of the relations derived from it by successively differentiating its both sides with respect to $`\tau `$: $`\ddot{v}\dot{u}+\dot{v}\ddot{u}=(1/f)^{}\dot{r}`$, and $`\stackrel{\dot{}\dot{}\dot{}}{v}\dot{u}+2\ddot{v}\ddot{u}+\dot{v}\stackrel{\dot{}\dot{}\dot{}}{u}=(1/f)^{\prime \prime }\dot{r}^2+(1/f)^{}\ddot{r}`$. Using these relations we find
$`\dot{s}(\tau =0)`$ $`=`$ $`1,`$ (92)
$`\ddot{s}(\tau =0)`$ $`=`$ $`{\displaystyle \frac{1}{2}}(f^{}/f)\dot{r},`$ (93)
$`\stackrel{\dot{}\dot{}\dot{}}{s}(\tau =0)`$ $`=`$ $`{\displaystyle \frac{1}{16f^2}}\left[\left(8f^{\prime \prime }f13f_{}^{}{}_{}{}^{2}\right)\dot{r}^2+8f^{}f\ddot{r}\right]+{\displaystyle \frac{1}{4}}f\ddot{v}\ddot{u}`$ (94)
(note that whereas $`\tau `$ is non-positive throughout the integration domain, $`z`$ and $`s`$ are, by definition, non-negative).
Finally, we need to similarly expand the various Bessel functions appearing in the integrand of Eq. (40). Using Eqs. (91) and (92) we find for any $`n0`$,
$`J_n(z)=J_n(\lambda )+{\displaystyle \frac{1}{2}}(r_0/L)\ddot{s}\lambda ^2J_n^{}(\lambda )+(r_0/L)^2\left({\displaystyle \frac{1}{8}}\ddot{s}^2\lambda ^4J_n^{\prime \prime }(\lambda ){\displaystyle \frac{1}{6}}\stackrel{\dot{}\dot{}\dot{}}{s}\lambda ^3J_n^{}(\lambda )\right)+O(1/L^3),`$ (95)
where a prime denotes $`d/d\lambda `$. Using this general form together with Eq. (69), we obtain the following expansions, needed for our analysis:
$`J_0(z)`$ $`=`$ $`J_0(\lambda ){\displaystyle \frac{1}{2}}(r_0/L)\ddot{s}\lambda ^2J_1(\lambda )+O(1/L^2)`$ (96)
$`J_1(z)`$ $`=`$ $`J_1(\lambda )+{\displaystyle \frac{1}{2}}(r_0/L)\ddot{s}\left(\lambda ^2J_0(\lambda )\lambda J_1(\lambda )\right)`$ (98)
$`+(r_0/L)^2\left[{\displaystyle \frac{1}{8}}\ddot{s}^2\left(\lambda ^3J_2(\lambda )\lambda ^4J_1(\lambda )\right){\displaystyle \frac{1}{6}}\stackrel{\dot{}\dot{}\dot{}}{s}\left(\lambda ^3J_0(\lambda )\lambda ^2J_1(\lambda )\right)\right]+O(1/L^3).`$
We now substitute the above expansions for $`\mathrm{\Delta }_r^{}`$, $`\mathrm{\Delta }_t`$, $`z`$, and the Bessel functions in Eqs. (74)–(78). We also substitute for the delta functions in Eqs. (76) and (80)
$$\delta (\widehat{w}_\pm )=\frac{\delta (\lambda )}{\left|d\widehat{w}_\pm /d\lambda \right|}=\frac{\delta (\lambda )}{f^{1/2}\dot{w}_\pm }=f^{1/2}\dot{w}_{}\delta (\lambda ),$$
(99)
where the last equality is due to the normalization of the four-velocity. We thereby obtain expressions for the various functions $`\widehat{H}_\alpha ^{(n=0,1,2)}`$, each expanded in powers of $`1/L`$ up to order $`O(L^{n2})`$ (with $`\lambda `$-dependent coefficients). Substitution of these expressions \[and of the expansion for $`1/r^{}`$, Eq. (90)\] into Eq. (73) finally yields the desired expression for the Green’s function’s gradient, as a power series in $`1/L`$ (with $`\lambda `$ held fixed). We find (for $`\alpha =r,t`$)
$$G_{,\alpha }^{l\pm }=H_\alpha ^{(0)\pm }L^2+H_\alpha ^{(1)}L+H_\alpha ^{(2)}+O(1/L),$$
(100)
where the various coefficients $`H_\alpha ^{(n)}`$ are functions of $`\lambda `$ along the worldline, given by
$$H_r^{(0)\pm }=\frac{1}{r_0^3}\left[\dot{r}_{}J_1(\lambda )\dot{w}_{}J_0(\lambda )\delta (\lambda )\right],$$
(101)
$`H_r^{(1)}={\displaystyle \frac{1}{4r_0^3}}[\lambda J_1(\lambda )(f_1/f^{1/2}4\dot{r}\dot{r}_{}+4r_0\ddot{s}\dot{r}_{}+2r_0\ddot{r}_{})`$ (102)
$`+\lambda ^2J_0(\lambda )(f^{1/2}f_1\dot{r}_{}^22r_0\ddot{s}\dot{r}_{})+4J_0(\lambda )],`$ (103)
$`H_r^{(2)}`$ $`=`$ $`{\displaystyle \frac{1}{96r_0^3}}[48\lambda J_0(\lambda )(2\dot{r}f_2\dot{r}_{})`$ (110)
$`+4\lambda ^3J_0(\lambda )(3f_1r_0\ddot{s}/f^{1/2}+6r_0^2\ddot{s}^2\dot{r}_{}+4r_0^2\stackrel{\dot{}\dot{}\dot{}}{s}\dot{r}_{}4ff_3\dot{r}_{}^3+6r_0^2\ddot{s}\ddot{r}_{}6f^{1/2}f_1r_0\dot{r}_{}\ddot{r}_{}`$
$`12\dot{r}r_0\ddot{s}\dot{r}_{}+6f^{1/2}f_1\dot{r}\dot{r}_{}^2)`$
$`+8\lambda ^2J_1(\lambda )(3f_1\dot{r}/f^{1/2}6r_0\ddot{s}3f_3\dot{r}_{}12\dot{r}^2\dot{r}_{}+6r_0\ddot{r}\dot{r}_{}+12r_0\dot{r}\ddot{s}\dot{r}_{}`$
$`6r_0^2\ddot{s}^2\dot{r}_{}4r_0^2\stackrel{\dot{}\dot{}\dot{}}{s}\dot{r}_{}+6r_0\dot{r}\ddot{r}_{}6r_0^2\ddot{s}\ddot{r}_{}2r_0^2\stackrel{\dot{}\dot{}\dot{}}{r}_{}+3f^{1/2}f_1\dot{r}_{})`$
$`+3\lambda ^4J_1(\lambda )\left(4r_0^2\ddot{s}^2\dot{r}_{}+ff_1^2\dot{r}_{}^34f^{1/2}f_1r_0\ddot{s}\dot{r}_{}^2\right)`$
$`+\lambda ^3J_2(\lambda )(7f_1^2\dot{r}_{}12r_0^2\ddot{s}^2\dot{r}_{})],`$
and
$$H_t^{(0)\pm }=\frac{f}{r_0^3}\left[J_0(\lambda )\dot{w}_{}\delta (\lambda )J_1(\lambda )\dot{t}\right],$$
(111)
$$H_t^{(1)}=\frac{f}{4r_0^3}\left\{\lambda J_1(\lambda )\left[4\dot{r}\dot{t}+2r_0\left(2\ddot{s}\dot{t}+\ddot{t}\right)\right]+\lambda ^2J_0(\lambda )\left(2r_0\ddot{s}\dot{t}+f^{1/2}f_1\dot{r}_{}\dot{t}\right)\right\},$$
(112)
$`H_t^{(2)}`$ $`=`$ $`{\displaystyle \frac{f}{96r_0^3}}\{48\lambda J_0(\lambda )f_2\dot{t}`$ (117)
$`4\lambda ^3J_0(\lambda )\left[r_0^2\left(6\ddot{s}^2\dot{t}+4\stackrel{\dot{}\dot{}\dot{}}{s}\dot{t}+6\ddot{s}\ddot{t}\right)3r_0\left(4\dot{r}\ddot{s}\dot{t}+f^{1/2}f_1\ddot{t}\dot{r}_{}+f^{1/2}f_1\dot{t}\ddot{r}_{}\right)+6f^{1/2}f_1\dot{r}\dot{t}\dot{r}_{}4ff_3\dot{t}\dot{r}_{}^2\right]`$
$`+8\lambda ^2J_1(\lambda )\left(f_3\dot{t}+12\dot{r}^2\dot{t}6r_0\ddot{r}\dot{t}12r_0\dot{r}\ddot{s}\dot{t}+6r_0^2\ddot{s}^2\dot{t}+4r_0^2\stackrel{\dot{}\dot{}\dot{}}{s}\dot{t}6r_0\dot{r}\ddot{t}+6r_0^2\ddot{s}\ddot{t}+2r_0^2\stackrel{\dot{}\dot{}\dot{}}{t}\right)`$
$`+3\lambda ^4J_1(\lambda )\left(4r_0^2\ddot{s}^2\dot{t}+4f^{1/2}f_1r_0\ddot{s}\dot{t}\dot{r}_{}ff_1^2\dot{t}\dot{r}_{}^2\right)`$
$`+\lambda ^3J_2(\lambda )(f_1^2\dot{t}+12r_0^2\ddot{s}^2\dot{t})\}.`$
Changing the integration variable in Eq. (40) from $`\tau `$ to $`\lambda `$, we now have
$$\delta F_\alpha ^{l(ϵ)\pm }=q^2r_0_0^{Lϵ/r_0}\left[LH_\alpha ^{(0)\pm }+H_\alpha ^{(1)}+H_\alpha ^{(2)}/L+O(L^2)\right]𝑑\lambda .$$
(118)
The desired regularization function $`h_\alpha ^l`$ is to be constructed such as to extract the large $`l`$ singular behavior of $`\delta F_\alpha ^{l(ϵ)\pm }`$ (while maintaining the simplest form possible). By virtue of Eq. (118) we take this function to be
$$h_\alpha ^{l\pm }=LA_\alpha ^\pm +B_\alpha +C_\alpha /L,$$
(119)
where
$`A_\alpha ^\pm `$ $``$ $`\stackrel{~}{lim}_l\mathrm{}\left(L^1\delta F_\alpha ^{l(ϵ)\pm }\right)=q^2r_0{\displaystyle _0^\stackrel{~}{\mathrm{}}}H_\alpha ^{(0)\pm }(\lambda )𝑑\lambda ,`$ (121)
$`B_\alpha `$ $``$ $`\stackrel{~}{lim}_l\mathrm{}\left(\delta F_\alpha ^{l(ϵ)\pm }LA_\alpha ^\pm \right)=q^2r_0{\displaystyle _0^\stackrel{~}{\mathrm{}}}H_\alpha ^{(1)}(\lambda )𝑑\lambda ,`$ (122)
$`C_\alpha `$ $``$ $`\stackrel{~}{lim}_l\mathrm{}L\left(\delta F_\alpha ^{l(ϵ)\pm }LA_\alpha ^\pm B_\alpha \right)=q^2r_0{\displaystyle _0^\stackrel{~}{\mathrm{}}}H_\alpha ^{(2)}(\lambda )𝑑\lambda ,`$ (123)
with $`^\stackrel{~}{\mathrm{}}()𝑑\lambda `$ standing for $`\stackrel{~}{lim}_x\mathrm{}^x()𝑑\lambda `$.<sup>\**</sup><sup>\**</sup>\**Here we extend the definition of the tilde-limit, given in App. A, from discrete functions with index $`l`$ to continuous functions of $`\lambda `$. In an analogous manner, the tilde-limit of a function $`f(\lambda )`$ as $`\lambda \mathrm{}`$ would be defined through the subtraction of a finite sum of functions of the form $`B^{(j)}(\lambda )=a_j\lambda ^{b_j}\mathrm{cos}(\alpha _j\lambda +\beta _j)`$ (with $`\alpha _j0`$ for all $`j`$). It is simple to verify that the tilde-limit of a function, when existing, is single-valued. That the second equality in each of Eqs. (122) and (123) is valid, and that the above choice of function $`h_\alpha ^{l\pm }`$ indeed satisfies the requirement that the tilde-sum $`_{l=0}^\stackrel{~}{\mathrm{}}\left(\delta F_\alpha ^{l(ϵ)\pm }h_\alpha ^{l\pm }\right)`$ would converge, will be shown in the next section, where we explicitly calculate the parameters $`A_\alpha `$, $`B_\alpha `$, and $`C_\alpha `$. The reason for using the tilde-limit instead of the standard limit in the definitions of these parameters will also become clear then. As to the parameter $`D_\alpha `$, substituting now Eqs. (118) and (119) into Eq. (38), one obtains<sup>††</sup><sup>††</sup>†† We assume here that the contribution associated with the $`O(L^2)`$ term in Eq. (118) vanishes upon taking the limit $`ϵ0`$, as this contribution is of order $`O(ϵ)`$ (this becomes clear from the calculation of $`D_\alpha `$ in the next section). However, a problem may occur if this term fails to yield a finite contribution when integrated over $`\lambda `$. Here we do not further investigate the behavior of the $`O(L^2)`$ term, and just assume that the above potential problem is not realized.
$$D_\alpha =q^2r_0\underset{ϵ0}{lim}\underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}_{Lϵ/r_0}^\stackrel{~}{\mathrm{}}\left[LH_\alpha ^{(0)\pm }+H_\alpha ^{(1)}+H_\alpha ^{(2)}/L\right]𝑑\lambda .$$
(124)
In conclusion, we find the tail part of the self force to be given by
$$F_\alpha ^{(\mathrm{tail})}=\underset{l=0}{\overset{\mathrm{}}{}}\left(F_\alpha ^{l\pm }A_\alpha ^\pm LB_\alpha C_\alpha /L\right)D_\alpha ,$$
(125)
where, from the above construction of $`h_\alpha ^l`$, it follows that the sum over $`l`$ converges at least as $`1/l`$. The implementation of our regularization scheme thus amounts to analytically determining the regularization parameters $`A_\alpha ^\pm `$, $`B_\alpha `$, $`C_\alpha `$, and $`D_\alpha `$, using Eqs. (119) and (124). For the calculation of the tail term, one may use either $`F_\alpha ^{l+}`$ (with $`A_\alpha ^+`$) or $`F_\alpha ^l`$ (with $`A_\alpha ^{}`$). Of course, one may also use any combination of these two one-sided quantities (e.g., their average). It should be emphasized here that the final result of the calculation, namely the tail term $`F_\alpha ^{(\mathrm{tail})}`$ (having a well defined value at the evaluation point), should be the same regardless of whether it is derived from one of the one-sided limits, or from the other, or, say, from their average.
We finally point out that, although Eq. (125) has been developed here for radial motion, an expression of this form is also valid for any other trajectory . The details of the specific trajectory under consideration would only affect the values of the various regularization parameters.
## VII Derivation of the regularization parameters for radial motion
To carry out the calculation of the regularization parameters in this section, we shall need the following integrals, the derivation of which will be described in Appendix B. For $`k,n`$ we have
$$_0^\stackrel{~}{\mathrm{}}\lambda ^kJ_n(\lambda )𝑑\lambda =\{\begin{array}{cc}(n+k1)!!/(nk1)!!,\hfill & 0kn,\hfill \\ (1)^{(kn)/2}(k+n1)!!(kn1)!!,\hfill & \text{even }kn>0,\hfill \\ 0,\hfill & \text{odd }kn>0\hfill \end{array}$$
(126)
\[in applying this formula for $`k=n`$, recall $`(1)!!=1`$\]. If the difference $`kn`$ is a positive odd integer, then we also have
$$_0^\stackrel{~}{\mathrm{}}\lambda ^kJ_n(\lambda )\mathrm{ln}\lambda d\lambda =(1)^{(kn+1)/2}(k+n1)!!(kn1)!!.$$
(127)
### A Derivation of $`A_\alpha ^\pm `$
Substituting Eqs. (101) and (111) into Eq. (121) and carrying out the tilde-integration \[with the help of Eq. (126)\] we find, recalling $`J_0(0)=1`$ and $`J_0(\mathrm{})=0`$,
$`A_r^\pm `$ $`=`$ $`{\displaystyle \frac{q^2}{r_0^2}}\left(\dot{r}_{}\dot{w}_{}\right)={\displaystyle \frac{q^2}{r_0^2}}\dot{t},`$ (128)
$`A_t^\pm `$ $`=`$ $`f{\displaystyle \frac{q^2}{r_0^2}}\left(\dot{t}+\dot{w}_{}\right)=\pm {\displaystyle \frac{q^2}{r_0^2}}\dot{r}.`$ (129)
Note that the two one-sided values of $`A_\alpha `$ are, in general, not the same. Consequently, as argued above, the function $`h_\alpha ^l`$ (and also $`\delta F_\alpha ^{l(ϵ)}`$) exhibits two different one-sided values. We note that the averaged value of the parameter $`A_\alpha `$, to be denoted by $`\overline{A}_\alpha `$, is found to vanish:
$$\overline{A}_\alpha \frac{1}{2}(A_\alpha ^++A_\alpha ^{})=0.$$
(130)
This vanishing of $`\overline{A}_\alpha `$ seems to occur for all trajectories of a scalar particle, not only the radial ones considered here .
### B Derivation of $`B_\alpha `$
By the definition of $`B_\alpha `$ in Eq. (122) we have, after substituting for $`A_\alpha ^\pm `$ from Eq. (121),
$$B_\alpha =q^2r_0\stackrel{~}{lim}_l\mathrm{}\left[L_{Lϵ/r_0}^\stackrel{~}{\mathrm{}}H_\alpha ^{(0)\pm }(\lambda )𝑑\lambda \right]+q^2r_0_0^\stackrel{~}{\mathrm{}}H_\alpha ^{(1)}(\lambda )𝑑\lambda .$$
(131)
The first term here cancels out upon taking the tilde-limit $`l\mathrm{}`$: At large $`l`$ (and fixed $`ϵ`$), each of the components $`H_\alpha ^{(0)\pm }`$ behaves as $`l^{1/2}`$ times oscillations with respect to $`l`$ \[these components are linear combinations of Bessel functions, the asymptotic form of which is described in Eq. (B6)\]. To leading order in $`1/l`$, this is also the form of the integral over $`H_\alpha ^{(0)\pm }`$ in Eq. (131) (which is carried out over asymptotically large values of $`\lambda `$). Hence, the expression in the squared brackets is found to diverge as $`l^{1/2}`$ times oscillations. When taking the tilde-limit, this divergent piece is removed, with the remaining part dying off at large $`l`$ as $`l^{1/2}`$ (times oscillations). Therefore, no contribution arises from the first term in Eq. (131), and the second equality of Eq. (122) is shown to be valid.
To calculate the parameter $`B_\alpha `$, we now substitute for $`H_\alpha ^{(1)}`$ from Eqs. (102) and (112) (for the $`r`$ and $`t`$ components, respectively). The calculation involves tilde-integrating over terms of the form $`\lambda J_1(\lambda )`$, $`\lambda ^2J_0(\lambda )`$, and $`J_0(\lambda )`$. Reading the values of the these integrals from Eq. (126), and substituting for $`f_1`$, $`\dot{r}_{}`$, $`\ddot{r}_{}`$, and $`\ddot{s}`$ \[using Eqs. (49), (86), and (93)\], we obtain
$$B_r=\frac{q^2}{2r_0^2}f^1\left(f+r_0f^{}/2\dot{r}^2+r_0\ddot{r}\right)$$
(132)
and
$$B_t=\frac{q^2}{2r_0^2}\left[fr_0\ddot{t}+\dot{t}\dot{r}(r_0f^{}f)\right].$$
(133)
Recalling $`a^r=\ddot{r}+f^{}/2`$ and $`a^t=\ddot{t}+f^{}\dot{r}_{}\dot{t}`$ in the case of radial motion considered here, we may express this result in a more compact form as
$$B_\alpha =\frac{q^2}{2r_0^2}\left(\delta _\alpha ^r+r_0a_\alpha \dot{r}u_\alpha \right).$$
(134)
### C Derivation of $`C_\alpha `$
By its definition in Eq. (123), we have for the parameter $`C_\alpha `$, after substituting for $`A_\alpha `$ and $`B_\alpha `$,
$$C_\alpha =q^2r_0\stackrel{~}{lim}_l\mathrm{}\left[_{Lϵ/r_0}^\stackrel{~}{\mathrm{}}\left(LH_\alpha ^{(0)}(\lambda )+H_\alpha ^{(1)}(\lambda )\right)𝑑\lambda \right]+q^2r_0_0^\stackrel{~}{\mathrm{}}H_\alpha ^{(2)}(\lambda )𝑑\lambda .$$
(135)
Again, there is a residual part left from the calculation of $`A_\alpha `$ and $`B_\alpha `$, which involves integration over asymptotically large values of $`\lambda `$. This part can again be shown to vanish as the tilde-limit $`l\mathrm{}`$ is taken, resulting in that only the second integral in Eq. (135) survives.
To calculate $`C_\alpha `$, we thus use the second equality of Eq. (123), in which we substitute for $`H_\alpha ^{(2)}`$ from Eqs. (110) and (117) (for the $`r`$ and $`t`$ components, respectively). One then has to evaluates the tilde-integral of a sum of various terms of the form $`\lambda ^kJ_n(\lambda )`$, all of which have $`k>n`$ and odd $`kn`$. According to Eq. (126), all such integrals vanish. Thus, we find
$`C_\alpha =0.`$ (136)
The vanishing of the parameter $`C_\alpha `$ seems to be a universal feature of our scheme, regardless of the specific trajectory under consideration . As we also find below, this vanishing constitutes a necessary condition for the self-consistency of the whole regularization scheme.
### D Derivation of $`D_\alpha `$
To calculate the parameter $`D_\alpha `$ we write Eq. (124) in the form
$$D_\alpha =D_\alpha ^{(0)}+D_\alpha ^{(1)}+D_\alpha ^{(2)},$$
(137)
where
$`D_\alpha ^{(n)}q^2r_0\underset{ϵ0}{lim}{\displaystyle \underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}}{\displaystyle _{Lϵ/r_0}^\stackrel{~}{\mathrm{}}}L^{1n}H_\alpha ^{(n)\pm }(\lambda )𝑑\lambda .`$ (138)
In calculating the above three pieces of $`D_\alpha `$, we shall transform from summation over $`l`$ to integration over a continuous variable. For this transformation we will make use of the relation
$$\underset{l=0}{\overset{\mathrm{}}{}}(ϵ/r_0)K(Lϵ/r_0)=_0^{\mathrm{}}\left[K(x)\frac{1}{24}(ϵ/r_0)^2K^{\prime \prime }(x)\right]𝑑x+O(ϵ^3),$$
(139)
where $`K(x)`$ is any (sufficiently regular) integrable real function, and $`x`$ is an integration variable. Here, the $`O(ϵ^0)`$ term on the right-hand side (RHS) is the standard (“Riemann type”) integral, which is obtained, by definition, when the $`ϵ0`$ limit of the left-hand side is taken. We also indicated here the $`O(ϵ^2)`$ correction to the integral, which we shall have to take into account in the calculation below (it is straightforward to verify the form of this correction term using standard calculus). Obviously, Eq. (139) also holds for the tilde-sum $`_{l=0}^\stackrel{~}{\mathrm{}}`$, where on the RHS we use the tilde-integral $`_0^\stackrel{~}{\mathrm{}}`$.
Beginning with the calculation of $`D_\alpha ^{(0)}`$, we write Eq. (138) for $`n=0`$ as
$$D_\alpha ^{(0)}=q^2r_0\underset{ϵ0}{lim}\left[(r_0/ϵ)^2\underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}(ϵ/r_0)\left(\left(Lϵ/r_0\right)_{Lϵ/r_0}^\stackrel{~}{\mathrm{}}H_\alpha ^{(0)\pm }(\lambda )𝑑\lambda \right)\right].$$
(140)
Comparing the form of the sum in this expression to the left-hand side of Eq. (139), we find
$$D_\alpha ^{(0)}=q^2r_0\underset{ϵ0}{lim}\left[(r_0/ϵ)^2_0^\stackrel{~}{\mathrm{}}𝑑x\left(K_\alpha ^{(0)}(x)\frac{1}{24}(ϵ/r_0)^2[K_\alpha ^{(0)}(x)]^{\prime \prime }\right)\right],$$
(141)
where
$$K_\alpha ^{(0)}(x)x_x^\stackrel{~}{\mathrm{}}H_\alpha ^{(0)\pm }(\lambda )𝑑\lambda .$$
(142)
Note here how the contribution to $`D_\alpha ^{(0)}`$ due to the $`O(ϵ^3)`$ term appearing in Eq. (139) vanishes upon taking the limit $`ϵ0`$. We recall that each of the two components $`H_r^{(0)\pm }`$ and $`H_t^{(0)\pm }`$, given explicitly in Eqs. (101) and (111), contains two terms: one proportional to $`\delta (\lambda )`$ and the other to $`J_1(\lambda )`$. The $`\delta (\lambda )`$ term has no contribution to $`K_\alpha ^{(0)}`$, resulting in that both two one-sided values $`H_\alpha ^{(0)+}`$ and $`H_\alpha ^{(0)}`$ yield the same function $`K_\alpha ^{(0)}`$ (for that reason, no $`\pm `$ sign has been assigned to this quantity). There is an apparent danger of divergence coming from the $`O(ϵ^2)`$ term in Eq. (141). Such a divergence is avoided, however, as we have
$$_0^\stackrel{~}{\mathrm{}}K_\alpha ^{(0)}(x)𝑑x=\frac{u_\alpha }{r_0^3}_0^\stackrel{~}{\mathrm{}}𝑑xx_x^\stackrel{~}{\mathrm{}}J_1(\lambda )𝑑\lambda =\frac{u_\alpha }{2r_0^3}_0^\stackrel{~}{\mathrm{}}x^2J_1(x)𝑑x=0,$$
(143)
where in the second equality we integrated by parts with respect to $`x`$, and where the vanishing of the last integral is implied by Eq. (126). Thus, the $`O(ϵ^2)`$ term in Eq. (141) vanishes, and the $`ϵ0`$ limit in this equation turns out well defined. The remaining $`O(ϵ^0)`$ contribution reads
$$D_\alpha ^{(0)}=\frac{q^2r_0}{24}_0^\stackrel{~}{\mathrm{}}[K_\alpha ^{(0)}(x)]^{\prime \prime }𝑑x=\frac{q^2r_0}{24}_0^\stackrel{~}{\mathrm{}}\left\{H_\alpha ^{(0)\pm }(x)+\left[xH_\alpha ^{(0)\pm }(x)\right]^{}\right\}𝑑x=\frac{q^2r_0}{24}_0^\stackrel{~}{\mathrm{}}H_\alpha ^{(0)\pm }(x)𝑑x,$$
(144)
as the surface term vanishes. Substituting for $`H_\alpha ^{(0)\pm }(x)`$ and integrating using Eq. (126), we finally obtain
$$D_\alpha ^{(0)}=\frac{q^2u_\alpha }{24r_0^2}_0^\stackrel{~}{\mathrm{}}J_1(\lambda )𝑑\lambda =\frac{q^2u_\alpha }{24r_0^2}.$$
(145)
We next turn to calculate $`D_\alpha ^{(1)}`$. Writing Eq. (138) for $`n=1`$ in the form
$$D_\alpha ^{(1)}=q^2r_0\underset{ϵ0}{lim}\left[(r_0/ϵ)\underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}(ϵ/r_0)_{Lϵ/r_0}^\stackrel{~}{\mathrm{}}H_\alpha ^{(1)}(\lambda )𝑑\lambda \right]$$
(146)
and applying the summation formula (139), we obtain
$$D_\alpha ^{(1)}=q^2r_0\underset{ϵ0}{lim}\left[(r_0/ϵ)_0^\stackrel{~}{\mathrm{}}𝑑x_x^\stackrel{~}{\mathrm{}}H_\alpha ^{(1)}(\lambda )𝑑\lambda \right],$$
(147)
with all $`O(ϵ^2)`$ terms appearing in Eq. (139) vanishing in the limit $`ϵ0`$. By integrating the last expression by parts with respect to $`x`$ (noticing the vanishing of the surface term) one finds
$`D_\alpha ^{(1)}=q^2r_0\underset{ϵ0}{lim}\left[(r_0/ϵ){\displaystyle _0^\stackrel{~}{\mathrm{}}}xH_\alpha ^{(1)}(x)𝑑x\right].`$ (148)
Here, the integrand contains only terms of the form $`x^kJ_n(x)`$, with $`kn`$ being positive odd integers \[see Eqs. (102) and (112)\]. Hence, by virtue of Eq. (126), the integral vanishes, yielding
$$D_\alpha ^{(1)}=0.$$
(149)
Finally, from Eq. (138) with $`n=2`$ we obtain
$`D_\alpha ^{(2)}=q^2r_0\underset{ϵ0}{lim}\left[{\displaystyle \underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}}(ϵ/r_0)(Lϵ/r_0)^1{\displaystyle _{Lϵ/r_0}^\stackrel{~}{\mathrm{}}}H_\alpha ^{(2)}(\lambda )𝑑\lambda \right]=q^2r_0{\displaystyle _0^\stackrel{~}{\mathrm{}}}(dx/x){\displaystyle _x^\stackrel{~}{\mathrm{}}}H_\alpha ^{(2)}(\lambda )𝑑\lambda ,`$ (150)
which, after integrating by parts, becomes
$$D_\alpha ^{(2)}=q^2r_0\left[_0^\stackrel{~}{\mathrm{}}\mathrm{ln}xH_\alpha ^{(2)}(x)𝑑x\underset{x0}{lim}\left(\mathrm{ln}x_x^\stackrel{~}{\mathrm{}}H_\alpha ^{(2)}(\lambda )𝑑\lambda \right)\right].$$
(151)
We notice here that the second integral on the RHS is just $`C_\alpha +O(x)`$ \[up to a multiplicative constant; see Eq. (123)\]. The above-deduced vanishing of the parameter $`C_\alpha `$ guarantees the definiteness of the $`x0`$ limit in Eq. (151), and makes the second term on its RHS vanish. Note the way the vanishing of the parameter $`C_\alpha `$ appears as a necessary self-consistency condition in our scheme: had we got $`C_\alpha 0`$, the parameter $`D_\alpha `$ would have been indefinite, and the whole regularization scheme would have been rendered meaningless.
As the $`x0`$ limit in Eq. (151) vanishes, we are left with
$$D_\alpha ^{(2)}=q^2r_0_0^\stackrel{~}{\mathrm{}}\mathrm{ln}xH_\alpha ^{(2)}(x)𝑑x.$$
(152)
With the explicit form of $`H_\alpha ^{(2)}`$, given in Eqs. (110) and (117), the integrand in the last expression is found to consist of various terms of the form $`x^k\mathrm{ln}xJ_n(x)`$, with $`kn`$ being positive odd integers. The integrals $`_0^\stackrel{~}{\mathrm{}}`$ of such terms can be read from the formula (127), which we derive in Appendix B. Using this formula we obtain expressions for $`D_r^{(2)}`$ and $`D_t^{(2)}`$, which, after substituting for $`\dot{r}_{}`$, $`\ddot{r}_{}`$, $`\stackrel{\dot{}\dot{}\dot{}}{r}_{}`$, $`\ddot{s}`$, $`\stackrel{\dot{}\dot{}\dot{}}{s}`$, $`f_1`$, $`f_2`$, and $`f_3`$, read
$`D_r^{(2)}={\displaystyle \frac{1}{3}}q^2\left(f^1\stackrel{\dot{}\dot{}\dot{}}{r}+\dot{r}\ddot{v}\ddot{u}\right)+{\displaystyle \frac{q^2\dot{r}}{24f^2r_0^2}}\left[f(4f3)+8r_0f^{}(fr_0\ddot{r})+2r_0^2(3ff^{\prime \prime }f_{}^{}{}_{}{}^{2})\right],`$ (153)
and
$`D_t^{(2)}`$ $`=`$ $`{\displaystyle \frac{1}{3}}q^2\left(f\stackrel{\dot{}\dot{}\dot{}}{r}+f^2\dot{t}\ddot{v}\ddot{u}\right){\displaystyle \frac{1}{2}}q^2f^{}\dot{r}\ddot{t}`$ (155)
$`+{\displaystyle \frac{q^2\dot{t}}{24r_0^2}}\left[3f4f^2+4f^1(f^{})^2r_0^2\dot{r}^28f^{\prime \prime }r_0^2\dot{r}^24f^{}r_0^2\ddot{r}2ff^{\prime \prime }r_0^28ff^{}r_0\right].`$
We are now in position to write an expression for the “overall” parameter $`D_\alpha `$. We have $`D_\alpha =D_\alpha ^{(0)}+D_\alpha ^{(2)}`$, yielding
$`D_r={\displaystyle \frac{1}{3}}q^2\left(f^1\stackrel{\dot{}\dot{}\dot{}}{r}+\dot{r}\ddot{v}\ddot{u}\right)+{\displaystyle \frac{q^2\dot{r}}{12f^2r_0^2}}\left[2f(f1)+4r_0f^{}(fr_0\ddot{r})+r_0^2(3ff^{\prime \prime }f_{}^{}{}_{}{}^{2})\right],`$ (156)
and
$`D_t`$ $`=`$ $`{\displaystyle \frac{1}{3}}q^2\left(f\stackrel{\dot{}\dot{}\dot{}}{r}+f^2\dot{t}\ddot{v}\ddot{u}\right){\displaystyle \frac{1}{2}}q^2f^{}\dot{r}\ddot{t}`$ (158)
$`+{\displaystyle \frac{q^2\dot{t}}{12r_0^2}}\left[2f(1f)+2f1(f^{})^2r_0^2\dot{r}^24f^{\prime \prime }r_0^2\dot{r}^22f^{}r_0^2\ddot{r}ff^{\prime \prime }r_0^24ff^{}r_0\right].`$
In the case of radial motion, the four-acceleration’s components admit the explicit form $`a^r=\ddot{r}+\frac{1}{2}f^{}`$ and $`a^t=\ddot{t}+f^{}\dot{r}_{}\dot{t}`$. Recalling also that in the spacetime class considered here the Ricci scalar reads $`R=[f^{\prime \prime }+4f^{}/r+2(f1)/r^2]`$, one can show that the above two expressions for $`D_r`$ and $`D_t`$ can be put into the simple vectorial form
$$D_\alpha =\frac{1}{3}q^2\left(\dot{a}_\alpha a^2u_\alpha \right)\frac{1}{12}q^2Ru_\alpha .$$
(159)
Comparing now this result with Eqs. (10) and (11), and recalling that in the case considered here (that of radial motion on static spherically-symmetric background) the first two terms in the expression for $`F_\alpha ^{(\mathrm{Ricci})}`$ \[Eq. (11)\] cancel out, we arrive at the remarkable conclusion that $`D_\alpha `$ is exactly the “standard” local part of the self-force:
$$D_\alpha =F_\alpha ^{(\mathrm{ALD})}+F_\alpha ^{(\mathrm{Ricci})}.$$
(160)
## VIII Summary and concluding remarks
The total self force acting on the radially moving scalar particle is obtained by substituting $`F_\alpha ^{(\mathrm{tail})}`$ from Eq. (125) in Eq. (9). By virtue of Eq. (160), the contribution of $`D_\alpha `$ to the tail term is then found to exactly cancel out the local term in the expression for the total self force. This, in addition to the vanishing of the parameter $`C_\alpha `$, leads to the simple result
$$F_\alpha ^{(\mathrm{total})}=\underset{l=0}{\overset{\mathrm{}}{}}\left(F_\alpha ^{l\pm }A_\alpha ^\pm LB_\alpha \right)$$
(161)
(where, we recall, $`L=l+1/2`$). An even simpler form is obtained when calculating $`F_\alpha ^{(\mathrm{total})}`$ using the averaged value of the modes $`F_\alpha ^l`$, obtained by averaging over their two one-sided values. Then, by virtue of Eq. (130), we find
$$F_\alpha ^{(\mathrm{total})}=\underset{l=0}{\overset{\mathrm{}}{}}\left(\overline{F}_\alpha ^lB_\alpha \right),$$
(162)
where $`\overline{F}_\alpha ^l\frac{1}{2}(F_\alpha ^{l+}+F_\alpha ^l)`$. Recall that the parameter $`B_\alpha `$, given in Eq. (134), is just the asymptotic value of the averaged $`l`$-mode $`\overline{F}_\alpha ^l`$ at the limit $`l\mathrm{}`$; namely, the total self force is obtained by simply subtracting from each (two one-sided averaged) mode its large $`l`$ asymptotic value, and then summing over all modes.<sup>‡‡</sup><sup>‡‡</sup>‡‡ The simplicity of our main result, Eq. (162), may lead one to wonder whether there could be simple arguments leading directly to this result. Such arguments might perhaps rely on general properties of the Hadamard expansion. This should make an interesting subject for further investigation.
To summarize, in this paper we have developed a method for calculating the self force on a scalar particle in curved spacetime, through regularization of the multipole mode sum. The basic difficulty in applying the mode decomposition approach—the apparent divergence of the sum over modes—has been taken care of by the introduction of an appropriate regularization scheme, providing a practical prescription for calculating the self force. It should be emphasized that the proposed method does not involve any weak-field or slow-motion approximations, and thus allows effective calculations of the self force even for strong field orbits.
The basic expression for the tail part of the self force is given in Eq. (125). This expression was developed here for radial motion; however, the same general form applies for any trajectory , with the details of the orbit encoded only in the values of the regularization parameters (as well as, of course, in the form of the “bare” modes $`F_\alpha ^{l\pm }`$). To apply this general expression for a given trajectory requires knowledge of (i) the modes $`F_\alpha ^l`$, to be derived by supplementary (basically straightforward) numerical analysis, as done in Refs. ; and (ii) four regularization parameters for each spacetime component of the force. In this paper we have worked out the entire calculation of the regularization parameters for the case of radial motion. For any other trajectory, the derivation of these parameters can be carried out along the same lines, based on the explicit form of the Green’s function’s $`l`$-mode given in Eq. (30), with Eqs. (64), (66), and (68). To that end, one first obtains an expression for the gradient of the Green’s function’s $`l`$-mode, as in Eq. (73) \[supplemented with Eqs. (74) and (78)\]. One next expands this gradient in powers of proper time $`\tau `$ along the worldline about the force’s evaluation point, and re-expresses the resulting expansion as an expansion in powers of $`1/L`$, by holding $`\tau L`$ fixed—as in Eq. (100) \[supplemented with Eqs. (101)–(117)\]. One finally uses the values of the above expansion coefficients \[denoted in this paper by $`H_\alpha ^{(n)}(\tau L)`$\] to construct the regularization parameters through Eqs. (119) and (124).
For a scalar particle moving radially in a spacetime of the class considered in this paper, we found the total self force to be given by Eq. (162). This constitutes our main result for the radial motion case, together with the explicit values of the regularization parameters given in Eqs. (130), (134), (136), and (159). We have found that, in the radial motion case, the parameters $`\overline{A}_\alpha `$ and $`C_\alpha `$ both vanish \[the one-sided values of $`A_\alpha `$ do not vanish; they are given in Eq. (128)\], and the parameter $`D_\alpha `$ is just the standard local part of the self force. The vanishing of $`C_\alpha `$, shown here explicitly, appeared as a necessary condition for the definiteness of the whole scheme (had $`C_\alpha 0`$, the parameter $`D_\alpha `$ would have diverge; see the discussion in Sec. VII). This point serves to demonstrate the self-consistency of the regularization scheme.
A question arises, whether the above results (the vanishing of $`\overline{A}_\alpha `$ and $`C_\alpha `$ and the special value of $`D_\alpha `$) represent generic features of the regularization scheme, or rather are special to radial trajectories. Preliminary investigation suggests that, indeed, $`\overline{A}_\alpha `$ and $`C_\alpha `$ vanish for all trajectories, at least in the Schwarzschild case. As to the parameter $`D_\alpha `$, this was shown so far to obey Eq. (160) at least in one more important example, that of a circular orbit around a Schwarzschild black hole . It might be conjectured (and be subject to further investigation) that Eq. (160) holds for any trajectory in any static spherically-symmetric background. In that case, the simple Eq. (162) for the total self force would be valid for all such trajectories.
Under the above conjecture, we find that regularization of the total self force requires knowledge of just one parameter, $`B_\alpha `$, representing the asymptotic value of the modes $`\overline{F}_\alpha ^l`$ as $`l\mathrm{}`$. The value of $`B_\alpha `$ for any specific radial trajectory on any given spherically-symmetric spacetime, can be read from Eq. (134) (valid regardless of the above conjecture). For example, in the special case of a static particle we find $`B_t^{(\mathrm{static})}=0`$ and
$$B_r^{(\mathrm{static})}=\frac{q^2}{2r_0^2}\left(1+\frac{r_0f^{}}{2f}\right).$$
(163)
For radial geodesic motion we find
$$B_t^{(\mathrm{geodesic})}=\frac{q^2}{2r_0^2}\dot{r}E,B_r^{(\mathrm{geodesic})}=\frac{q^2}{2r_0^2}\left(2E^2/f\right),$$
(164)
where $`Eu_t`$ is the energy parameter (which is a constant of motion in the absence of self force effect). For the value of $`B_r`$ in the case of uniform circular motion in Schwarzschild spacetime (the derivation of which will be presented elsewhere ) we refer the reader to Eq. (34) of Ref. .
The applicability of the regularization prescription described here was demonstrated recently in actual calculations of the self force for various scenarios. Burko first studied the cases of static and circular orbits in the Schwarzschild spacetime. For these stationary scenarios, the modes $`F_\alpha ^l`$ were obtained by summing over the Fourier-multipole modes $`F_\alpha ^{lm\omega }`$, first derived by solving the appropriate ordinary field equations in the frequency domain. Later, Barack and Burko analyzed the case of radial motion in Schwarzschild. In this case, which is no longer stationary, numerical evolution of the appropriate partial DE in the time domain was applied to directly infer the modes $`F_\alpha ^l`$. In each of these studies, the overall force acting on the scalar particle was finally deduced by summing over all modes, using the above regularization scheme. In each of the cases analyzed, the vanishing of $`C_\alpha `$ was demonstrated, and the analytically-derived expressions for $`A_\alpha `$ and $`B_\alpha `$ were verified. Aside from demonstrating the applicability of the regularization scheme and providing verification for the values of the regularization parameters, the above studies yielded valuable physical results, as mentioned in the Introduction.
Of course, the analysis of the scalar self force merely serves as a toy model for more realistic cases. Generalization of the regularization scheme to the electromagnetic self force seems possible, based on the existing formalism . Such a generalization is necessary, for example, for resolving the interesting question re-raised recently by Hubeny , whether a nearly extreme electrically-charged black hole might be overcharged (and its event horizon by destroyed) by throwing in a charged particle: as pointed out by Hubeny, knowledge of the exact radiation reaction effect is crucial for obtaining a definite answer. More difficult to accomplish would be the important generalization of the scheme to the gravitational self force acting on a mass particle.
Finally, it should be mentioned that a closely related approach was recently applied by Lousto for studying the gravitational self force on a mass particle in Schwarzschild spacetime. This approach is also based on the multipole expansion, yet it employs a different regularization method for the mode sum (it is argued that the correct self force can be deduced by applying the zeta-function regularization technique). For the geodesic motion case studied by Lousto, this approach leads to an expression analogous to Eq. (162).
## acknowledgements
I wish to thank Amos Ori for suggesting the basic idea for the regularization scheme, and for his assistance in developing it. I would also like to thank Lior Burko for discussions and for reading the manuscript.
## A Multipole expansion of the Green’s function using “tilde-summation”
DeWitt and Brehme wrote a general expression for the scalar Green’s function in curved spacetime \[see Eq. (2.21) of Ref. \], of the form
$$G(x^\mu ;x_{}^{}{}_{}{}^{\mu })=a(x^\mu ;x_{}^{}{}_{}{}^{\mu })\delta (\sigma )+b(x^\mu ;x_{}^{}{}_{}{}^{\mu })\mathrm{\Theta }(\sigma ).$$
(A1)
Here, $`\sigma `$ is plus or minus half the squared geodesic distance between the source point $`x^\mu `$ and the evaluation point $`x_{}^{}{}_{}{}^{\mu }`$, according to whether the geodesic connecting the points (along which the invariant distance is measured) is timelike ($`\sigma >0`$) or else ($`\sigma 0`$); $`a`$ is a certain function having a well defined value at $`\sigma =0`$; and $`b`$ is a function which may be written as a Taylor expansion in $`\sigma `$ about $`\sigma =0`$. \[This expansion was shown by Hadamard (see pp. 96–98 in Ref. ) to converge uniformly at least inside the region where $`\sigma `$ is single valued.\] While the first term in Eq. (A1) is associated with the familiar delta function exhibited already in flat spacetime, the second term represents a curvature-induced tail, which “fills” the light cone (defined by $`\sigma =0`$). Note that the Green’s function is strongly irregular along the light cone of the source point $`x_{}^{}{}_{}{}^{\mu }`$.
Now, the question to consider is whether the above Green’s function may be expanded in terms of the standard spherical harmonic functions $`Y^{lm}(\theta ,\phi )`$ on a sphere of constant $`r,t`$. Standard theorem (see, e.g., Ref. , p. 513) states that a sufficient condition for (absolute and uniform) convergence of the spherical harmonic expansion is the expanded function being $`C^2`$ on the sphere. This condition is not satisfied here, as the Green’s function diverges along the curve generated by the intersection of the future light cone of $`x_{}^{}{}_{}{}^{\mu }`$ and the sphere of constant $`r,t`$. Therefore, it is not guaranteed, in advance, that such an expansion could be naively applied. Indeed, already in flat spacetime the attempt to apply the multipole expansion to the Green’s function turns out to yield a divergent sum—see Eq. (14) and the discussion proceeding it in Sec. III.
Let us introduce the “modified” Green’s function (which is not a“Green’s function” for the scalar field anymore), defined by
$$G_{\mathrm{mod}}=G\delta G,$$
(A2)
where
$$\delta Ga_0\delta (\sigma )b_0\mathrm{\Theta }(\sigma ),$$
(A3)
with $`a_0a(\sigma =0)`$ and $`b_0b(\sigma =0)`$. The function $`G_{\mathrm{mod}}`$ has two essential features: (i) its support inside the light cone is identical to that of the Green’s function $`G`$ (as $`\delta G`$ has no support there), and (ii) it is continuous throughout any sphere of constant $`r,t`$. The first feature implies that we can use $`G_{\mathrm{mod}}`$ instead of $`G`$ in calculating the self force: We may re-write Eq. (13) as
$$F_\alpha ^{(ϵ)}q^2_{\mathrm{}}^ϵ\left\{G_{\mathrm{mod}}[x_0^\mu ;x_p^\mu (\tau )]\right\}_{,\alpha }𝑑\tau ,$$
(A4)
as the $`\delta G`$ term in $`G_{\mathrm{mod}}`$ contributes nothing to the integral along the particle’s worldline.
The second of the above features of $`G_{\mathrm{mod}}`$, its continuity, may imply that the multipole expansion could now be applied to it. Strictly speaking, $`G_{\mathrm{mod}}`$ does not satisfy the above sufficient condition for absolute and uniform convergence of the mode sum (i.e., being $`C^2`$ on the sphere); yet, we shall assume here that $`G_{\mathrm{mod}}`$, being continuous, is already regular enough to admit a convergent mode sum. The results of our analysis turn out to be consistent with this assumption, as the mode-sum of $`G`$ considered (based on the function $`G_{\mathrm{mod}}`$ through the use of “tilde summation”—see below) is found to be (absolutely) convergent. The validity of this assumption can also be demonstrated in the flat space case: the $`b_0`$ tail term vanishes in this case, and by expanding the $`\delta (\sigma )`$ term in Eq. (A3) in spherical harmonics one can easily verify that the $`l`$-mode of $`\delta G`$ is exactly the $`l`$-mode of $`G`$, given in Eq. (14). The $`l`$-mode of $`G_{\mathrm{mod}}`$ then vanishes, and the mode sum converges. Although trivial, this flat-space example may serve to demonstrate how the subtraction of $`\delta G`$ from $`G`$ already removes the divergent piece from the $`l`$-mode, making the mode sum converge.<sup>\**</sup><sup>\**</sup>\** One may similarly construct a more sophisticated function $`\delta G`$, designed to yield a $`C^2`$ modified function $`G_{\mathrm{mod}}`$ \[by canceling also the $`O(\sigma )`$ and $`O(\sigma ^2)`$ terms in the Taylor expansion of $`b`$\] being sufficiently regular to assure uniform and absolute convergence of the multipole expansion, by standard mathematical theorem. Such an improved construction will not be examined here. We thus expand $`G_{\mathrm{mod}}`$ as
$$G_{\mathrm{mod}}=\underset{l=0}{\overset{\mathrm{}}{}}G_{\mathrm{mod}}^l=\underset{l=0}{\overset{\mathrm{}}{}}\left(G^l\delta G^l\right),$$
(A5)
where $`G_{\mathrm{mod}}^l`$ and $`\delta G^l`$ are the spherical harmonic modes of $`G_{\mathrm{mod}}`$ and $`\delta G`$, respectively (obtained by summing over all azimuthal numbers $`m`$).
To proceed, let us now define the new operation $`\stackrel{~}{lim}_l\mathrm{}`$ (“tilde-limit”) as follows: Consider a series of numbers $`A_l`$ (with $`l=0,1,\mathrm{},\mathrm{}`$). Let $`B_l^{(j)}`$ be any expression of the form
$$B_l^{(j)}=a_jl^{b_j}\mathrm{cos}(\alpha _jl+\beta _j),$$
(A6)
where $`a_j`$, $`b_j`$, $`\alpha _j`$, and $`\beta _j`$ are some $`l`$-independent real numbers, with $`\alpha _j0`$ for all $`j`$. If there exists a finite number $`k`$ of expressions $`B_l^{(j)}`$ of this form (with $`j=1,2,\mathrm{},k`$), such that subtracting their sum from the original series $`A_l`$ would yield a well-defined finite limit as $`l\mathrm{}`$, then we define the “tilde-limit” $`\stackrel{~}{lim}_l\mathrm{}A_l`$ as in Eq. (16).
One may easily be convinced that the tilde-limit is single-valued (when existing). For, suppose that for a given series $`A_l`$ there were two different sets of quantities $`B_l^{(j)}`$, one (denoted by $`\overline{B}_l^{(\overline{j})}`$) yielding $`\stackrel{~}{lim}_l\mathrm{}A_l=\overline{c}`$, and the other (denoted by $`\widehat{B}_l^{(\widehat{j})}`$) yielding $`\stackrel{~}{lim}_l\mathrm{}A_l=\widehat{c}\overline{c}`$. Then, for the difference between the two limits one would have found $`lim_l\mathrm{}\left[_{\widehat{j}}\widehat{B}_l^{(\widehat{j})}_{\overline{j}}\overline{B}_l^{(\overline{j})}\right]=\widehat{c}\overline{c}0`$. This, however, is impossible, as the (standard) limit $`l\mathrm{}`$ of any quantity of the type $`B_l^{(j)}`$ is either diverging or zero, and so is the limit of any finite sum of such quantities. Hence, we must have $`\widehat{c}=\overline{c}`$, and the tilde-limit is single valued. In particular, we find that if there exists a finite standard limit $`lim_l\mathrm{}A_l`$, then $`lim_l\mathrm{}A_l=\stackrel{~}{lim}_l\mathrm{}A_l`$.
We can now also define the “tilde-sum” of a series $`A_l`$, as in Eq. (17). Again, if the “tilde-sum” of a series exists, then it is unique. Also, if the standard sum $`_l\mathrm{}A_l`$ converges, then we may replace it with a “tilde-sum” operation. In particular, we may replace the convergent standard sum of Eq. (A5) with a tilde-summation:
$$G_{\mathrm{mod}}=\underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}\left(G^l\delta G^l\right).$$
(A7)
Below we show that
$$\underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}\delta G^l=0$$
(A8)
(for any evaluation point $`x_0^\mu `$ lying inside the future light cone of the source point $`x_p^\mu `$). As a consequence, the tilde-sum of $`G^l`$ is found to be finite and equal to $`G_{\mathrm{mod}}`$—as indicated in Eq. (18) of Sec. III.
Combining Eqs. (A4) and (18) we conclude that the self-force can be calculated by analyzing the modes $`G^l`$ of the original Green’s function, provided that in order to sum over all modes one applies the tilde-summation instead of the standard summation. The validity of this statement crucially depends on the vanishing of the tilde-sum over $`\delta G^l`$ \[Eq. (A8)\], which we now prove.
### Proof of Eq. (A8)
To calculate the $`l`$-modes of $`\delta G`$, it is convenient to use a spherical coordinate system in which the source point $`x_\mathrm{p}^\mu `$ lies on the polar axis (i.e., $`\theta _p=0`$). In this coordinate system, contributions to $`\delta G`$ would come only from the $`m=0`$ modes:
$`\delta G^l`$ $`=`$ $`{\displaystyle \underset{m=l}{\overset{l}{}}}Y_{lm}(\theta ,\phi ){\displaystyle _0^{2\pi }}𝑑\phi ^{}{\displaystyle _1^1}d(\mathrm{cos}\theta ^{})\delta G(\sigma ^{})Y_{lm}^{}(\theta ^{},\phi ^{})=`$ (A10)
$`L{\displaystyle _1^1}d(\mathrm{cos}\theta ^{})\left[a_0\delta (\sigma ^{})b_0\mathrm{\Theta }(\sigma ^{})\right]P_l(\mathrm{cos}\theta )P_l(\mathrm{cos}\theta ^{}),`$
where the integration is carried out over a sphere spanned by $`\theta ^{},\phi ^{}`$ (containing the evaluation point $`x^\mu `$), $`L(l+1/2)`$, and $`\sigma ^{}`$ is half the squared geodesic distance between the source point and the integration point $`\theta ^{},\phi ^{}`$. Now, the future light cone of the source point (along which $`\sigma ^{}=0`$) intersects the integration sphere along a circle $`\theta ^{}=\mathrm{const}\theta _0`$. The $`\delta (\sigma ^{})`$ term in Eq. (A10) contributes to the integration only along this circle, while the $`\mathrm{\Theta }(\sigma )`$ term contributes only across the part of the sphere outside it. We thus find
$$\delta G^l=LP_l(\mathrm{cos}\theta )\left[\widehat{a}_0P_l(\mathrm{cos}\theta _0)b_0_1^{\mathrm{cos}\theta _0}P_l(\mathrm{cos}\theta ^{})d(\mathrm{cos}\theta ^{})\right]\delta G_{(\delta )}^l+\delta G_{(\mathrm{\Theta })}^l,$$
(A11)
where $`\widehat{a}_0a_0/\left|d\sigma ^{}/d(\mathrm{cos}\theta ^{})\right|_{\theta ^{}=\theta _0}`$ is a constant, and where the symbols $`\delta G_\delta ^l`$ and $`\delta G_\mathrm{\Theta }^l`$ represent the two terms proportional to $`\widehat{a}_0`$ and $`b_0`$, respectively.
To calculate the tilde-sum over $`\delta G^l`$ we make use of the finite-sum identity \[referred to as “Christoffel’s first summation formula”—see, e.g., Eq. 3.8(20) of Ref. \]
$$(xy)\underset{l^{}=0}{\overset{l}{}}LP_l^{}(x)P_l^{}(y)=\frac{1}{2}(l+1)\left[P_{l+1}(x)P_l(y)P_{l+1}(y)P_l(x)\right],$$
(A12)
valid for $`|x|1`$ and $`|y|1`$. Applying this formula to Eq. (A11), we obtain
$`{\displaystyle \underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}}\delta G^l`$ $`=`$ $`\stackrel{~}{lim}_l\mathrm{}{\displaystyle \underset{l^{}=0}{\overset{l}{}}}\left(\delta G_{(\delta )}^l^{}+\delta G_{(\mathrm{\Theta })}^l^{}\right)=`$ (A15)
$`{\displaystyle \frac{1}{2}}\widehat{a}_0\stackrel{~}{lim}_l\mathrm{}\left[(l+1){\displaystyle \frac{P_{l+1}(x)P_l(x_0)P_{l+1}(x_0)P_l(x)}{xx_0}}\right]`$
$`{\displaystyle \frac{1}{2}}b_0\stackrel{~}{lim}_l\mathrm{}\left[(l+1){\displaystyle _1^{x_0}}{\displaystyle \frac{P_{l+1}(x)P_l(x^{})P_{l+1}(x^{})P_l(x)}{xx^{}}}𝑑x^{}\right],`$
where $`x`$ and $`x_0`$ stand for $`\mathrm{cos}\theta `$ and $`\mathrm{cos}\theta _0`$. (Recall that for any evaluation point $`x^\mu `$ lying inside the future light cone of the source point $`x_p^\mu `$, we have $`\mathrm{cos}\theta >\mathrm{cos}\theta _0\mathrm{cos}\theta ^{}`$; hence, the denominators appearing in the last expression are strictly positive.)
Let us consider first the tilde-sum of $`\delta G_{(\delta )}^l`$. For values of $`\theta `$ satisfying $`ϵ\theta \pi ϵ`$ (where $`ϵ>0`$) we have the large-$`l`$ asymptotic form
$$P_l(\mathrm{cos}\theta )\sqrt{2}(l\pi \mathrm{sin}\theta )^{1/2}\mathrm{cos}(L\theta +\pi /4)+O(l^{3/2})$$
(A16)
\[see, e.g., Eq. 3.9(2) of Ref. \]. Using this asymptotic form with Eq. (A15), we may easily write $`_{l^{}=0}^l\delta G_{(\delta )}^l^{}`$ as a sum of a few terms of the form $`a_j\mathrm{cos}(\alpha _jl+\beta _j)+O(l^1)`$ (in the case $`0<\theta <\pi `$), or of the form $`a_jl^{1/2}\mathrm{cos}(\alpha _jl+\beta _j)+O(l^{1/2})`$ (in the case $`\theta =0`$ or $`\pi `$), where $`a_j`$, $`\alpha _j0`$, and $`\beta _j`$ are certain functions of $`\theta `$ and $`\theta _0`$ (independent of $`l`$). Such terms all vanish at the tilde-limit $`l\mathrm{}`$. Thus, we clearly have
$$\underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}\delta G_{(\delta )}^l=0.$$
(A17)
We next turn to calculate the tilde-sum of $`\delta G_{(\mathrm{\Theta })}^l`$. Integrating by parts in Eq. (A15), and using
$$P_l(x)𝑑x=\frac{P_{l+1}(x)P_{l1}(x)}{2l+1}$$
(A18)
\[see Eq. 7.111 of Ref. , together with Eq. 8.733-4 therein\], we obtain
$`{\displaystyle \underset{l^{}=0}{\overset{l}{}}}\delta G_{(\mathrm{\Theta })}^l^{}={\displaystyle \frac{1}{2}}b_0(l+1)`$ $`\{{\displaystyle \frac{P_{l+1}(x)}{2l+1}}[{\displaystyle \frac{P_{l+1}(x^{})P_{l1}(x^{})}{xx^{}}}|_1^{x_0}{\displaystyle _1^{x_0}}{\displaystyle \frac{P_{l+1}(x^{})P_{l1}(x^{})}{(xx^{})^2}}dx^{}]`$ (A19)
$`{\displaystyle \frac{P_l(x)}{2l+3}}[{\displaystyle \frac{P_{l+2}(x^{})P_l(x^{})}{xx^{}}}|_1^{x_0}{\displaystyle _1^{x_0}}{\displaystyle \frac{P_{l+2}(x^{})P_l(x^{})}{(xx^{})^2}}dx^{}]\}.`$ (A20)
The surface terms here vanish at the lower boundary, as $`P_{l+1}(1)=P_{l1}(1)`$ and $`P_{l+2}(1)=P_l(1)`$ \[recalling $`P_l(1)=(1)^l`$\]. The contribution from the upper boundary dies off at large $`l`$ \[by virtue of Eq. (A16)\] as $`l^1`$ (times oscillations) for $`\theta 0`$, or as $`l^{1/2}`$ (times oscillations) for $`\theta =0`$. In both cases, this contribution thus vanishes at the standard limit $`l\mathrm{}`$, and hence also at the tilde-limit. Consider next the integral terms: The difference between two Legendre functions appearing in these terms may be globally bounded (in absolute value) as
$$\left|P_{l+1}(\mathrm{cos}\theta )P_{l1}(\mathrm{cos}\theta )\right|C_0[\pi (l1)]^{1/2},$$
(A21)
where $`C_0`$ is a number independent of $`l`$ and $`\theta `$ \[see Eq. 8.838 of Ref. \]. For any $`\theta `$ and $`\theta _0`$, the integrals of Eq. (A19) are thus each bounded (in absolute value) by $`C_1(\theta ,\theta _0)\times l^{1/2}`$, and we find
$$\left|\underset{l^{}=0}{\overset{l}{}}\delta G_{(\mathrm{\Theta })}^l^{}\right|C_2(\theta ,\theta _0)l^{1/2}\left[\left|P_{l+1}(\mathrm{cos}\theta )\right|+\left|P_l(\mathrm{cos}\theta )\right|\right]0$$
(A22)
as $`l\mathrm{}`$ (the above coefficients $`C_1`$ and $`C_2`$ are $`l`$-independent). Thus, the standard infinite sum over $`\delta G_{(\mathrm{\Theta })}^l`$ vanishes, and hence also the tilde-sum:
$$\underset{l=0}{\overset{\stackrel{~}{\mathrm{}}}{}}\delta G_{(\mathrm{\Theta })}^l=0.$$
(A23)
With Eqs. (A17) and (A23), Eq. (A8) is verified.
## B Derivation of integrals
In this appendix we obtain the tilde-integrals given in Eqs. (126) and (127), which are needed in the calculation of the regularization parameters. We start with Eq. (126). Let $`I_n^k(\lambda )`$ denote the primitive function of $`\lambda ^kJ_n(\lambda )`$, where $`\lambda `$ is a real variable, $`J_n`$ is the Bessel function of the first kind, of order $`n`$, and $`k,n`$:
$$I_n^k(\lambda )\lambda ^kJ_n(\lambda )𝑑\lambda .$$
(B1)
Let also $`\stackrel{~}{I}_n^k`$ stand for the definite integral
$$\stackrel{~}{I}_n^k\stackrel{~}{lim}_\lambda \mathrm{}_0^\lambda (\lambda ^{})^kJ_n(\lambda ^{})𝑑\lambda ^{}_0^\stackrel{~}{\mathrm{}}\lambda ^kJ_n(\lambda )𝑑\lambda .$$
(B2)
Consider first the case $`k=0`$. The standard integral $`_0^{\mathrm{}}J_n(\lambda )𝑑\lambda =1`$ is well defined and finite. Thus, in this case, the tilde-integration in Eq. (B2) may be replaced with a standard integration, yielding
$$\stackrel{~}{I}_n^{k=0}=1,n0.$$
(B3)
Consider next the case $`k>0`$. Writing in Eq. (B1) $`\lambda ^kJ_n(\lambda )=\lambda ^{kn1}\left[\lambda ^{n+1}J_n(\lambda )\right]=\lambda ^{kn1}\left[\lambda ^{n+1}J_{n+1}(\lambda )\right]^{}`$ \[where use is made of Eq. (69) and a prime denotes $`d/d\lambda `$\], and integrating by parts, we arrive at the recursive formula
$$I_n^k(\lambda )=\lambda ^kJ_{n+1}(\lambda )(kn1)I_{n+1}^{k1}(\lambda ).$$
(B4)
If $`0<kn`$, then by $`k`$ successive applications of this recursive formula we obtain
$$I_n^k(\lambda )=\underset{j=0}{\overset{k1}{}}\left[\frac{(nk1+2j)!!}{(nk1)!!}\lambda ^{kj}J_{n+1+j}(\lambda )\right]+\frac{(n+k1)!!}{(nk1)!!}I_{n+k}^0(\lambda ),$$
(B5)
Now, the Bessel functions $`J_n(\lambda )`$ admit the asymptotic form
$$J_n(\lambda \mathrm{})(2/\pi \lambda )^{1/2}\mathrm{cos}\left(\lambda n\pi /2\pi /4\right)$$
(B6)
(see, e.g., Eq. 8.451-1 in ). Therefore, each of the $`k`$ terms in the sum over $`j`$ in Eq. (B5) diverges at large $`\lambda `$ as some positive (half-integer) power of $`\lambda `$ times oscillations with respect to $`\lambda `$. Clearly, all such terms are eliminated when the tilde-limit $`\lambda \mathrm{}`$ is taken. Also, all of these terms vanish at $`\lambda =0`$. We are thus left with
$$\stackrel{~}{I}_n^k=\frac{(n+k1)!!}{(nk1)!!}\stackrel{~}{I}_{n+k}^0=\frac{(n+k1)!!}{(nk1)!!},\text{for }\text{0¡k≤n},$$
(B7)
where the last equality is due to Eq. (B3).
If $`k>n`$ and the difference $`kn`$ is even, then by $`p(kn)/2`$ applications of the recursive formula (B4) we obtain
$$I_n^k(\lambda )=\underset{j=0}{\overset{p1}{}}(1)^j\frac{(kn1)!!}{(kn12j)!!}\lambda ^{kj}J_{n+1+j}(\lambda )+(1)^p(kn1)!!I_{(k+n)/2}^{(k+n)/2}(\lambda ).$$
(B8)
Again, all terms in the sum over $`j`$ vanish at the tilde-limit $`\lambda \mathrm{}`$ and at $`\lambda =0`$, leading to
$`\stackrel{~}{I}_n^k`$ $`=`$ $`(1)^p(kn1)!!I_{(k+n)/2}^{(k+n)/2}(\lambda )=`$ (B10)
$`(1)^{(kn)/2}(kn1)!!(k+n1)!!,\text{for even }kn>0,`$
where the last equality is due to Eq. (B7).
The situation is different in case $`k>n`$ and the difference $`kn`$ is odd. Then, following $`q=(kn1)/2`$ applications of the recursive formula (B4) one obtains
$`I_n^k(\lambda )`$ $`=`$ $`{\displaystyle \underset{j=0}{\overset{q1}{}}}(1)^j{\displaystyle \frac{(kn1)!!}{(kn12j)!!}}\lambda ^{kj}J_{n+1+j}(\lambda )+(1)^q(kn1)!!I_{(k+n1)/2}^{(k+n+1)/2}=`$ (B12)
$`{\displaystyle \underset{j=0}{\overset{q}{}}}(1)^j{\displaystyle \frac{(kn1)!!}{(kn12j)!!}}\lambda ^{kj}J_{n+1+j}(\lambda ),`$
with no residual integral \[notice that when Eq. (B4) is applied with the upper index of $`I_n^k`$ greater by $`1`$ than its lower index, then the second term on the RHS of this recursive formula vanishes\]. This leads, when applying the tilde-limit, to
$$\stackrel{~}{I}_n^k=0,\text{for odd }\text{k-n¿0}.$$
(B13)
The above results, Eqs. (B3), (B7), (B10), and (B13), are summarized by Eq. (126) in Sec. VII.
We further need now to calculate the integral given in Eq. (127). Integrating by parts, we express the primitive function of the integrand, $`\lambda ^kJ_n(\lambda )\mathrm{ln}\lambda `$, for odd $`kn>0`$, as
$$I_n^{k(\mathrm{log})}\lambda ^kJ_n(\lambda )\mathrm{ln}\lambda d\lambda =I_n^k(\lambda )\mathrm{ln}\lambda \left[I_n^k(\lambda )/\lambda \right]𝑑\lambda .$$
(B14)
By virtue of Eq. (B12), the surface term on the RHS here is dominated at small $`\lambda `$ by $`\lambda ^{(k+n+1)/2}J_{(k+n+1)/2}(\lambda )\mathrm{ln}\lambda \lambda ^{k+n+1}\mathrm{ln}\lambda `$, and therefore it vanishes at the limit $`\lambda 0`$. However, this surface term diverges at the tilde-limit $`\lambda \mathrm{}`$ (as $`\lambda ^{k1/2}\mathrm{ln}\lambda `$ times oscillations with respect to $`\lambda `$). This can be avoided by slightly modifying the definition of a function’s tilde-limit, by allowing the quantities $`B_l^{(j)}`$ in Eq. (A6) to also admit the form $`B_l^{(j)}=a_jl^{b_j}\mathrm{ln}l\mathrm{cos}(\alpha _jl+\beta _j)`$ (with $`\alpha _j0`$ for all $`j`$). It can be shown that all features and results discussed in Appendix A concerning the tilde-limit remain valid also under this wider definition. With the revised definition of the tilde-limit, the surface term in Eq. (B14) vanishes at the tilde-limit $`\lambda \mathrm{}`$ as well as for $`\lambda 0`$. Denoting $`\stackrel{~}{I}_n^{k(\mathrm{log})}\stackrel{~}{lim}_\lambda \mathrm{}I_n^{k(\mathrm{log})}`$, we then have
$`\stackrel{~}{I}_n^{k(\mathrm{log})}={\displaystyle _0^\stackrel{~}{\mathrm{}}}\left[I_n^k(\lambda )/\lambda \right]𝑑\lambda ={\displaystyle \underset{j=0}{\overset{(kn1)/2}{}}}(1)^j{\displaystyle \frac{(kn1)!!}{(kn12j)!!}}\stackrel{~}{I}_{n+1+j}^{kj1},`$ (B15)
where we have substituted for $`I_n^k(\lambda )`$ from Eq. (B12). Provided that $`kn`$ is an odd number, the difference between the upper and lower indices of $`\stackrel{~}{I}_{n+1+j}^{kj1}`$ is also an odd number, $`kn22j`$. Therefore, by virtue of Eq. (B13), we have $`\stackrel{~}{I}_{n+1+j}^{kj1}=0`$ for any $`j`$ satisfying $`kj1>n+1+j`$, i.e., $`j<(kn2)/2`$. We find that of all the terms summed up in Eq. (B15), the only nonvanishing one is the one with $`j=(kn1)/2`$. Hence,
$`\stackrel{~}{I}_n^{k(\mathrm{log})}=(1)^{(kn1)/2}(kn1)!!\stackrel{~}{I}_{(k+n+1)/2}^{(k+n1)/2}=`$ $`(1)^{(kn+1)/2}(kn1)!!(k+n1)!!,`$ (B16)
where, in the last equality, the value of $`\stackrel{~}{I}_{(k+n+1)/2}^{(k+n1)/2}`$ has been inferred from Eq. (B7). This proves Eq. (127).
|
warning/0005/quant-ph0005019.html
|
ar5iv
|
text
|
# Symmetric Hybrid Dynamics: A canonical formulation of coupled classical-quantum dynamics
##
### Introduction.
The interaction between the space time structure and the matter/energy content of the universe is described by General Relativity. The fact that all forms of matter/energy interact with the space time structure lies at the foundations of the theory. A pertinent question is then what is the form of the gravitational interaction for quantum fields . In the absence of a theory of quantum gravity several attempts to describe the space time - matter/energy interaction by a semiclassical formulation of gravity have been made . In such a theory the space time structure is described by a classical metric while the matter content is described by standard quantum observables. An interesting debate has been whether these semiclassical formulations will reproduce the predictions of quantum gravity in the appropriate limit .
This is one of several examples where what seems to be missing is a consistent general formulation of coupled classical-quantum dynamics. That is, a theory derived as the appropriate limit of quantum mechanics that provides a consistent description of a general interaction between classical and quantum subsystems .
It seems reasonable to expect that such a theory should satisfy the two following prerequisites: a) under some general conditions - concerning the initial data and the dynamical structure of the particular system under consideration - its predictions should be consistent with the full quantum ones, and b) the theory should display a consistent dynamical structure.
In two different proposals for a theory of coupled classical-quantum dynamics were postulated and motivated in terms of the thus resulting properties. It was also proved that the theory presented in can be derived as the appropriate limit of quantum mechanics. Unfortunately, the two proposals do not display a consistent canonical structure. The theories are not formulated over a Lie algebra of observables and thus time evolution is problematic at several levels . In fact, it was later proved that standard classical and quantum mechanics cannot be consistently coupled .
Recently a new formulation of hybrid dynamics was presented by Diosi, Gisin and Strunz . The starting point of this proposal is quantum mechanics in the Schrödinger picture. Hybrid dynamics is then derived using coherent state methods under some assumptions concerning the initial data and the dynamical behaviour. However, it is not clear in that context whether it will display a fully developed canonical structure.
In this letter we shall present an alternative proposal for coupled classical quantum dynamics. The starting point will be the full quantum formulation (in the Heisenberg picture) of a general dynamical system. Using a specific order dequantization we will be able to derive symmetric hybrid dynamics. The procedure automatically ensures that, firstly, the theory is consistent with quantum mechanics in the appropriate limit and secondly, it displays a standard canonical structure.
### Canonical Structure.
Let us then start by establishing the conventions and assembling some general results. Let us consider a generic $`(N+M)`$-dimensional quantum system with fundamental observables $`(\widehat{q}_k,\widehat{p}_k)`$ $`(k=1,\mathrm{},N+M)`$ or sucinctly just $`\widehat{O}_k`$ $`(k=1,\mathrm{},2(N+M))`$. These operators act on the Hilbert space $`=_1_2`$. The Hilbert space $`_1`$ is spanned by the eigenvectors $`|z_1,\mathrm{},z_N>`$ of the classical sector observables $`\widehat{O}_i`$ $`(i=1,\mathrm{},2N)`$ or $`(\widehat{q}_i,\widehat{p}_i)`$ $`(i=1,\mathrm{},N)`$. The remaining subspace $`_2`$ is spanned by the eigenvectors $`|w_1,\mathrm{},w_M>`$ of the quantum sector observables $`\widehat{O}_\alpha `$ $`(\alpha =2N+1,\mathrm{},2(N+M))`$ or $`(\widehat{q}_\alpha ,\widehat{p}_\alpha )`$ $`(\alpha =N+1,\mathrm{},N+M)`$.
The algebra of linear operators of the quantum system is generated by the set:
$$\widehat{𝒢}\{\widehat{O}_k,k=1,\mathrm{},2(N+M)\}.$$
(1)
This algebra is an infinite dimensional complex vector space. In , we saw that a possible basis for this vector space is given by the set of completely symmetric operators:
$$\widehat{}=\left\{\widehat{O}_{k_1\mathrm{}k_n}=\left(\widehat{O}_{k_1}\mathrm{}\widehat{O}_{k_n}\right)_+;1k_1,\mathrm{},k_n2(N+M);n𝒩\right\}.$$
(2)
Since any observable from the classical sector commutes with all observables of the quantum sector, the set $`\widehat{}`$ can be written as the set of elements of the form:
$$\widehat{}=\left\{\widehat{O}_{i_1\mathrm{}i_n}\widehat{O}_{\alpha _1\mathrm{}\alpha _m}=\left(\widehat{O}_{i_1}\mathrm{}\widehat{O}_{i_n}\right)_+\left(\widehat{O}_{\alpha _1}\mathrm{}\widehat{O}_{\alpha _m}\right)_+\right\},$$
(3)
with $`1i_1,\mathrm{},i_n2N;2N+1\alpha _1,\mathrm{}\alpha _m2(N+M);n,m𝒩`$. These elements are linearly independent and generate all elements in the algebra $`\widehat{𝒜}(_1_2)`$. Therefore $`\widehat{}`$ constitutes a basis for $`\widehat{𝒜}(_1_2)`$.
Let us now define the symmetric half-dequantization map. The classical phase-space associated with the classical sector is denoted by $`T^{}M_1`$.
Definition 1: Dequantization $`V_S^{H.Q.}`$
$$V_S^{H.Q.}:\widehat{𝒜}(_1_2)𝒮=\{f:T^{}M_1\widehat{𝒜}(_2)\}.$$
1) $`V_S^{H.Q.}`$ is a linear map,
2) $`V_S^{H.Q.}(\widehat{1})=1_C\widehat{1}_Q`$,
3) $`V_S^{H.Q.}\left(\widehat{O}_{i_1\mathrm{}i_n}\widehat{O}_{\alpha _1\mathrm{}\alpha _m}\right)=O_{i_1\mathrm{}i_n}\widehat{O}_{\alpha _1\mathrm{}\alpha _m}`$,
where $`O_{i_1\mathrm{}i_n}=O_{i_1}\mathrm{}O_{i_n}T^{}M_1`$.
The dequantization $`V_S^{H.Q.}`$ satisfies the following properties:
1) Dequantization of the product: Let $`\widehat{A},\widehat{B}\widehat{𝒜}`$. We have: $`V_S^{H.Q.}(\widehat{A}\widehat{B})=V_S^{H.Q.}(\widehat{A})V_S^{H.Q.}(\widehat{B})`$, where $``$ is an extension of the well-known $``$-product and is given by:
$$:𝒮\times 𝒮𝒮;\stackrel{~}{A}\stackrel{~}{B}=\stackrel{~}{A}\mathrm{exp}\left(\frac{i\mathrm{}}{2}\widehat{𝒥}\right)\stackrel{~}{B},$$
(4)
where $`\stackrel{~}{A}=V_S^{H.Q.}(\widehat{A})`$Notice that $`\stackrel{~}{A}`$ can be seen as an element of $`\widehat{𝒜}(_2)`$ of the form $`\stackrel{~}{A}=_nA_n^C\widehat{A}_n^Q`$, where $`A_n^CT^{}M_1`$ and $`\widehat{A}_n^Q\widehat{𝒜}(_2)`$., $`\stackrel{~}{B}=V_S^{H.Q.}(\widehat{B})`$ and
$$\widehat{𝒥}\underset{i=1}{\overset{N}{}}\left(\frac{\stackrel{}{}}{q_i}\frac{\stackrel{}{}}{p_i}\frac{\stackrel{}{}}{p_i}\frac{\stackrel{}{}}{q_i}\right).$$
(5)
The product $``$ is associative, distributive with respect to the sum and has a neutral element. Therefore the set $`(𝒮,,+)`$ is a complex ring.
2) Dequantization of the hermitean conjugate:
$$V_S^{H.Q.}(\widehat{A}^{})=V_S^{H.Q.}\left(\underset{i=1}{\overset{n}{}}c_i^{}\left(\underset{j=1}{\overset{m}{}}\widehat{X}_{ij}\right)_+^{}\right)=\left[V_S^{H.Q.}(\widehat{A})\right]^{},$$
(6)
where we considered the expansion of $`\widehat{A}`$: $`\widehat{A}=_{i=1}^nc_i_{j=1}^m\widehat{X}_{ij}`$ and $`\widehat{X}_{ij}\widehat{𝒢}`$.
3) Dequantization of the bracket: Using the map $`V_S^{H.Q.}`$, we can define a new bracket structure in $`𝒮`$. Let $`\widehat{A},\widehat{B}\widehat{𝒜}`$. We have:
$$\left[[,]\right]_M:𝒮\times 𝒮𝒮;\left[[V_S^{H.Q.}(\widehat{A}),V_S^{H.Q.}(\widehat{B})]\right]_M=V_S^{H.Q.}\left([\widehat{A},\widehat{B}]\right).$$
The explicit form of the new bracket, which is an extension of the Moyal bracket , is given by:
$$\begin{array}{c}\left[[\stackrel{~}{A},\stackrel{~}{B}]\right]_M=\stackrel{~}{A}\stackrel{~}{B}\stackrel{~}{B}\stackrel{~}{A}=\stackrel{~}{A}\mathrm{exp}\left(\frac{i\mathrm{}}{2}\widehat{𝒥}\right)\stackrel{~}{B}\stackrel{~}{B}\mathrm{exp}\left(\frac{i\mathrm{}}{2}\widehat{𝒥}\right)\stackrel{~}{A}=\\ \\ =[\stackrel{~}{A},\stackrel{~}{B}]+\frac{i\mathrm{}}{2}\{\stackrel{~}{A},\stackrel{~}{B}\}\frac{i\mathrm{}}{2}\{\stackrel{~}{B},\stackrel{~}{A}\}+𝒪(\mathrm{}^2).\end{array}$$
(7)
If $`\stackrel{~}{A},\stackrel{~}{B}𝒜(T^{}M_1)`$ -the algebra of classical sector observables-, then the bracket $`\left[[,]\right]_M`$ reduces to the Moyal bracket. Likewise, for $`\stackrel{~}{A},\stackrel{~}{B}\widehat{𝒜}(_2)`$, the bracket reduces to the quantum commutator. Futhermore if we disregard the terms of order $`\mathrm{}^2`$ or higher the bracket reduces to the Boucher-Traschen bracket . The properties of the new bracket follow immediately from the properties of the full quantum one: it is antisymmetric, linear and satisfies the Jacobi identity. Therefore, $`(𝒮,\left[[,]\right]_M)`$ is a Lie algebra. Moreover, the map $`V_S^{H.Q.}`$ is a Lie algebra isomorphism. A quantization prescription for symmetric hybrid dynamics is given by the symmetric half quantization map $`\mathrm{\Lambda }_S^{H.Q.}`$ which is defined by $`\mathrm{\Lambda }_S^{H.Q.}V_S^{total}=V_S^{H.Q}`$ (where $`V_S^{total}:\widehat{𝒜}(_1_2)𝒜(T^{}M_1T^{}M_2)`$ named total dequantization map, is the extension of the map $`V_S^{H.Q}`$ to the case where the entire system is dequantized ). The map $`\mathrm{\Lambda }_S^{H.Q.}`$ is also a Lie algebra isomorphism.
4) Dynamical Structure: Consider the algebraic structure of symmetric hybrid dynamics. We can easily obtain the dynamical equations of the theory. They follow directly from dequantizing the full quantum equations. The time evolution of a general observable $`\stackrel{~}{A}`$ is thus given by:
$$\dot{\stackrel{~}{A}}=\frac{1}{i\mathrm{}}\left[[\stackrel{~}{A},\stackrel{~}{H}]\right]_M,$$
(8)
where $`\stackrel{~}{H}=V_S^{H.Q.}(\widehat{H})`$ or, alternatively, $`\stackrel{~}{H}=\mathrm{\Lambda }_S^{H.Q.}(H)`$ is the hybrid hamiltonian. The formal solution of this equation is given by:
$$\stackrel{~}{A}(t)=\underset{n=0}{\overset{+\mathrm{}}{}}\frac{1}{n!}\left(\frac{it}{\mathrm{}}\right)^n\left[[\stackrel{~}{H},\left[[\stackrel{~}{H},\mathrm{}\left[[\stackrel{~}{H},\stackrel{~}{A}]\right]_M\mathrm{}]\right]_M]\right]_M.$$
(9)
Alternatively, the time evolution of $`\stackrel{~}{A}`$ is generated by an unitary operator $`\stackrel{~}{U}`$:
$$\stackrel{~}{A}(t)=\stackrel{~}{U}(t)^{}\stackrel{~}{A}(0)\stackrel{~}{U}(t),$$
(10)
where $`\stackrel{~}{U}(t)`$ is the solution of the hybrid Schrödinger equation:
$$i\mathrm{}\frac{\stackrel{~}{U}}{t}=\stackrel{~}{H}\stackrel{~}{U},\stackrel{~}{U}(0)=1_C\widehat{1}_Q,$$
(11)
and satisfies $`\stackrel{~}{U}^{}=\stackrel{~}{U}^1`$. Time evolution is a canonical transformation. More generally, all unitary transformations are canonical transformations:
$$\stackrel{~}{U}^1\left[[\stackrel{~}{A},\stackrel{~}{B}]\right]_M\stackrel{~}{U}=\left[[\stackrel{~}{U}^1\stackrel{~}{A}\stackrel{~}{U},\stackrel{~}{U}^1\stackrel{~}{B}\stackrel{~}{U}]\right]_M,$$
(12)
and so the dynamical structure of the algebra $`(𝒮,)`$ is invariant under unitary transformations.
### Predictions.
Let the initial data for a general hybrid system be given by the initial wave function $`|\varphi ^Q>_2`$ for the quantum sector plus a set of values $`(O_i^0,\delta _i),i=\mathrm{1..2}N`$ for the classical sector (where $`\delta _i`$ are the classical error margins associated to the initial values $`O_i^0`$ of the classical sector observables $`O_i`$). The aim now is to obtain physical predictions for the time evolution of the system. The naive procedure would be to determine a set of eigenvectors of $`\stackrel{~}{O}_k(t)=\stackrel{~}{O}_k(\widehat{O}_\alpha ,O_i^0,t)`$ spanning the Hilbert space $`_2`$ (let these eigenvectors be $`|b_k,m_k>`$, so that $`\stackrel{~}{O}_k(t)|b_k,m_k>=b_k|b_k,m_k>`$, where $`b_k`$ is the associated eigenvalue and $`m_k`$ is the degeneracy index) and then assume that the predictions of hybrid dynamics, for the outputs of a measurement of the observable $`O_k`$ at the time $`t`$, consist of the set of values $`b_k`$ with associated probabilities $`p(b_k)=_{m_k}|<b_k,m_k|\varphi ^Q>|^2`$.
However there is no reason to believe that such predictions are physically valid. The reason is straightforward. The real physical observable is $`\widehat{O}_k(t)=\widehat{O}_k(\widehat{O}_\alpha ,\widehat{O}_i,t)`$ obtained using the full quantum formulation of the dynamical system. $`\widehat{O}_k(t)`$ has the eigenvectors $`|a_k,n_k>`$, where $`a_k`$ is the associated eigenvalue and $`n_k`$ is the corresponding degeneracy index. The physical predictions for the output of a measurement of the observable $`O_k`$ are given by the values $`a_k`$ with probability $`p(a_k)=_{n_k}|<a_k,n_k|\varphi >|^2`$ where $`|\varphi >`$ is the total wave function describing both the classical and the quantum sectors at the initial time. Clearly there is no reason why these predictions should be the same as the ones obtained by using the operator $`\stackrel{~}{O}_k`$.
Our best chance will be to use the operator $`\stackrel{~}{O}_k`$ to obtain some knowledge (but not the complete knowledge) about the outputs of a measurement of $`\widehat{O}_k`$. This can be done through a procedure similar to the one presented in . Let us then summarize the main points of that method:
i) The first step is the following: quantum and hybrid dynamics provide two alternative descriptions of the initial time configuration of the dynamical system. The true, physical, description is the quantum one given by a total wave function $`|\varphi >`$ that we assume, for simplicity, to be of the form $`|\varphi >=|\varphi ^Q>|\varphi ^c>`$ where $`|\varphi ^Q>_2`$ and $`|\varphi ^c>_1`$. These two wave functions describe the quantum and the classical sector initial time configurations, respectively. On the other hand hybrid dynamics provides only an approximate description of the initial time configuration. Concerning the quantum sector, we are given exactly the same description: $`|\varphi ^Q>`$, whereas for the classical sector, we only have the set of values $`(O_i^0,\delta _i)`$ available. Our main task is then to understand under which conditions the two descriptions of the classical sector are consistent. Clearly, if they are not consistent, then we can not expect hybrid dynamics to yield sensible predictions. This problem was studied in detail in in the context of the semiclassical limit of quantum mechanics and the result turned out to be a set of criteria establishing a notion of classicality. These criteria were then used in to test the physical validity of a proposed theory of hybrid dynamics originally suggested in .
The classicality criterion consisted of a set of conditions relating the classical and the quantum description of the initial time configuration of a general dynamical system that, when satisfied, ensure that the classical predictions be consistent (in some precise sense) with the quantum ones at all times. Here we shall also use this classicality criterion. We thus impose that $`|\varphi ^c>`$ be first order classical (i.e. it should satisfy the first order classicality criterion) with respect to the classical data $`(O_i^0,\delta _i)`$ (notice, however, that higher order classicality criteria, imposing more stringent conditions on the functional form of the classical sector initial data wave function, can also be used ). This means that $`|\varphi ^c>`$ should satisfy:
$$<E(\widehat{S}_{i_n},|\varphi ^c>,S_{i_n})|E(\widehat{S}_{i_n},|\varphi ^c>,S_{i_n})>\delta _{S_{in}}^2$$
(13)
where $`|E(\widehat{S}_{i_n},|\varphi ^c>,S_{i_n})>`$ is the error ket associated to the sequence of observables $`S_{i_n}=O_{i_1}\mathrm{}.O_{i_n};1i_1,..,i_n2N;n𝒩`$, defined by:
$$|E(\widehat{S}_{i_n},|\varphi ^c>,S_{i_n})>=(\widehat{O}_{i_1}O_{i_1}^0)\mathrm{}.(\widehat{O}_{i_n}O_{i_n}^0)|\varphi ^c>,$$
(14)
and $`S_{i_n}`$ is any sequence of observables $`S_{i_n}=O_{i_1}\mathrm{}.O_{i_n}`$ such that:
$$\frac{^n\stackrel{~}{O}_k(t)}{S_{i_n}}=\frac{^n\stackrel{~}{O}_k(t)}{O_{i_1}\mathrm{}.O_{i_n}}0,$$
(15)
for any of the observables $`\stackrel{~}{O}_k(t),k=\mathrm{1..2}(N+M)`$. Moreover $`\delta _{S_{in}}=\delta _{i_1}\mathrm{}.\delta _{i_n}`$.
In summary, a general quantum system admits a hybrid description with initial data $`(|\varphi ^Q>,O_i^0,\delta _i)`$ if the initial data wave function $`|\varphi ^c>`$ describing the classical sector in the full quantum description admits a proper classical description. This in turn means that $`|\varphi ^c>`$ should satisfy the set of conditions (13) for all sequences satisfying (15).
ii) In the second step of our approach, a general relation between the operators $`\widehat{O}_k(t)=\widehat{A}`$ and $`\stackrel{~}{O}_k(t)=\widehat{B}`$ is presented and then used to establish a relation between the eigenvectors of $`\stackrel{~}{O}_k(t)`$ and the eigenvectors of $`\widehat{O}_k(t)`$. Let us then proceed along these lines. It can be proved , that for $`\widehat{B}=V_S^{H.Q.}(\widehat{A})=\stackrel{~}{A}`$:
$$\widehat{A}\widehat{B}=\underset{i=1}{\overset{2N}{}}\frac{\widehat{B}}{O_i}(\widehat{O}_iO_i)+\frac{1}{2}\underset{i,j=1}{\overset{2N}{}}\frac{^2\widehat{B}}{O_iO_j}(\widehat{O}_iO_i)(\widehat{O}_jO_j)+\mathrm{}$$
(16)
We now construct the set of eigenstates of $`\widehat{B}`$, $`|\psi _k,m_k>=|b_k,m_k>|\varphi ^c>_1_2`$, which can be used to expand $`|\varphi >`$, and derive the explicit form of the error ket of these states, in the representation of $`\widehat{A}`$, around the corresponding eigenvalue:
$$\begin{array}{c}|E(\widehat{A},|\psi _k,m_k>,b_k)>=(\widehat{A}\widehat{B})|\psi _k,m_k>=_{i=1}^{2N}|E(\widehat{O}_i,|\varphi ^c>,O_i)>\frac{\widehat{B}}{O_i}|b_k,m_k>+\\ \\ +\frac{1}{2}_{i,j=1}^{2N}|E(\widehat{O}_i,\widehat{O}_j,|\varphi ^c>,O_i,O_j)>\frac{^2\widehat{B}}{O_iO_j}|b_k,m_k>+\mathrm{}\end{array}$$
(17)
Using this quantity it is straightfoward to prove that $`\mathrm{\Delta }(\widehat{A},|\psi _k,m_k>,b_k,p)=<E(\widehat{A},|\psi _k,m_k>,b_k)|E(\widehat{A},|\psi _k,m_k>,b_k)>^{1/2}/(1p)^{1/2}`$ (where $`0p<1`$ is a probability) satisfy:
$$\mathrm{\Delta }(\widehat{A},|\psi _k,m_k>,b_k,p)\frac{1}{(1p)^{1/2}}\underset{i=1}{\overset{2N}{}}|<b_k,m_k|\frac{\widehat{B}^{}}{O_i}\frac{\widehat{B}}{O_i}|b_k,m_k>|^{1/2}\delta _i+\mathrm{}$$
(18)
where we explicitly used the requirement (13). $`\mathrm{\Delta }(\widehat{A},|\psi _k,m_k>,b_k,p)`$ is named the spread of the state $`|\psi _k,m_k>`$ in the representation of $`\widehat{A}`$ and can be used to study the properties of $`|\psi _k,m_k>`$ in that representation. Namely it can be proved that in this representation the state $`|\psi _k,m_k>`$ has at least a probability $`p`$ confined to the interval of eigenvalues of $`\widehat{A}`$, $`I=[b_k\mathrm{\Delta }(\widehat{A},|\psi _k,m_k>,b_k,p),b_k+\mathrm{\Delta }(\widehat{A},|\psi _k,m_k>,b_k,p)]`$. This result is valid for all $`0p<1`$. The functional form of $`\mathrm{\Delta }`$ suggests that we write it as $`\mathrm{\Delta }(\widehat{A},|\psi _k,m_k>,b_k,p)=\delta _B/(1p)^{1/2}`$ where $`\delta _B=<E(\widehat{A},|\psi _k,m_k>,b_k)|E(\widehat{A},|\psi _k,m_k>,b_k)>^{1/2}`$ is named the error of $`\widehat{B}`$ and is of the size of a classical error margin.
iii) Using the former relation between the eigenvectors of $`\widehat{B}`$ and the eigenvectors of $`\widehat{A}`$ we are able to derive a relation between the probabilities in the representation of $`\widehat{B}`$ and those in the representation of $`\widehat{A}`$ and thus to obtain predictions for the outputs of a measurement of $`\widehat{A}`$ using only the knowledge about the hybrid operator $`\widehat{B}`$ and its eigenvectors $`|b_k,m_k>`$. Let us then consider an arbitrary interval of eigenvalues of $`\widehat{A}`$: $`I_0=[a^0D,a^0+D]`$, $`D>2\mathrm{\Delta }(p)`$ and let $`P(a_kI_0)`$ be the probability of a measurement of the observable $`\widehat{A}`$ yielding a value $`a_k`$ inside the interval $`I_0`$. The full quantum mechanical predictions are given by:
$$P(a_kI_0)=\underset{n_k,a_kI_0}{}|<\varphi |a_k,n_k>|^2.$$
(19)
Now $`P(a_kI_0)`$ can be expanded using the eigenvectors of $`\widehat{B}`$. We have:
$$P(a_kI_0)=\underset{n_k,a_kI_0}{}\left|\underset{m_k,b_k}{}<\varphi |\psi _k,m_k><\psi _k,m_k|a_k,n_k>\right|^2,$$
(20)
where we used the fact that $`|\varphi >=_{b_k,m_k}<\psi _k,m_k|\varphi >|\psi _k,m_k>`$. Using the properties of the states $`|\psi _k,m_k>`$ in the representation of $`\widehat{A}`$ we can finally prove, through a quite long calculation, that:
$$1\{P(b_kI_{min})^{1/2}+(1p)^{1/4}\}^2P(a_kI_0)\{P(b_kI_{max})^{1/2}+(1p)^{1/4}\}^2$$
(21)
where $`I_{min}=[a^0(D2\mathrm{\Delta }),a^0+(D2\mathrm{\Delta })]`$, $`I_{max}=[a^0(D+2\mathrm{\Delta }),a^0+(D+2\mathrm{\Delta })]`$ and $`\mathrm{\Delta }=\mathrm{\Delta }(\widehat{A},|\psi _k,m_k>,b_k,p)`$ is given by (18). Notice that we can make $`(1p)`$ as small as desired. However, this will affect the value of $`\mathrm{\Delta }`$ and thus the range of the intervals $`I_{min}`$ and $`I_{max}`$.
To see this explicitly let us make $`p=0.99`$. $`\mathrm{\Delta }`$ in (18) is given by $`\mathrm{\Delta }=10\delta _B`$ where $`\delta _B=<E(\widehat{A},|\psi _k,m_k>,b_k)|E(\widehat{A},|\psi _k,m_k>,b_k)>^{1/2}`$ is of the order of magnitude of the classical error margins. Using this value of the spread we can state that, in the representation of $`\widehat{A}`$, the eigenvalue of $`\widehat{B}`$, $`|\psi _k,m_k>`$ has at least 99% of its probability distribution confined to the interval $`[b_k\mathrm{\Delta },b_k+\mathrm{\Delta }]`$. The predictions for the outputs of a measurement of $`\widehat{A}`$ are then:
$$\begin{array}{c}1(P(b_kI_{min})^{1/2}+0.32)^2P(a_kI_0)(P(b_kI_{max})^{1/2}+0.32)^2\\ \\ P(b_kI_{min})0.74P(a_kI_0)P(b_kI_{max})+0.74\end{array}$$
(22)
where $`I_{max},I_{min}=[a_0(D\pm 20\delta _B),a_0+(D\pm 20\delta _B)]`$. An error of $`74\%`$ in a prediction of a probability is, of course, inadmissible.
Let us now increase the value of $`p`$ and see what happens. Let $`p=0.9999`$. $`\mathrm{\Delta }`$ is now given by $`\mathrm{\Delta }=100\delta _B`$ and the predictions for the outputs of a measurement of $`\widehat{A}`$:
$$\begin{array}{c}1(P(b_kI_{min})^{1/2}+0.1)^2P(a_kI_0)(P(b_kI_{max})^{1/2}+0.1)^2\\ \\ P(b_kI_{min})0.21P(a_kI_0)P(b_kI_{max})+0.21\end{array}$$
(23)
An error of $`21\%`$ is more reasonable. Notice however that the difference between the range of $`I_0`$ and $`I_{max},I_{min}`$ increased considerably - from $`20\delta _B`$ to $`200\delta _B`$. We can continue to increase the value of $`p`$ but this will affect the value of $`\mathrm{\Delta }`$ and thus it is not likely that we might obtain more precise predictions.
We should have expected symmetric hybrid dynamics to provide imprecise predictions since the theory uses classical data to describe the initial time configuration of one of its sectors and classical data are imprecise in nature. However, the degree of imprecision is very large which is due to the fact that the conditions imposed on the classical sector initial data wave function are the least restrictive possible. It is worth pointing out that more accurate predictions are possible either by completely reformulating the procedure by which symmetric hybrid dynamics makes predictions for the outputs of a measurement of the full quantum operators or, albeit using the former procedure, by imposing more restrictive conditions on the classical sector initial data wave function .
To see what happens in this latter case let us assume that the classical sector initial data wave function satisfies the 2nd order classicality criterion. This means that $`|\varphi ^c>`$ should satisfy the conditions (13) for all 2nd order sequences of the form $`S_{i_n}=(S_{i_n^{}},S_{i_n^{\prime \prime }})`$ where $`S_{i_n^{}}`$ and $`S_{i_n^{\prime \prime }}`$ are any of the 1-order sequences determined in (15). In this case it can be proved that $`\mathrm{\Delta }`$ in (18) is (with a good aproximation) given by $`\mathrm{\Delta }=<E|E>^{1/2}/(1p)^{1/4}=\delta _B/(1p)^{1/4}`$ and, moreover that the new spread allows for the same statement concerning the confinement of the eigenvectores of $`\widehat{B}`$. The predictions for the outputs of a measurement of $`\widehat{A}`$ become considerably more precise:
$$1\{P(b_kI_{min})^{1/2}+(1p)^{3/8}/3^{1/2}\}^2P(a_kI_0)\{P(b_kI_{max})^{1/2}+(1p)^{3/8}/3^{1/2}\}^2$$
(24)
Let us make, once again, $`p=0.9999`$. This time $`\mathrm{\Delta }=\delta _B/(1p)^{1/4}=10\delta _B`$ and:
$$\begin{array}{c}1(P(b_kI_{min})^{1/2}+0.02)^2P(a_kI_0)(P(b_kI_{max})^{1/2}+0.02)^2\\ \\ P(b_kI_{min})0.04P(a_kI_0)P(b_kI_{max})+0.04\end{array}$$
(25)
That is an error of at most $`4\%`$, with the difference between the range of $`I_0`$ and $`I_{min},I_{max}`$ decreasing to $`20\delta _B`$. If we keep increasing the degree of the classicality conditions to be satisfied by the classical sector initial data wave function we will certainly increase the degree of precision of the hybrid predictions. However, we will also narrow the range of validity of hybrid dynamics to those systems with a classical subsystem satisfying higher order classicality criteria.
### Example.
To illustrate the former results let us consider the simple hybrid system described by the hamiltonian:
$$\stackrel{~}{H}=\frac{1}{2}(x^2+q^2)+kx\widehat{Q}$$
(26)
where $`(q,x)`$ are the canonical variables of a classical harmonic oscillator with $`m=w=1`$, $`(\widehat{Q},\widehat{P})`$ are the fundamental observables of the quantum system to which the harmonic oscillator is coupled and $`k`$ is the coupling constant. By solving the system of equations (8) we obtain the time evolution of the fundamental observables:
$`\stackrel{~}{q}(t)`$ $`=`$ $`q(0)\mathrm{cos}t+\{x(0)+k\widehat{Q}(0)\}\mathrm{sin}t`$ (27)
$`\stackrel{~}{x}(t)`$ $`=`$ $`q(0)\mathrm{sin}t+x(0)\mathrm{cos}t+k\{\mathrm{cos}t1\}\widehat{Q}(0)`$ (28)
$`\stackrel{~}{Q}(t)`$ $`=`$ $`\widehat{Q}(0)`$ (29)
$`\stackrel{~}{P}(t)`$ $`=`$ $`\widehat{P}(0)k\{q(0)\mathrm{cos}t+x(0)\mathrm{sin}t\}k^2\{\mathrm{sin}tt\}\widehat{Q}(0)`$ (30)
where $`(q(0),x(0))`$ and $`(\widehat{Q}(0),\widehat{P}(0))`$ are the initial time classical and quantum sector fundamental observables, respectively. Finally the $`L`$-order spreads (18) ($`L𝒩`$) are given by:
$`\mathrm{\Delta }_L(\widehat{q},p)`$ $`=`$ $`\delta _{\stackrel{~}{q}}/(1p)^{1/2L}=\{|\mathrm{cos}t|\delta _q(0)+|\mathrm{sin}t|\delta _x(0)\}/(1p)^{1/2L}`$ (31)
$`\mathrm{\Delta }_L(\widehat{x},p)`$ $`=`$ $`\delta _{\stackrel{~}{x}}/(1p)^{1/2L}=\{|\mathrm{sin}t|\delta _q(0)+|\mathrm{cos}t|\delta _x(0)\}/(1p)^{1/2L}`$ (32)
$`\mathrm{\Delta }_L(\widehat{Q},p)`$ $`=`$ $`\delta _{\stackrel{~}{Q}}/(1p)^{1/2L}=0`$ (33)
$`\mathrm{\Delta }_L(\widehat{P},p)`$ $`=`$ $`\delta _{\stackrel{~}{P}}/(1p)^{1/2L}=\{|k\mathrm{cos}t|\delta _q(0)+|k\mathrm{sin}t|\delta _x(0)\}/(1p)^{1/2L}`$ (34)
where $`0p<1`$ is a probability, $`\delta _q(0),\delta _x(0)`$ are the initial data classical error margins and the results are valid up to the order of classicality $`L`$ of the classical sector initial data wave function $`|\varphi ^c>`$. In particular they are valid for the first and second order spreads $`L=1,2`$ (if the initial data wave function $`|\varphi ^c>`$ is first or second order classical) that were previously used to obtain the general predictions (21,24).
### Conclusions.
A canonical formulation of coupled classical and quantum dynamics was presented. The theory satisfies an interesting set of properties. On the one hand, and most importantly, for a general dynamical system satisfying some general conditions concerning its initial data and dynamical behaviour, the predictions of symmetric hybrid dynamics are consistent with the predictions of full quantum mechanics (and, in fact, can be used to obtain full quantum predictions) and are therefore physically valid.
On the other hand, symmetric hybrid dynamics displays a fully consistent canonical structure: i) it is formulated over a Lie algebra of observables, ii) time evolution is unitary and probabilities are positive defined, iii) the theory admits a set of general canonical transformations which are generated by the action of unitary operators and iv) time evolution, in particular, is a canonical transformation.
Moreover, the limit of symmetric hybrid dynamics when the classical sector does not exist is quantum mechanics. However, one should notice that the limit when the quantum sector does not exist is not standard classical mechanics. Instead this limit is symmetric classical mechanics, an alternative theory of classical mechanics that was proposed and studied in , and this is the reason why the results of this letter do not contradict those of . Therefore symmetric hybrid dynamics properly generalizes both quantum and symmetric classical mechanics. Futhermore, the quantization prescription from symmetric classical mechanics to symmetric hybrid dynamics is just a Lie algebra isomorphism and is thus not plagued with ordering ambiguities.
Finally, notice that the limit of symmetric classical mechanics when $`\mathrm{}0`$ is just standard classical mechanics. Likewise, in this limit, the bracket structure and thus the dynamical structure of symmetric hybrid dynamics are just the ones originally proposed by Boucher and Traschen in . This theory was later proved to be consistent with quantum mechanics under the same assumptions that were made in this letter in what concerns the initial data but using an approximation procedure that discarded the contributions of the terms proportional to $`\mathrm{}^2`$ or smaller. The results of the present letter corroborate, by straightforward order of magnitude considerations, the procedure used in and thus the validity of the Boucher-Traschen bracket as a possible dynamical structure (unfortunately ill behaved) for coupled classical-quantum dynamics.
### Acknowledgments
This work was partially supported by the grants ESO/PRO/1258/98 and CERN/P/Fis/15190/1999.
|
warning/0005/cond-mat0005310.html
|
ar5iv
|
text
|
# Dielectric nonlinearity of relaxor ferroelectric ceramics at low ac drives
## I Introduction
Since Pb(Mg<sub>1/3</sub>Nb$`{}_{2/3}{}^{})`$O<sub>3</sub> (PMN) was first synthesized by Smolenski et al. in the late 1950s, there has been a series of relaxor ferroelectrics (relaxors) with complex perovskite structure whose dielectric and ferroelectric properties are rather different from that of normal ferroelectrics. For the relaxor ferroelectrics, the dielectric permittivity is unusually high, the sintering temperature is rather low, and the temperature coefficient of capacitance is quite small due to the diffuse phase transition (DPT), which lead to the successful application as Multi-layered Capacitors (MLC). In addition, the field-induced piezoelectric effect of relaxors is strong, and the pulse echo response of transducer can be controlled by bias voltage. So the relaxors are competent in the range of actuators, medical diagnostic transducers, etc.. Recently, the observation of the highly excellent electromechanical properties in some single crystal of relaxors (for example, the PMN-PT solid solution) has brought great interest to the research, development and application of this kind of materials.
In actual applications, the components and device made of relaxors usually work under dc bias or ac drive voltages. So the performances of the material under external field always cause great interests. In recent years, some works focus on the dielectric nonlinear response under various ac drive voltages. Apart from the strong application background, these works can also provide some clues of the polarization mechanism of relaxors in theory. Although several possible models have been proposed, the nature of the dielectric response of relaxors, especially PMN, keeps unclear. Experiments such as high-resolution transmission electron microscopy (HTEM) have confirmed a main feature of PMN in structure: a great amount of nanoscaled ordered microregions are embedded randomly in the disordered matrix. The ordered microregions are probably nonstoichiometric. The assumption that the ordered microregions are just the centers of polar microregions presented by the superparaelectric model need further testification of experiments, but the theoretical calculation based on the assumption can explain the particularly large permittivility of PMN.
For normal ferroelectrics such as BaTiO<sub>3</sub>, the nature of dielectric nonlinearity at different ac drives has been universally accepted, i. e., the nonlinearity is caused by the movement of domain walls among the ferroelectric domains with different polarization directions. For relaxor ferroelectrics, opposite with normal ferroelectrics, no macro phase transition takes place, and no ferroelectric domain with micrometric dimensions appears. Under the usual conditions, the nanoscaled polar microregions in relaxors do not grow into ferroelectric domains even if the temperature is much lower than $`T_{max}`$ (the temperature where the permittivity reaches a maximum). Therefore, it is greatly interesting to investigate the connections between the dielectric nonlinear response and the polar microregions.
The low frequency dielectric properties of PMN at different ac drives were first reported by Bokov et al. in the early 1960s. But there have been little works reported on this subject until 1990s. In the recent decade, with the widely application of PMN-type relaxors and the increasing interests on the polarization mechanism of relaxors, the research of nonlinearity was emphasized again. Experiments showed that $`T_{max}`$ shifts to lower temperature with increasing ac drive amplitude for PMN. And the case of PLZT with relaxor behavior is similar. The curves of permittivity at various drives are similar to the frequency dispersion in relaxors. Glazounov et al. presented that the nonlinear behavior was controlled by the domain wall motion rather than the reorientation of polar clusters (i. e., superparaelectric approach). However, they did not explain neither the type of domains nor the process of domain wall motion. Colla et al. investigated the experiments of PMN-PT single crystals and pointed out that the nonlinearity mechanism is related to the drive amplitude: a glass-like dynamics of the polarization freezing process is dominant at low drives; at intermediate drives, the movements and reconstruction of the boundaries of polar nanodomains take place; at higher drives, the interactions between polar regions cause the formation of normal micron-sized domains and the movement of domain walls.
In this paper, the nonlinear dielectric response of 0.9PMN-0.1PT ceramics at various ac drive amplitudes and frequencies was studied. The experimental results were qualitatively explained by the phase transition theory of ergodic space shrinking in succession, and a Monte Carlo simulation was conducted to verify the theory.
## II Experimental procedure
0.9PMN-0.1PT powder was prepared by the columnbite precursor method. The starting materials are analytically pure PbO, Nb<sub>2</sub>O<sub>5</sub>, TiO<sub>2</sub>, and (MgCO<sub>2</sub>)<sub>4</sub>.Mg(OH)<sub>2</sub>.6H<sub>2</sub>O. MgNb<sub>2</sub>O<sub>6</sub> was synthesized by (MgCO<sub>2</sub>)<sub>4</sub>.Mg(OH)<sub>2</sub>.6H<sub>2</sub>O and Nb<sub>2</sub>O<sub>5</sub> at 1000 C. MgNb<sub>2</sub>O<sub>6</sub>, PbO, and TiO<sub>2</sub> powders were mixed and calcined at 870 C for 2 h. Then the PMN-PT powders obtained were pressed into pellets ($`\varphi `$10$`\times `$1–2 mm) at 100 MPa, and sintered in the PbO-rich atmosphere for 2 h at 1200 C. The specimen were analyzed by the X-ray diffraction technique on a diffractometer (Science D/max-RA) using CuK<sub>α</sub> raciation, and a pure perovskite structure was confirmed. Finally, the specimen were polished to 0.4 mm, and plated by silver. The dielectric permittivity was measured using a HP4284 LCR meter over the frequency range 1–100 kHz at a heating rate 3 K/min. The amplitude of the ac measuring field is 0.05, 0.25, 0.40, and 0.50 kV/cm.
## III Results and Discussions
The usual amplitude of the ac signal used to measure the dielectric permittivity of relaxors is 0.01 kV/cm. The results obtained correspond to the slope of the hysteresis loop, $`P/E`$, at the starting point. Because the amplitude is small enough to fall in the linear-response region, $`P/E`$ is constant, which represents the dielectric permittivity $`\epsilon ^{}`$, i. e., the $`\epsilon ^{}`$ value is independent on the amplitude of external field. However, with the increasing of ac field amplitude, the nonlinear terms can not be ignored.
Fig.1–Fig.3 show the change in the dielectric permittivity measured at various amplitudes when the ac field frequency is 1 kHz, 10 kHz and 100 kHz, respectively. From the figures one can list the following features: (1) when the measuring frequency is fixed, the dielectric permittivity keeps constant for various ac drives at high temperatures, while it increases with increasing the ac amplitude at low temperatures; (2) with increasing the amplitude, the dielectric permittivity maximum, $`\epsilon _m^{}`$, increases and shifts to lower temperature; (3) the diffusion behavior is more evident at larger ac amplitude; (4) the dielectric nonlinear effect is weakened at lower frequencies. Fig. 4 demonstrates the dielectric permittivity at various frequencies when the amplitude is fixed as 0.05 kV/cm. (The curves for $`E=`$0.25, 0.4 and 0.5 kV/cm are omitted since they are similar to Fig.4.) It is noted that increasing amplitude has the same effects on the permittivity as decreasing frequency.
It can be seen from the results above that the effect of ac field amplitude on the permittivity maximum $`\epsilon _m^{}`$ is obvious. The relation between $`\epsilon _m^{}`$ and amplitude is depicted in Fig. 5. A linear law is found in the range of amplitude and frequency under study, which is consistent with the results of single crystal. With increasing the frequency, the effect of amplitude on $`\epsilon _m^{}`$ is weakened. Extrapolate the curves in Fig. 5 to the zero field, we can obtain the permittivity maximum without nonlinear effect. Table I gives the variation of $`\epsilon _m^{}`$ when the amplitude increases from 0.05 kV/cm to 0.5 kV/cm. It shows that $`\epsilon _m^{}`$ increases by 7.4% at 1 kHz (which is the frequency in usual measurements). This means that the measuring result of permittivity is affected by the weak-field nonlinearity. As a result, the thickness and the applied voltage of specimen, i. e., the field strength, should be specified to avoid the influence of nonlinear effect, so that the results in different experiments are comparable.
It was mentioned in Sec. I that there were different explains on the nonlinear effect at weak fields. Glazounov et al. denied the mechanism of reorientation of polar clusters, but they have ignored the interactions between polar clusters. Colla et al. presented that a glass-like dynamics of the polarization freezing process dominates at low drives. However, the connections between the dynamic process and the nonlinear effect were not explained.
We proposed that the dielectric nonlinearity of relaxor ferroelectrics at low drives, as well as the frequency dispersion, can be explained in the theory of the phase transition of ergodic space shrinking in succession. The TEM dark-field image of 0.9PMN-0.1PT proved that there are a great amount of nanoscaled ordered microregions embedded in the disordered matrix. At a certain temperature, the homogeneous crystal structure of the ordered microregions causes the cooperative displacement of B-site cations along one of eight $`111`$-equivalent directions. When the temperature is high enough, the thermal energy, $`k_BT`$, is much larger than the energy barriers between different directions, which results that the probabilities of displacement along eight directions are equal, i. e., the system is ergodic. Thus the ordered microregions are unpolar. However, the environments of the microregions along different directions are not identical, and the spherical symmetry of ordered microregions breaks down. Since the potential wells of different $`111`$ directions are different, the B-site ions tend to stay along the direction with the lowest well for more time in the thermal flipping process when the temperature decreases to a certain value. Thus the ordered microregions transform into the polar microregions, and dipole behavior appears. Due to the cooperative displacement, the ions in the same microregion flip as a whole unit under external drives, so the dielectric permittivity of relaxor ferroelectrics is extraordinarily high. The polar microregions are random the magnitude and direction of polarization. Under zero field, $`p_i=0`$, while $`p_i^20`$. When the temperature is much higher than the freezing temperature, the relaxation times of polar microregions are much shorter than the observation time. All the polar microregions flip dynamically with ac drives, so the sum of polarizations, $`P`$, is proportional to the external field, i. e., the dielectric permittivity, $`\epsilon =P/E`$, is independent on the frequency and the amplitude of external field. In this temperature range, the deviation of micro-polarization direction from that of external field by thermal fluctuation is weakened with decreasing temperature, and then the permittivity increases with decreasing temperature. This corresponds to the high temperature regions in Fig.1–Fig.4, where no frequency dispersion and nonlinearity is found. The polar microregions can be regarded as independent dipoles. When the temperature further decreases, the electrostatic interactions between dipoles get more and more strong. Under the ac drives with a certain frequency, the flip of a dipole is affected by both the external field applied and the internal field generated by other dipoles. Some dipoles cannot keep up with the switching of the measuring field, and become “slow dipoles”. Some are even frozen along a certain direction, and become “frozen dipoles”. Thus the phase space with ergodicity is shrinking in succession. When the frequency increases, the time scale of dipole flipping is shortened. More dipoles cannot reach the equilibrium states in the observation time, i.e., the proportion of slow dipoles and frozen dipoles increases. Slow dipoles and frozen dipoles give no or little contribution to the flipping polarization. So the dielectric permittivity decreases, which is the frequency dispersion in relaxors. When the ac field amplitude increases, the driving force on dipoles is enhanced. Slow dipoles and frozen dipoles are forced to flip faster and give more contribution to the flipping polarization. The proportion of slow dipoles and frozen dipoles decreases, and the dielectric permittivity increases, which is the nonlinear effects in relaxors. This is the origin of the special dielectric properties in relaxors (frequency dispersion, nonlinearity, etc.).
It should be emphasized that the external field discussed above is in the range of weak drives (less than 0.6 kV/cm according to Ref. 16). Only at weak drives can nanoscaled dipoles exist, and the long-range interactions between dipoles dominate in the dynamic process. If the external field increases, motions and reconstruction of the boundaries of polar nanodomains would take place. Under the strong fields, through interacting with the disordered matrix surrounding, polar nanodomains will switch, coalesce, and grow into the conventional micron-sized ferroelectric domains as that in normal ferroelectrics. Then the model discussed above is not applicable to describe the polarization dynamics.
In order to better understand the dielectric nonlinearity in relaxors and verify the model above, a Monte Carlo simulation is conducted in the next section to investigate the dynamic flipping process of the polar microregions .
## IV Monte Carlo simulation
Gui et al. have used the Monte Carlo method to simulate the dynamics of freezing process in relaxor ferroelectrics. In the theory framework of Ref. 21, the polar microregions are modeled as point dipoles. The interaction between two dipoles with moment $`\underset{i}{\overset{}{\mu }}`$ and $`\underset{j}{\overset{}{\mu }}`$ is expressed as
$`J_{ij}=J_{ji}`$ $`=`$ $`\stackrel{}{\mu _j}{\displaystyle \frac{1}{4\pi \epsilon _0}}({\displaystyle \frac{3\stackrel{}{\mu _i}\widehat{r}_{ij}}{r^3}}\widehat{r}_{ij}{\displaystyle \frac{\stackrel{}{\mu _i}}{r^3}})`$ (1)
$`=`$ $`{\displaystyle \frac{1}{4\pi \epsilon _0}}{\displaystyle \frac{3\mathrm{cos}\phi _i\mathrm{cos}\phi _j\mathrm{cos}\varphi }{r^3}}\mu _i\mu _j,`$ (2)
where $`\widehat{r}_{ij}`$ is the unit vector between the two dipoles. $`r`$ is the distance between dipoles. $`\phi _i`$ ($`\phi _j`$) is the angle between $`\stackrel{}{\mu _i}`$ ($`\stackrel{}{\mu _j}`$) and $`\widehat{r}_{ij}`$. $`\varphi `$ is the angle between $`\stackrel{}{\mu _i}`$ and $`\stackrel{}{\mu _j}`$. The Hamiltonian of relaxors at dc bias is obtained as
$$H=\frac{1}{2}\underset{ij}{}J_{ij}E\underset{i}{}\mu _i\mathrm{cos}\theta _i,$$
(3)
where $`E`$ is the dc field strength. $`\theta _i`$ is the angle between $`\stackrel{}{E}`$ and $`\underset{i}{\overset{}{\mu }}`$.
The effective interaction energy, $`\stackrel{}{J_{ij}}`$, is introduced as
$$\stackrel{}{J_{ij}}\sigma _i\sigma _j=J_{ij}\mu _i\mu _j/2,$$
(4)
where $`\sigma _i=\pm 1`$ is the projection of $`\stackrel{}{\mu _i}`$ on the direction of the external field. Eq. (2) can be rewritten as
$$H=\underset{ij}{}\stackrel{}{J_{ij}}\sigma _i\sigma _jE\overline{\mu }\underset{i}{}\frac{|\mu _i\mathrm{cos}\theta _i|}{\overline{\mu }}\sigma _i,$$
(5)
where $`\overline{\mu }`$ is the maximal projection of dipole moments on the external field.
Ref. 21 used Eq. (4) to study the dielectric origins of relaxor ferroelectrics under dc external field. It is unsuitable for the case of ac field. However, the polarization mechanism should be similar in both ac and dc fields. Therefore, an ac field term is introduced as
$$E(t)=E\mathrm{cos}\left(2\pi \frac{t}{t_L}\right),$$
(6)
where $`t_L`$ is the period of the ac field, which corresponds to the frequency. Thus a Hamiltonian similar to Eq. (4) is obtained:
$$H=\underset{ij}{}\stackrel{}{J_{ij}}\sigma _i\sigma _jE(t)\overline{\mu }\underset{i}{}\frac{|\mu _i\mathrm{cos}\theta _i|}{\overline{\mu }}\sigma _i.$$
(7)
A Gaussian distribution is assumed for $`\stackrel{}{J_{ij}}`$, i. e.,
$$P(\stackrel{}{J_{ij}})exp[\frac{\stackrel{2}{\stackrel{}{J_{ij}}}}{2(\mathrm{\Delta }J)^2}],$$
(8)
where $`\mathrm{\Delta }J`$ is the distribution width.
There are N dipoles in the system. (N=$`16\times 16\times 16`$) The flipping probability of the $`i`$th dipole is defined as
$$W=\frac{1}{e^{\delta H/k_BT}+1},$$
(9)
where $`\delta H`$ is the change of energy when the dipole flip from $`\sigma _i`$ to -$`\sigma _i`$. The details of simulation process can be found in Ref. 21. Then the ac permittivity can be obtained as
$$\chi =\frac{1}{E}\frac{1}{t_{obs}}\underset{t_0}{\overset{t_0+t_{obs}}{}}p(t)\mathrm{exp}\left(i2\pi \frac{t}{t_L}\right)𝑑t,$$
(10)
where $`p(t)`$ is the average polarization.
Fig. 6 shows the temperature dependence of dielectric permittivity at various ac amplitudes when the frequency is fixed as $`t_L`$=5 MCS/dipole. It can be seen that the diffusion behavior is enhanced with increasing amplitude. The simulation results are consistent with experiments in main features, which verifies the polarization mechanism described by the above model. It is noted that the curve of $`E`$=3.0$`\mathrm{\Delta }J/\overline{\mu }`$ (which corresponds to a stronger field) lies below other curves at high temperatures. In this case, maybe the corresponding field is too strong and cause the growth of dipoles, so the weak-field model is not applicable.
## V Conclusions
The dielectric nonlinear response of 0.9PMN-0.1PT ceramics was revealed over the field range 0.05–0.50 kV/cm. When the measurement frequency is fixed, the dielectric permittivity is invariant with field amplitude at high temperatures. At low temperatures, the permittivity maximum, $`\epsilon _m^{}`$, increases and shifts to lower temperatures with increasing amplitude. A linear law between $`\epsilon _m^{}`$ and the amplitude was observed at all frequencies and amplitudes in the experiment. The nonlinearity is weakened at higher frequencies. The effects of increasing amplitude are similar to that of decreasing frequency. It was proposed that the nonlinearity of relaxors at low drives can be explained in the theory of the phase transition of ergodic space shrinking in succession. A Monte Carlo simulation was conducted to investigate the dynamic flips of polar microregions at low drives and verify the proposition.
This work was supported by the Chinese National Science Foundation (Grant NO. 59995520).
|
warning/0005/nlin0005018.html
|
ar5iv
|
text
|
# Spatial competition and price formation
## 1 Introduction
There are several basic concepts which lie at the heart of economic theory. They are the ”economic atom” which is usually considered to be the individual, profits, money, price and markets and the more complex organism the firm. Much of economic theory is based on utility maximizing individuals and profit maximizing firms. The concept of a utility function attributes to individuals a considerable amount of sophistication. The proof of its existence poses many difficult problems in observation and measurement. In this study of market and price formation we consider simplistic social individuals who must buy to eat and who look for where to shop for the best price. In this foray into dynamics we opt for a simple model of consumer price formation. Our firms are concerned with survival rather than a sophisticated profit maximization. Yet we relate these simple behaviors to the more conventional and complex ones.
A natural way to approach the economic physics of monopolistic competition is to introduce space explicitly. For much of economic analysis of competition space and information are critical factors. The basic aspects of markets involve an intermix of factors, such as transportation costs and delivery times which depend explicitly on physical space. But for pure information, physical distance is less important than direct connection. For questions concerning the growth of market areas, the spatial representation is appropriate. Consideration of space is sufficient to provide a justification of Chamberlin’s model of monopolistic competition as is evident from the work of Hotelling . Furthermore it is reasonably natural to consider space on a grid with some form of minimal distance. Many of the instabilities found in economic models such as the Bertrand model are not present with an appropriate grid.
When investigating these topics, one quickly finds that many aspects of price formation can be understood in terms of generalized evolutionary dynamics. In consequence, our first models in this paper study spatial competition and cluster formation without the generation of price (Sec. 3). This generates cluster size distributions, which can be compared to real world data. We spend some time investigating theoretical models which can explain our simulation data (Sec. 4). We then, finally, move on to price formation, where we implement the price dynamics “on top” of the already analyzed spatial competition models (Sec. 5). The paper is concluded by a discussion and a summary.
## 2 Related work
The model is an open one related to the partial equilibrium models of much of micro-economics. In particular money and its acceptance in trade is taken as a primitive concept. There is a literature on the acceptance of money both in a static equilibrium context (see for example ) and in a ”bootstrap” or dynamic context (see for example ). These are extremely simple closed models of the economy where each individual is both a buyer and seller. Eventually we would like to construct a reasonable model where the acceptance of money, the emergence of competitive price and the emergence of market structure all arise from the system dynamics. This will call for an appropriate combination of the features of the model presented here with the closed models noted above. We do not pursue this further here. Instead by taking the acceptance of money as given our observations are confined to the emergence of markets and the nature of price. The static economic theories of monopoly and mass homogeneous competitive equilibrium provide natural upper and lower benchmarks to gauge market behavior. The intermediate zone between $`n=2`$ and very many is covered in the economic literature by various oligopoly models, of which those of Cournot , Bertrand and Chamberlin serve as exemplars. The Chamberlin model unlike the earlier models stresses that all firms trade in differentiated goods. They are all in part differentiated or partially monopolistic. When one considers both information and physical location this is a considerable step towards greater realism. Other work on evolutionary or behavioral learning in price formation are Refs. .
## 3 Spatial competition
As mentioned in the introduction, we will start with spatial models without price. We will add price dynamics later.
### 3.1 Basic spatial model (domain coarsening)
We use a 2-dimensional $`N=L\times L`$ grid with periodic boundary conditions. Sites are numbered $`i=1..N`$. Each site belongs to a cluster, denoted by $`c(i)`$. Initially, each site belongs to “itself”, that is, $`c(i)=i`$, and thus cluster numbers also go from $`1`$ to $`N`$.
The dynamics is such that in each time step we randomly pick a cluster, delete it, and the corresponding sites are taken over by neighboring clusters. Since the details, in particular with respect to the time scaling, make a difference, we give a more technical version of the model. In each time step, we first select a cluster for deletion by randomly picking a number $`C`$ between $`1`$ and $`N`$. All sites belonging to the cluster (i.e. $`c(i)=C`$) are marked as “dead”. We then let adjoining clusters grow into the “dead” area. Because of the interpretation later in the paper, in our model the “dead” sites play the active role. In parallel, they all pick randomly one of their four nearest neighbors. If that neighbor is not dead (i.e. belongs to a cluster), then the previously dead site will join that cluster. This step is repeated over and over, until no dead sites are left. Only then, time is advanced and the next cluster is selected for deletion.
In physics this is called a domain coarsening scheme (e.g. ): Clusters are selected and deleted, and their area is taken over by the neighbors. This happens with a total separation of time scales, that is, we do not pick another cluster for deletion before the distribution of the last deleted cluster has finished. Fig. 1 shows an example. We will call a cluster of size larger than zero “active”.
Note that it is possible to pick a cluster that has already been deleted. In that case, nothing happens except that the clock advances by one. This implies that there are two reasonable definitions of time:
* Natural time $`t`$: This is the definition that we have used above. In each time step, the probability of any given cluster to be picked for deletion is a constant $`1/N`$, where $`N=L^2`$ is the system size. Note that it is possible to pick a cluster of size zero, which means that nothing happens except that time advances by one.
* Cluster time $`\stackrel{~}{t}`$: An alternative is to chose between the *active* clusters only. Then, in each time step, the probability of any given cluster to be picked for deletion is $`1/n(\stackrel{~}{t})`$, where $`n(\stackrel{~}{t})=N\stackrel{~}{t}`$ is the number of remaining active clusters in the system.
Although the dynamics can be described more naturally in cluster time, we prefer natural time because it is closer to our economics interpretation.
At any particular time step, there is a typical cluster size. In fact, in cluster time, since there are $`n(\stackrel{~}{t})=N\stackrel{~}{t}`$ clusters, the average cluster size as a function of cluster time is $`\overline{S}(\stackrel{~}{t})=N/n(\stackrel{~}{t})=1/(1\stackrel{~}{t}/N)`$. However, if one *averages over all time steps*, we find a scaling law. In cluster time, it is numerically close to $`\stackrel{~}{n}(s)s^3\text{ or }\stackrel{~}{n}(>s)s^2,`$ where $`s`$ is the cluster size, $`n(s)`$ is the number of clusters of size $`s`$, and $`n(>s)`$ is the number of clusters with size larger than $`s`$.<sup>1</sup><sup>1</sup>1In this paper, we will also use $`N(s)=sn(s)`$ for the cluster size distribution in logarithmic bins, in particular for the figures. In natural time, the large clusters have more weight since time moves more slowly near the end of the coarsening process. The result is again a scaling law (Fig. 2 (left)), but with exponents increased by one:
$$n(s)s^2\text{ or }n(>s)s^1.$$
(1)
It is important to note that this is not a steady state result. The result emerges when averaging over the whole time evolution, starting with $`N`$ clusters of size one and ending with one cluster of size $`N`$.
### 3.2 Random injection with space
In view of evolution, for example in economics or in biology, it is realistic to inject new small clusters. A possibility is to inject them at random positions. So in each time step, before the cluster deletion described above, in addition with probability $`p_{\mathrm{𝑖𝑛𝑗}}`$ we pick one random site $`i`$ and inject a cluster of size one at $`i`$. That is, we set $`c(i)`$ to $`i`$. This is followed by the usual cluster deletion. It will be explained in more detail below what this means in terms of system-wide injection and deletion rates.
This algorithm maintains the total separation of time scales between the cluster deletion (slow time scale) and cluster growth (fast time scale). That is, no other cluster will be deleted as long as there are still “dead” sites in the system. Note that the definition of time in this section corresponds to natural time.
The probability that the injected cluster is really new is reduced by the probability to select a cluster that is already active. The probability of selecting an already active cluster is $`n(t)/N`$, where $`n(t)`$ is again the number of active clusters. In consequence, the effective injection rate is
$$r_{\mathrm{𝑖𝑛𝑗},\mathrm{𝑒𝑓𝑓}}=p_{\mathrm{𝑖𝑛𝑗}}n(t)/N.$$
(2)
Similarly, the effective cluster deletion depends on the probability of picking an active cluster, which is $`n(t)/N`$. In consequence, the effective deletion rate is
$$r_{\mathrm{𝑑𝑒𝑙},\mathrm{𝑒𝑓𝑓}}=n(t)/N.$$
(3)
This means that, in the steady state, there is a balance of injection and deletion, $`n_{}/N=p_{\mathrm{𝑖𝑛𝑗}}n_{}/N`$, and thus the steady state average cluster number is
$$n_{}=Np_{\mathrm{𝑖𝑛𝑗}}/2.$$
(4)
In consequence, the steady state average cluster size is
$$s_{}=N/n_{}=2/p_{\mathrm{𝑖𝑛𝑗}}.$$
(5)
The cluster size distribution for the model of this section is numerically close to a log-normal distribution, see Fig. 2 (right). Indeed, the position of the distribution moves with $`1/p_{\mathrm{𝑖𝑛𝑗}}`$ (not shown). In contrast to Sec. 3.1, this is now a steady state result.
### 3.3 Injection on a line
It is maybe intuitively clear that the injection mechanism of the model described in Sec. 3.2 destroys the scaling law from the basic model without injection (Sec. 3.1), since injection at random positions introduces a typical spatial scale. One injection process that actually generates steady-state scaling is injection along a 1-d line. Instead of the random injection of Sec. 3.2, we now permanently set
$$c(i)=i$$
(6)
for all sites along a line. Fig. 3 (left) shows a snapshot of this situation.
In this case, we numerically find a stationary cluster size distribution (Fig. 3 (right)) with
$$n(s)s^{1.5}\text{ or }n(>s)s^{0.5}.$$
(7)
Since the injection mechanism here does not depend on time, and since the cluster size distribution itself is stationary, it is independent from the specific definition of time.
### 3.4 Random injection without space
One could ask what would happen without space. A possible translation of our model into “no space” is: Do in parallel: Instead of picking one of your four nearest neighbors, you pick an arbitrary other agent (random neighbor approximation). If that agent is not dead, copy its cluster number. Do this over and over again in parallel, until all agents are part of a cluster again. A cluster is now no longer a spatially connected structure, but just a set of agents. In that case, we obtain again power laws for the size distribution, but this time with slopes that depend on the injection rate $`p_{\mathrm{𝑖𝑛𝑗}}`$ (Fig. 4); see Sec. 4.4 for details.
### 3.5 Real world company size distributions
Fig. 5 shows actual retail company size distributions from the 1992 U.S. economic census , using annual sales as a proxy for company size. We use the retail sector because we think that it is closest to our modelling assumptions — this is discussed at the end of Sec. 6. We show two curves: establishment size, and firm size.<sup>2</sup><sup>2</sup>2An establishment is “a single physical location at which business is conducted. It is not necessarily identical with a company or enterprise, which may consist of one establishment or more.” . It is clear that in order to be comparable with our model assumptions, we need to look at establishment size rather than at company size.
Census data comes in unequally spaced bins; the procedure to convert it into useable data is described in the appendix. Also, the last four data points for firm size (not for the establishment size, however) were obtained via a different method than the other data points; for details, again see the appendix.
From both plots, one can see that there is a typical establishment size around $400 000 annual sales; and the typical firm size is a similar number. This number intuitively makes sense: With, say, income of 10% of sales, smaller establishments will not provide a reasonable income.
One can also see from the plots that the region around that typical size can be fitted by a log-normal. We also see, however, that for larger numbers of annual sales, such a fit is impossible since the tail is much fatter. A scaling law with
$$n(>s)s^1\text{ corresponding to }n(s)s^2$$
(8)
is an alternative here.<sup>3</sup><sup>3</sup>3Remember again, that slopes from log-log plots in logarithmic bins are different by one from the exponent in the distribution. So $`n(s)s^2`$ corresponds to a slope $`1`$ *both* in the accumulated distribution $`n(>s)`$ and when plotting logarithmic bins $`N(s)/N(1)`$.
This is, however, at odds with investigations in the literature. For example, Ref. find a log-normal, and by using a Zipf plot they show that for large companies the tail is *less* fat than a log-normal. However, there is a huge difference between our and their data: They only use *publicely traded* companies, while our data refers to all companies in the census. Indeed, one finds that their plot has its maximum at annual sales of $`\$10^8`$, which is already in the tail of our distribution. This implies that the small scale part of their distribution comes from the fact that small companies are typically not publicely traded. In consequence, it reflects the dynamics of companies entering and exiting from the stock market, not entry and exit of the company itself.
We conclude that from available data, company size distributions are between a log-normal and a power law with $`n(s)s^2`$ or $`n(>s)s^1`$. Further investigation of this goes beyond the scope of this paper.
## 4 Theoretical considerations
### 4.1 Spatial coarsening model (slope -2 in natural time)
We are looking again at the “basic model”. In cluster time this was: randomly pick one of the clusters, and give it to the neighbors. The following heuristic model gives insight:
1. We start with $`N`$ clusters of size 1.
2. We need $`N/2`$ time steps to delete $`N/2`$ of them and with that generate $`N/2`$ clusters of size 2.
3. In general, we need $`N/2^k`$ time steps to move from $`N/2^{k1}`$ clusters of size $`2^{k1}`$ to $`N/2^k`$ clusters of size $`2^k`$.
4. If we sum this over time, then in each logarithmic bin at $`s=2^k`$ the number of contributions is $`N/2^k\times N/2^k`$, i.e. $`s^2`$.
5. Since these are logarithmic bins, this corresponds to $`\stackrel{~}{n}(s)s^3\text{ or }\stackrel{~}{n}(>s)s^2,`$ which was indeed the simulation result in cluster time.
6. In natural time, we need a constant amount of time to move from $`k1`$ to $`k`$, and thus obtain via the same argument $`n(s)s^2\text{ or }n(>s)s^1,`$ which was the simulation result in natural time.
### 4.2 Random injection in space (log-normal)
At the moment, we do not have a consistent explanation for the log-normal distribution in the spatial model. A candidate is the following: Initially, most injected clusters of size one are *within* the area of some larger and older cluster. Eventually, that surrounding cluster gets deleted, and all the clusters of size one spread in order to occupy the now empty space. During this phase of fast growth, the speed of growth is proportional to the perimeter, and thus to $`\sqrt{s}`$, where $`s`$ is the area. Therefore, $`\sqrt{s}`$ follows a biased multiplicative random walk, which means that $`\mathrm{log}(\sqrt{s})=\mathrm{log}(s)/2`$ follows a biased additive random walk. In consequence, once that fast growth process stops, $`\mathrm{log}(s)`$ should be normally distributed, resulting in a log-normal distribution for $`s`$ itself. In order for this to work, one needs that this growth stops at approximately the same time for all involved clusters. This is apprixomately true because of the “typical” distance between injection sites which is inversely proportional to the injection rate. More work will be necessary to test or reject this hypothesis.
### 4.3 Injection on a line (slope -3/2)
If one looks at a snapshot of the 2D picture for “injection on a line” (Fig. 3), one recognizes that one can describe this as a structure of cracks which are all anchored at the injection line. There are $`L`$ such cracks (some of length zero); cracks merge with increasing distance from the injection line, but they do not branch.
According to Ref. , this leads naturally to a size exponent of $`3/2`$, as found in the simulations. The argument is the following: The whole area, $`L^2`$, is covered by
$$𝑑ssn(s),$$
(9)
where $`n(s)`$ is the number of clusters of size $`s`$ on a linear scale. We assume $`n(s)s^\tau `$, however the normalization is missing. If all clusters are anchored at a line of size $`L`$, then a doubling of the length of the line will result in twice as many clusters. In consequence, the normalization constant is $`L`$, and thus $`n(s)Ls^\tau `$. Now we balance the total area, $`L^2`$, with what we just learned about the covering clusters:
$$L^2𝑑ssLs^\tau =L𝑑ss^{1\tau }Ls^{2\tau }|_0^S.$$
(10)
Assuming that $`\tau <2`$, then the integral does not converge for $`S\mathrm{}`$, and we need to take into account how the cut-off $`S`$ scales with $`L`$. This depends on how the cracks move in space as a function of the distance from the injection line. If the cracks are roughly straight, then the size of the largest cluster is $`L^2`$. If the cracks are random walks, then the size of the largest cluster is $`L^{3/2}`$. In consequence:
* For “straight” lines: $`L^2L(L^2)^{2\tau }`$ $``$ $`2=1+2(2\tau )`$ $``$ $`\tau =3/2.`$
* For random walk: $`2=1+3/2(2\tau )\tau =4/3.`$
Since our simulations result in $`\tau 3/2`$, we conclude that our lines between clusters are not random walks. This is intuitively reasonable: When a cluster is killed, then the growth is biased towards the center of the deleted cluster, thus resulting in random walks which are all differently biased. This bias then leads to the “straight line” behavior. — This implies that the $`s^{3/2}`$ steady state scaling law hinges on two ingredients (in a 2D system): (i) The injection comes from a 1D structure. (ii) The boundaries between clusters follow something that corresponds to straight lines. As we have seen, the biasing of a random walk is already enough to obtain this effect.
### 4.4 Injection without space (variable slope)
Without space, clusters do not grow via neighbors, but via random selection of one of their members. That is, we pick a cluster, remove it from the system, and then give its members to the other clusters one by one. The probability that the agent choses a cluster $`i`$ is proportional to that cluster’s size $`s_i`$. If for the moment we assume that time advances with each member which is given back, we obtain the rate equation
$$\frac{dn(s)}{dt}=(s1)n(s1)ϵn(s)sn(s)ϵp_{\mathrm{𝑖𝑛𝑗}}n(s)+ϵp_{\mathrm{𝑖𝑛𝑗}}n(s+1).$$
(11)
The first and second term on the RHS represent cluster growth by addition of another member; the third term represents random deletion; the fourth and fifth term the decrease by one which happens if one of the members is converted to a start-up via injection. $`ϵ`$ is the rate of cluster deletion; since we first give all members of a deleted cluster back to the population before we delete the next cluster, it is proportional to the inverse of the average cluster size and thus to the injection rate: $`ϵ1/sp_{\mathrm{𝑖𝑛𝑗}}`$. This is similar to an urn process with additional deletion.
Via the typical approximations $`sn(s)(s1)n(s1)\frac{d}{ds}(sn(s))`$ etc. we obtain, for the steady state, the differential equation
$$0=Ns\frac{dN}{ds}ϵN+ϵp_{\mathrm{𝑖𝑛𝑗}}\frac{dN}{ds}.$$
(12)
This leads to
$$n(s)(sϵp_{\mathrm{𝑖𝑛𝑗}})^{(1+ϵ)}s^{(1+ϵ)}.$$
(13)
That is, the exponent depends on the injection rate, and in the limit of $`p_{\mathrm{𝑖𝑛𝑗}}0`$ it goes to $`1`$. This is indeed the result from Sec. 3.4 (see Fig. 4).<sup>4</sup><sup>4</sup>4Note that the approach in this section corresponds to measuring the cluster size distribution every time we give an agent back to the system, while in the simulations we measured the cluster size distribution only just before a cluster was picked for deletion. In how far this is important is an open question; preliminary simulation results indicate that it is important for the spatial case with injection but not important for the non-spatial case in this section.
## 5 Price formation
What we will do now is to add the mechanism of price formation to our spatial competition model. For this, we identify sites with consumers/customers. Clusters correspond to domains of consumers who go to the same shop/company. Intuitively, it is clear how this should work: Companies which are not competitive will go out of business, and their customers will be taken over by the remaining companies. The reduction in the number of companies is balanced by the injection of start-ups. Companies can go out of business for two reasons: losing too much money, or losing too many customers. The first corresponds to a price which is too low; the second corresponds to a price which is too high.
We model these aspects as follows: We again have $`N`$ sites on an $`N=L\times L`$ grid with periodic boundary conditions (torus). On each site, we have a consumer and a firm. These are not connected in any way except by the spatial position – one can imagine that the firm is located “downstairs” while the consumer lives “upstairs”. Firms with customers are called “active”, the other ones “inactive”. A time step consists of the following sub-steps:
* Trades are executed.
* Companies with negative profit go out of business.
* Companies change prices.
* New companies are injected.
* Consumers can change where they shop.
These steps are described in more detail in the following:
Trade: All customers have an initial amount $`M`$ of money, which is completely spent in each time step and replenished in the next. Every customer $`i`$ also knows which firm $`j=f(i)`$ he/she buys from. Thus, he/she orders an amount $`Q_i=M/P_j`$ at his/her company, where $`P_j`$ is that company’s price. The companies produce to order, and then trades are executed. That is, a company that has $`n_j`$ customers and price $`P_j`$ will produce and sell $`Q_j=n_jM/P_j`$ units and will collect $`n_jM`$ units of money.
Company exit: We assume an externally given cost function for production, $`C(Q)`$, which is the same for everybody. If profit $`\mathrm{\Pi }_j:=n_jMC(n_jM/P_j)`$ is less than zero, then the company is losing money and will immediately go out of business.<sup>5</sup><sup>5</sup>5In this model no accumulation of assets is allowed. This simplification will be relaxed in future work. The prices of such a company is set to infinity. We will use $`C(Q)=Q`$, corresponding to a linear cost of production. With this choice, companies with prices $`P_j<1`$ will exit according to this rule as soon as they attract at least one customer.
Price changes: With probability one, pick a random integer number between $`1`$ and $`N`$. If there is an active company with that number, its price is randomly increased or decreased by $`\delta `$.
Company injection: Companies are made active by giving them one customer: With probability $`p_{\mathrm{𝑖𝑛𝑗}}`$, pick a random site $`i`$ and make the consumer $`i`$ go shopping at company $`i`$. The price of the injected company is set to the price that the customer has paid before, randomly increased or decreased by $`\delta `$.
Consumer adaptation: All customers whose prices got increased (either via “company exit” or via “price changes”) will search for a new shop.<sup>6</sup><sup>6</sup>6The simplification that customers react to price changes only is useful because it leads to the separation of time scales between consumer behavior and firm behavior. These “searching” consumers correspond to dead sites in the basic spatial models (Sec. 3), and the dynamics is essentially a translation of that: All searching consumers in parallel pick a random nearest neighbor. If that neighbor is also searching, nothing happens. If that neighbor is however not searching, and if that neighbor is paying a lower price, our consumer will accept the neighbor’s shop. Otherwise the customer will remain with her old shop, and she will no longer search. We keep repeating this until no consumer is searching any more.
This model does not invest much in terms of rational or organized behavior by any of the entities. Firms change prices randomly; and they exit without warning when they lose money. New companies are injected as small variations of existing companies. Consumers only make moves when they cannot avoid it (i.e. their company went out of business and they need a new place to go shopping) or when prices just went up. Only in the last case they actively compare some prices. It will turn out (see below) that even that price comparison is not necessary.
In the above model, price converges to the unit cost of production, which is the competetive price. In Fig. 6 (left, bottom curve) we show how an initially higher price slowly decreases towards a price of one. The reason for this is that, as long as prices are larger than one, there will be companies that, via random changes or injection, have a lower price than their neighbors. Eventually, these neighbors raise prices, thus driving their customers away and to the companies with lower prices. If, however, a company lowers its price below one, then it will immediately exit after it has attracted at least one customer.<sup>7</sup><sup>7</sup>7If *all* prices in the system are more than $`\delta `$ below one, then the model is not well-defined. In the limit of large systems and when starting with prices above one, such a state cannot be reached via the dynamics. – Also note that if the model allowed credit, the exit of such a company would be delayed, allowing losses for limited periods of time.
As already mentioned above, it turns out that the price comparison by the consumers is not needed at all. We can replace the rule “if price goes up, try to find a better price” by “if price goes up, go to a different shop no matter what the price there”. In both cases, we find the alternative shop via our neighbors, as we have done throughout this paper. The top curve in Fig. 6 shows the resulting price adjustment. Clearly, the price still moves towards the critical value of one, but it moves more slowly and the trajectory displays more fluctuations. This is what one would expect, and we think it is typical for the situation: If we reduce the amount of “rationality”, we get slower convergence and larger fluctuations.
In terms of cluster size distribution, the price model is similar to the earlier spatial competition model with random injection. They would become the same if we separated bankruptcy and price changes.
In Fig. 6 (right) we also show that our model is able to track slowly varying costs of production. For this, we replace $`C(Q)=Q`$ by a sinus-function which oscillates around $`Q`$. The plot implies that prices lag behind the costs of production.
This is also visible in the asymmetry of the cross correlation function between both series. In order to be able to compare with non-stationary real world series, we look at relative changes, $`R_x(t)=x(t)/x(t1)`$. The cross correlation function between price increases and cost increases then is
$$\mathrm{𝑋𝑐𝑜𝑟𝑟}(\tau )=R_P(t)R_C(t\tau ),$$
(14)
where $`.`$ averages over all $`t`$. In Fig. 7 (left) one can clearly see that prices are indeed lagging behind costs for our simulations. In order to stress the asymmetry, we also plot $`\mathrm{𝑋𝐶𝑜𝑟𝑟}(\tau )`$. In Fig. 7 (right) we show the same analysis for the Consumer Price Index vs. the Production Price Index (seasonally adjusted). Although the data is much more noisy, it is also clearly asymmetric.
## 6 Discussion and outlook
The modelling approach with respect to economics in this paper is admittedly simplistic. Some obvious and necessary improvements concern credit and bankruptcy (i.e. rules for companies to operate with a negative amount of cash). Instead of those, we want to discuss some issues here that are closer to this paper. These issues are concerned with time, space, and communication.
In this paper, in order to reach a clean model with possible analytic solutions, we have described the models in a language which is rather unnatural with respect to economics. For example, instead of “one company per time step” which changes prices one would use rates (for example a probability of $`p_{\mathrm{𝑐ℎ}}`$ for each company to change prices in a given time step). However, in the limiting case of $`p_{\mathrm{𝑐ℎ}}0`$, at most one and usually zero companies change prices in a given time step. If one also assumes that consumers adaptation is fast enough so that it is always completed before the next price change occurs, then this will result in the same dynamics as our model. Thus, our model is not “different” from reality, but it is a limiting case for the limit of fast customer adaptation and slow company adaptation. Our approach is to understand these limiting cases first before we move to the more general cases.
Similar comments refer to the utilization of space. We have already seen that moving from a spatial to a non-spatial model is rather straightforward. There is an even more systematic way to make this transition, which is the increase of the dimensions. In two dimensions on a square grid, every agent has four nearest neighbors. In three dimensions, there are six nearest neighbors. In general, if $`D`$ is the dimension, there are $`2D`$ nearest neighbors. If we leave the number $`N`$ of agents constant and keep periodic boundary conditions ($`D`$-dimensional torus), then at $`D=(N1)/2`$ everybody is a nearest neighbor of everybody. Thus, a non-spatial model is the $`D\mathrm{}`$ limiting case of a spatial model.<sup>8</sup><sup>8</sup>8Furthermore, models such as the ones discussed in this paper often have a so-called upper critical dimension, where some aspects of the model become the same as in infinite dimensions. This upper critical dimension often is rather low (below 10).
These concepts can be generalized beyond grids and nearest neighbors – the only two ingredients one needs is that (i) the probability to interact with someone else decreases fast enough with distance, and that (ii) if one doubles distance from $`r`$ to $`2r`$, then the number of interactions up to $`2r`$ is $`2^D`$ times the number of interactions up to distance $`r`$.
This should also make clear that space can be seen in a generalized way if one replaces distance by generalized cost. For example, how many more people can you call for “20 cents a minute or less” than for “10 cents a minute or less”? If the answer to this is “two times as many”, then for the purposes of this discussion you live in a one-dimensional world.
Given this, it is important to note that we have used space only for the communication structure, i.e. the way consumers aquire information (by asking neighbors). This is a rather weak influence of space, as opposed to, for example, transportation costs; it however also assumes a not very sophisticated information structure, as for example contrast to today’s internet. The details of this need to be left to future work.
Last, one needs to consider which part of the economy one wants to model. For example, a stockmarket is a centralized institution, and space plays a weak role at best. In contrast, we had the retail market in mind when we developed the models of this paper. In fact, we implicitely assume perishable goods, since agents have no memory of what they bought and consumed the day before. Also, we assume that consumers spend little effort in selecting the “right” place to shop, which excludes major personal investments such as cars or furniture. Also note that our companies have no fixed costs, which implies that there are no capital investments, which excludes for example most manufacturing.
## 7 Summary
Price formation is an important aspect of economic activity. Our interest was in price formation in “everyday” situations, such as for retail prices. For this, we assumed that companies are price setters and agents are price takers, in the sense that their only strategy option is to go someplace else. In our abstracted situation, this means that companies with too low prices will exit because they cannot cover costs, while companies with too high prices will exit because they lose their customers.
We use space in order to simplify and structure the way in which information about alternative shopping places is found. This prevents the singularity of “Bertrand-style” models, where the market share of each company is independent from history, leading to potentially huge and unrealistic fluctuations.
By doing this, one notices that the spatial dynamics can be separated from the price formation dynamics itself. This makes intuitively sense since, in generalized terms, we are dealing with evolutionary dynamics, which often does not depend on the details of the particular fitness function. We have therefore started with an investigation of a spatial competition model without prices. For this model, we have looked at cluster size distributions, and compared them with real world company size distributions. In contrast to investigations in the literature, which find log-normal distributions, we find a scaling law a better fit of our data. In the models, we find log-normal distributions or scaling laws, depending on the particular rules.
We then added price formation to our spatial model. We showed that the price, in simple scenarios, converges towards the competitive price (which is here the unit cost of production), and that it is able to track slowly varying production costs, as it should. This predicts that prices should lag behind costs of production. We indeed find this in the data of consumer price index vs. production price index for the United States since 1941.
## Acknowledgments
KN thanks Niels Bohr Institute for hospitality during the summer 1999, where this work was started. All of the authors thank Santa Fe Institute, where some of the authors met, which provided a platform for continuous discussion, and where some of the work was done in spring 2000. We also thank H. Flyvbjerg and K. Sneppen for invaluable hints and discussions.
## Appendix A Converting the aggregated census data
#### Non-equidistant bins
The size data in the 1992 U.S. economic census comes in non-equidistant bins. For example, we obtain the number of establishments with annual sales above 25 000 k$, between 10 000 k$ and 25 000 k$, etc. For an accumulated function, such as Fig. 5 (right), this is straightforward to use. For distributions, such as Fig. 5 (left), this needs to be normalized. We have done this in the following way: (1) We first divide by the weight of each bin, which is its width. In the above example, we would divide by $`(\mathrm{25\hspace{0.17em}000}k\$\mathrm{10\hspace{0.17em}000}k\$)=\mathrm{15\hspace{0.17em}000}k\$`$. Note that this immediately implies that we cannot use the data for the largest companies since we do not know where that bin ends. (2) For the log-normal distribution
$$\rho (x)\frac{1}{x}\mathrm{exp}\left[(\mathrm{ln}(x)\mathrm{ln}(\mu ))^2\right]$$
(15)
(note the factor $`1/x`$), one typically uses logarithmic bins, since then the factor $`1/x`$ cancels out. This corresponds to a weight of $`x`$ of each census data point. (3) Now we have to decide where we plot the data for a specific bin. We used the arithmic mean between the lower and the upper end. In our example case, $`\mathrm{17\hspace{0.17em}500}k\$`$. (4) In summary, say the number of establishments between $`s_i`$ and $`s_{i+1}`$ is $`N_i`$. Then the transformed number $`\stackrel{~}{N}_i`$ is calculated according to
$$\stackrel{~}{N}_i=\frac{N_i}{s_{i+1}s_i}\frac{s_i+s_{i+1}}{2}.$$
(16)
#### The largest firms
For the largest firms (but not for the large establishments), the census also gives the combined sales of the four (eight, twenty, fifty) largest firms. We used the combined sales of the four largest firms divided by four as a (bad) proxy for the sales of each of these four companies. We then substracted the sales of the four largest firms from the sales of the eight largest firms, divided again, etc. Those data points should thus be seen as an indication only, and it probably explains the “kink” near $`2\times 10^9`$ in Fig. 5.
|
warning/0005/astro-ph0005035.html
|
ar5iv
|
text
|
# ISO observations of a sample of Compact Steep Spectrum and GHz Peaked Spectrum Radio Galaxies
## 1 Introduction
Compact Steep Spectrum (CSS) and GHz Peaked Spectrum (GPS) radio sources are powerful extragalactic radio sources with radio emission confined well within their host galaxy/quasar ($`<15`$kpc)<sup>2</sup><sup>2</sup>2Throughout this paper we use H<sub>0</sub>=100 (Km/sec)/Mpc and q<sub>0</sub>=0.5.. They are as powerful as the classical FRII radio sources but are of much smaller physical size and yet have normal/steep radio spectra at GHz frequencies. A discussion of the properties of CSS can be found in Fanti et al. (fanti (1990); fanti2 (1995)). A comprehensive and updated review on CSS and GPS is presented by O’Dea (odea3 (1998)).
Some 15 to 25 % of sources in a flux density limited sample, depending on the selection frequency, belong to this class. Many of them have radio spectra which show a flattening or a more marked turnover at frequencies between 50 MHz and a few GHz; those sources with a turnover at about 1 GHz are the GPS sources. The turnover is usually attributed to synchrotron self–absorption, but also free–free absorption (Bicknell et al. bicknell (1997)) and induced Compton scattering (Kuncic et al. kuncic (1998)) have been considered.
Given that CSS and GPS sources are physically small, i.e. of sub–galactic dimensions, it has been suggested that they are either (1) young objects (youth scenario, Phillips & Mutel phillips (1982)), which have yet to develop extended radio lobes or (2) sources where the radio emitting plasma is trapped by an unusually dense interstellar medium (frustration scenario, van Breugel et al. vanbreugel (1984)).
In the latter hypothesis, CSS/GPS radio sources would be as old as those of larger size but their jets would spend their whole lifetime trying to escape, without success, out of the interstellar medium. This scenario requires a rather dense interstellar medium (average density $`n_{\mathrm{ISM}}1`$ cm<sup>-3</sup> and total mass within 1 kpc $`M_{\mathrm{ISM}}>10^8M_{}`$; see De Young deyoung (1993); Fanti et al. fanti2 (1995)) which is in the path of the radio source. Even if the host galaxy contains a lot of gas, if this is distributed in a disk perpendicular to the radio source axis, it will have little effect on the radio source propagation, therefore any “frustrating” gas has to be distributed over a large volume. Attempts to probe the different phases of this interstellar gas (e.g. optical observations of the Narrow Lines, polarization studies, X–ray emission) have given, in our opinion, little support to the frustration scenario.
Recently, proper motion measurements of the hot spots in a few GPS sources by Owsianik et al. (owsianik (1998)) and Owsianik & Conway (owsianik2 (1998)) have shown separation velocities $`0.1c`$. This finding strongly suggests that these sources are very young. Also a recent study of the integrated radio spectra of a number of CSS/GPS (Murgia et al. murgia (1999)) provides additional evidence for short lifetimes.
It is nevertheless important to support these recent results with information on the ambient medium around these sources.
The frustration scenario requires that an unusually dense ISM be present in CSS/GPS sources. This may contain a substantial cold phase and a large amount of dust, and, in order to stop the advancing jet, this medium has to have a large covering factor. The dust will therefore absorb and re-process a fraction of the optical and UV radiation from the AGN (and young stars, if present) higher than in the larger size, “non–frustrated”, radio sources, giving to the CSS/GPS sources an extra IR emission component in addition to the one from the disk/torus.
Since this dense medium has to extend over several kpc from the nucleus (in order to stop also radio sources 10–15 kpc in size), a large fraction of the dust will be moderately cold so that the emission is expected predominantly in the medium–far IR (MFIR) spectral region.
By searching for emission from cold dust, we have therefore tried to obtain some constraints on this scenario. Since the emission peak is expected at relatively long wavelengths, ISO, with its capability of carrying out photometry at $`\lambda 60\mu `$m, (Kessler et al. kessler (1996)) was well suitable for such an investigation.
This search is better suited to radio galaxies, rather than quasars, as in the former the contamination from the IR emission produced in the innermost regions by the AGN is lower by a factor 4–5 (Heckman et al. heckman2 (1992)). No definitive explanation, within the “unified models” scenario, has yet been given for such a difference (see also Sect. 6.5), but it is an observational effect that we have taken into consideration in planning the experiment. Moreover there are several ISO programs in the Guaranteed Time allocation aimed at observing different samples of quasars, including a number of CSS/GPS, from which the relevant information could be retrieved.
The project of which we present the results is the merging of two independent programs (P.I.s C.F. and H.F.) which both got observing time in the first ISO Call for Proposals. A representative sample of CSS/GPS radio galaxies, unbiased with respect to FIR emission, was selected from various samples of CSS/GPS radio sources. A sample of extended radio galaxies, matched in redshift and radio luminosity, was also selected for observations at the same wavelengths, in order to determine the level of MFIR emission in extended radio galaxies as compared to CSS/GPS galaxies.
The layout of the paper is the following.
Sect. 2 describes the selection of the two samples.
Sects. 3 and 4 describe the ISO observations and the data reduction. In these sections we mention only what is relevant to the paper. For more details we refer the reader to the specific ISO and PIA<sup>3</sup><sup>3</sup>3PIA is a joint development by the ESA Astrophysics Division and the ISOPHOT Consortium led by the Max Planck Institute for Astronomy (MPIA), Heidelberg. Contributing ISOPHOT Consortium institutes are DIAS, RAL, AIP, MPIK, and MPIA. (IsoPhot Interactive Analysis) literature.
Sects. 5 and 6 present the results and discuss them.
Conclusions are summarized in bf Sect. 7
## 2 The sample
Eighteen CSS/GPS radio galaxies in the redshift range $`0.2z0.8`$ and with radio size $`<10`$ kpc were originally selected. They are mostly from the 3CR catalogue at 178 MHz (Laing et al. laing (1983)) and from the Peacock & Wall (peacock (1982)) (PW) catalogue of radio sources at 2.7 GHz, (see Fanti et al. fanti2 (1995)), but some have been selected also from other lists provided the selection was unbiased with respect to IR emission (see Table 1).
Of these radio sources, 3C318 has recently been discovered to be a quasar at a redshift higher than originally believed (Willott et al. willott (2000)) and therefore it has to be dropped from the sample, which counts then 17 CSS/GPS galaxies. However, for completeness, we have included in Table 5 also the data we obtained on this source.
A sample of 16 3CR radio galaxies, with radio sizes $`>`$ 20 kpc and spanning a similar range in redshift and radio luminosity, was selected for comparison purposes.
The two samples are presented in Tables 1 & 2.
Four additional GPSs, at redshifts $`<0.2`$ were also selected for observations (see Sect. 5.3), although they are not going to be used in the discussion. They also are listed in Table 1, separate from the representative sample.
## 3 Observations
The sources were observed with ISOPHOT (Lemke et al. lemke (1996)), sub–instruments C100 & C200, at the wavelengths of 60, 90, 174 & 200 $`\mu `$m. These wavelengths were chosen in order to cover a spectral range as broad as possible with the best detectors. Actually some sources, mainly in the control sample, were observed in other programmes, including Guaranteed Time Observations, requiring us to observe them only at the wavelengths not planned in the other experiments (see Tables 1 & 2, where the wavelengths missing in our programme are marked by a “—”).
The C100 detector consists of a 3$`\times `$3 matrix of pixels $`43.5^{\prime \prime }\times 43.5^{\prime \prime }`$ in size and C200 of a 2$`\times `$2 matrix of $`89.4^{\prime \prime }\times 89.4^{\prime \prime }`$ pixels. Each pixel is an independent detector, which requires its own calibration.
The diameter of the Airy disks of the Point Spread Functions (PSF) are given in Table 3. By comparison with the pixel sizes, it is clear that for C100 the Airy disk matches approximately one pixel size, while for C200 the PSF covers most of the matrix. Therefore detected sources will be visible mostly on the central pixel #5 with C100, while with C200 the four pixels should give approximately the same values (within the noise). Table 3 gives also the fraction, $`f_{\mathrm{PSF}}`$, of light falling onto the central pixel (# 5) for C100 and on the four pixels for C200.
Observations were made in chopper rectangular mode, with chopper throw 180”, in order to have, every few seconds (chopper plateau), an ON–source and an OFF–source measure of the same time length. The OFF data then have to be subtracted from the ON data, in order to extract the source signal. In this observing mode, the telescope points halfway between the source and the background positions, and a small mirror switches between the two sky positions. This introduces a vignetting error which is different for ON and OFF positions (see Sect. 4).
We applied for and obtained time also in the second “Call for Proposals”, mainly for the purpose of increasing the observing time (originally planned on the basis of the ground estimates of the instrument performances). Therefore some of the sources were observed twice.
The total ON–source time ranges from 32 to 256 sec, although the glitch removal and other options of the data reduction (Sect. 4) have shortened the nominal observing time. In Tables 1 & 2 however, we give the total “nominal” ON–source time; for the sources observed twice we give the total.
## 4 Data reduction
The data analysis was carried on using mainly the PIA V7.2 software (Gabriel et al. gabriel (1997)). Some further software has been kindly made available to us by Dr. M. Haas from MPIA (Heidelberg) or has been written by ourselves.
PIA removes glitches due to cosmic rays, subtracts the dark current, corrects for drifts, derives the signals from the ramps<sup>4</sup><sup>4</sup>4The signal is read from the detector every 1/32 sec and accumulated until the voltage reaches a saturation limit. At this time a “destructive” read–out occurs, which resets the integration. The integral signal between two consecutive destructive read–outs is called a “ramp” and its slope gives the signal in Volt/sec., calibrates the data and makes corrections for vignetting.
Special PIA features that we used are:
1) We used the method of ramp subdivision, i.e. we divided each ramp into pieces 0.25 sec long (pseudo–ramps) to improve the accuracy of signal determination.
2) to remove glitches we used the “two threshold glitch recognition” which is claimed to work better than the original one. In a number of cases, however, glitch “tails” remain in the data. We wrote a simple IDL program to further statistically remove the glitch residuals. The noise is reduced by approximately a further 20%.
3) Due to “memory effects” the chopper plateaus are not always flat, but in many cases show upwards/downwards trends in the ON/OFF measurements. In absence of an appropriate program which fits these trends and extrapolates the asymptotic value, we used the quite crude PIA option which removes the first half of each chopper plateau, thus cutting in half the observing time.
More detailed remarks on some other of these reduction steps are discussed here below.
### 4.1 Calibration
The conversion Volt/sec into Watt is made by using an internal calibration source which provides the conversion factors for each of the matrix pixels (or detectors). Then the conversion to Jy is made by using the known instrument and filter characteristics. The flux density scale may still be $`<`$ 20% uncertain at 60 and 90 $`\mu `$m (see Sect. 5.2) and perhaps more at 174 and 200 $`\mu `$m.
In addition, the matrices of the OFF measurements are not “flat” after PIA calibration, the pixel–to–pixel fluctuations being much larger than the instrumental noise. This is explained as a residual calibration error of the individual pixels (flat fielding), that we estimate to be up to $`\pm 15`$%. This causes fluctuations from pixel to pixel of several tens of mJy, essentially due to the background, as the “true” signal from the sources is rather small. These errors, however, largely cancel out in the (ON – OFF) data and we have not attempted any correction.
Finally, according to the ISO team, it seems that flux densities have to be increased by an amount which depends on the chopper frequency (i.e. the rate at which the chopper switches between ON and OFF positions) and on the signal difference between ON and OFF. At the time of writing this paper this correction has not been firmly established yet, although it is believed to be small for the C200 data. For the C100 data only the strong sources may be affected by this problem (see also Sect. 5.2), while for the average detections (see Sect. 5.4) we expect the correction to be negligible.
### 4.2 Vignetting corrections
The vignetting corrections applied by PIA have an uncertainty of a few percent. As they are different in the ON and OFF measures, they do not cancel in computing (ON – OFF), as residual flat–field errors do. Given the typical values of the background (Table 4), this uncertainty introduces errors which are not negligible as compared to the source flux densities which are rather weak.
At 60 and 90 $`\mu `$m, pixel # 5, where the source signal is largely concentrated, is not affected by vignetting and the source flux density is safely derived as explained in Sect. 4.3 by dividing the signal on pixel # 5, by the $`f_{\mathrm{PSF}}`$.
At 174 and 200 $`\mu `$m, where the source signal is spread over the four pixels, vignetting errors are a major cause of uncertainty, due to the higher background levels, not only as they increase the final noise, but also as they introduce systematic effects. Indeed, after application of PIA vignetting corrections, we got the peculiar result that the (ON – OFF) flux densities were, on average, systematically negative, with a clear dependence on the background brightness. This effect is very likely due to residual vignetting errors.
We have therefore tried to estimate such corrections independently, from our own data. For this purpose we used a method suggested by Dr. M. Haas, from MPIA, whose application we take the responsibility for. We assumed that the pixels which are more central with respect to the telescope pointing, i.e. # 2 and #3 at ON, and # 1 and #4 at OFF position, are, at the first order, unaffected by vignetting, while the other pixels are.
As the PSF fills essentially all four pixels, they do see the same signal. Any deviation from this is attributed to vignetting. Considering only the sources with very strong background ($`>`$20 MJy/ster), assuming that the source signal is negligible as compared to the background, and that the background does not change appreciably from ON to OFF, the four pixels at ON position ought to see approximately the same signal as the four pixels at OFF. Therefore the ratios of the signal in pixels # 1 and # 4 ON (assumed to be the only ones affected by vignetting) to the signal of the corresponding pixels at OFF position (assumed to be unaffected by vignetting) give the vignetting errors for these two pixels. The same procedure is applied to pixels \# 2 and #3 at OFF position. Note that flat–fielding errors cancel out in taking the ratios of corresponding pixels.
The correction factors deduced from the different sources are consistently in the range from 3 % to 5 % and were applied to all the sources. The estimated residual uncertainty is $``$ 1.0 %. The application of these corrections eliminates the problem of the systematic negative (ON – OFF) values and of their dependence on the background brightness.
### 4.3 Flux density determination
For C100 the source flux density should be derived by fitting a PSF of known width to the (ON – OFF) matrix values. In practice our signal is always very weak and will only be detectable on pixel #5. Therefore we have computed the source flux density simply by dividing the flux density falling onto pixel #5 by the appropriate value of $`f_{\mathrm{PSF}}`$ given in Table 3.
For C200 the flux density has been obtained by summing the (ON–OFF) four matrix pixels and dividing the result by the $`f_{\mathrm{PSF}}`$ in Table 3.
In the following we shall therefore use as units mJy/pixel to refer to the flux density falling onto one single pixel and mJy to indicate source flux density.
### 4.4 Estimate of statistical flux density errors
Besides the calibration and vignetting errors, additional uncertainty in the flux density measurements is caused by: a) instrumental noise (or simply noise, $`\sigma _\mathrm{n}`$), which includes all noise contributions along the signal chain (i.e. photon noise due to source and background, plus detector noise), b) confusion due to foreground galactic cirrus ( $`\sigma _{\mathrm{cc}}`$) and c) confusion due to extragalactic background ($`\sigma _{\mathrm{egc}}`$). Typical values are reported in Table 4.
In the next paragraphs we discuss how they have been evaluated in the paper.
a) Instrumental noise, $`\sigma _\mathrm{n}`$
The nominal instrumental noise per pixel is given by the equation:
$$\sigma _\mathrm{n}^{}=\sqrt{NEP_{\mathrm{source}}^2+2NEP_{\mathrm{bck}}^2+2NEP_{\mathrm{rec}}^2}$$
(1)
where the $`NEP`$’s are the Noise Equivalent Power of source, background and receiver, to be computed using parameters given in the ISO manuals and which depend on the ON–source time. The factors of 2 in Eq. (1) derive from the fact that the signal is obtained from (ON – OFF) measurements. ¿From the values of the parameters necessary to compute the $`NEP`$’s it is clear that, except for very high backgrounds ($`>100`$ MJy/ster) or for very strong sources ($`>`$ 50–100 Jy), the contribution of the detector to the nominal noise is expected to dominate.
Because the resulting performance of the instrument was worse than expected, we adopted a pragmatic method and estimated $`\sigma _\mathrm{n}^{}`$ from the data themselves, instead of using the parameters given in the manuals.
To estimate $`\sigma _\mathrm{n}^{}`$ directly from the data, within each observation we used the chopper sequence (i.e. the sequence of ON–OFF pairs of each source) to determine the statistics of the (ON – OFF) values. For C100 we used only pixel # 5, thus avoiding the residual flat–field (Sect. 4.1) and vignetting (Sect. 4.2) problems. For C200 we used the four pixels together, after application of vignetting corrections, and ignoring the residual flat–field corrections.
The average (ON – OFF) gives the flux density within pixel # 5 for C100 and the flux density within any of the four pixels for C200. The r.m.s. distribution of the ON–OFF pairs gives the “actual instrumental noise” $`\sigma _\mathrm{n}^{}`$ (in mJy/pixel) of that one observation. The software for this analysis is due to M.Haas (private communication).
We find that $`\sigma _\mathrm{n}^{}`$ can change significantly from source to source, even for similar backgrounds and observing times. The median values, however, scale rather well with the inverse of the square root of the ON–source time. We do not find any difference as a function of the background brightness, which confirms that the detector noise is the dominant factor. The noise computed in this way is in agreement with that from Eq. 1 when adopting the parameters of the ISO manual, except for $`\lambda =200\mu `$m where our estimate is lower. This proves, to us, that no significant systematic error is present in our data and that glitches have been removed fairly well.
Given the good behavior of the instrumental (mostly detector’s) noise with $`1/\sqrt{t}`$, we give in Table 4, for each observing wavelength, one single value for $`\sigma _\mathrm{n}^{}`$, corresponding to an integration time of 64 sec. Note that with C200, since the four pixels are independent detectors, the flux density error due to the noise will be computed as $`\sigma _\mathrm{n}=2\times \sigma _\mathrm{n}^{}/f_{\mathrm{PSF}}`$.
b) Cirrus confusion noise, $`\sigma _{\mathrm{cc}}`$
This term has been evaluated by Gautier et al. (gautier (1992)) who provide useful tables to compute $`\sigma _{\mathrm{cc}}`$ on any angular scales (3 arcmin, in our case, the value of the chopper throw) and also show that cirrus confusion depends strongly on the galactic background brightness, $`B`$, according to $`\sigma _{\mathrm{cc}}B^{1.5\pm 0.2}`$.
Since our measurements include a contribution from the celestial fore/back–grounds and, at least at 60 and 90 $`\mu m`$, a strong contribution from zodiacal light, in order to estimate, $`\sigma _{\mathrm{cc}}^{}`$ (within a pixel), at all four wavelengths, for both C100 and C200 we used IRSKY software by IPAC which applies a model to remove the zodiacal light, providing also the values for a pure galactic foreground. This estimate of $`\sigma _{\mathrm{cc}}^{}`$ is accurate to within a factor of two.
Note that, for $`\lambda >100\mu `$m, the foreground provided by IRSKY is extrapolated from the IRAS shorter wavelengths, using the dust emission model by Desert et al. (desert (1990)). This increases the uncertainty on $`\sigma _{\mathrm{cc}}^{}`$. For consistency and to be conservative we generally used the IRAS background, which is, on average, higher than ISO’s, except in those cases in which, by comparison with the other satellite values, the extrapolation from IRAS was obviously wrong. In those cases we used the ISO background.
We have also compared IRSKY values of $`\sigma _{\mathrm{cc}}^{}`$ with those reported in ISO manuals: while at 60 and 90 $`\mu `$m the two estimates do agree, at $`\lambda =174\mu `$m the ISO manual estimate is over one order of magnitude greater than IRSKY’s. We found no estimate at 200$`\mu `$m.
As a reference, we report in Table 4 the values of $`\sigma _{\mathrm{cc}}^{}`$ at each wavelength, from IRSKY, for a foreground brightness typical of our observations.
c) Extragalactic confusion noise, $`\sigma _{\mathrm{egc}}`$
This is simply due to the piling up, within a beam, of faint galaxies not individually resolved by the instrument. The estimate of the magnitude of this effect, within each pixel, is reported in Table 4 and is derived by IRSKY, in the hypothesis of cosmological evolution of faint galaxies. These values are about a factor of two higher than in the case of a non–evolving model.
Note that with C200, since the PSF covers most of the four pixels, we considered the cirrus and extragalactic confusion correlated across the matrix and therefore we adopted, as the flux density error, $`4\times `$(pixel error)/$`f_{\mathrm{PSF}}`$.
The total error on source flux density is then computed as:
$$\sigma {}_{\mathrm{T}}{}^{}=\sqrt{\sigma _\mathrm{n}^2+\sigma _{\mathrm{egc}}^2+\sigma _{\mathrm{cc}}^2}$$
(2)
where $`\sigma _\mathrm{n}`$ is the individual source noise, $`\sigma _{\mathrm{egc}}`$ is the error due to the extragalactic confusion (which is the same in all fields at a given wavelength) and $`\sigma _{\mathrm{cc}}`$ the cirrus confusion noise appropriate for the specific source background (zodiacal light subtracted). In Table 4 total flux density errors, appropriate to the parameters in the Table, are also given as an example.
As it can be seen from Table 4, the extragalactic source and cirrus confusion ($`\sigma _{\mathrm{egc}}`$ and $`\sigma _{\mathrm{cc}}`$) may be not negligible as compared to the instrumental noise and may limit the instrument sensitivity considerably or lead to spurious detections, as it will be further discussed in Sect. 5.1 .
As an internal test, we used those sources observed twice (with different backgrounds) to check the goodness of error evaluation.
At each wavelength, we compared the difference $`\mathrm{\Delta }S`$ of the two measured flux densities with the combined total error ($`\sigma _\mathrm{T}^c`$) obtained by adding in quadrature the $`\sigma _\mathrm{T}`$ of the two observations. If errors are estimated properly, the ratio $`\mathrm{\Delta }S/\sigma _\mathrm{T}^c`$, computed for all the available pairs, should have zero average and standard deviation =1.
In spite of the very poor statistics (8 sources observed twice with C100 and 4 with C200) the agreement is satisfactory.
## 5 Results
### 5.1 Individual Detections
About half of the observed radio galaxies, both in the representative sample of CSS/GPS and in the comparison sample, have $`S>3\times \sigma _\mathrm{n}`$, at least at one wavelength. These could be considered “formal” detections, in the sense that we got some signal out of the instrumental noise. This does not mean, however, that we have detected our target sources, since, as seen in Table 5, extragalactic and/or cirrus confusion could cause spurious detections: we might just have detected a cirrus filament or a fluctuation in the confusion.
We have checked the reality of these “formal” detections by looking for “negative” detections. We have only three such cases (see Tables 5 and 6) suggesting that, after all, in the “formal” detections some signal from the target is present. This is why we have kept them in Table 5. The test for “claiming” a detection, however, is made by comparing $`S`$ with $`\sigma _\mathrm{T}`$ (Eq. 2).
We considered detected the sources for which $`S5\times \sigma _\mathrm{T}`$ at least at one wavelength (value in bold in Table 5) and possibly detected those for which $`3\times \sigma _\mathrm{T}S<5\times \sigma _\mathrm{T}`$ (value in italic in Table 5).
Table 5 gives the detected, possibly detected and “formally” detected sources, listing the flux densities at all four wavelengths (significant or not) and related errors. The sources detected at least at one wavelength are marked with $``$; those possibly detected are marked with $``$. In each column we give $`S,\sigma _\mathrm{n}`$ and $`\sigma _\mathrm{T}`$ in mJy (1 $`\sigma `$ level).
In Table 5 only two sources from the representative (1819+39) and control (3C459) samples are clearly detected at least at one wavelength. Nine more sources have flux densities which exceed $`3\sigma _\mathrm{T}`$, but are $`<5\sigma _\mathrm{T}`$, and are therefore possible detections.
The quasar 3C318 is detected at a $`3\sigma `$ level at both 60 and 90 $`\mu `$m.
We also detect clearly at all four wavelengths two of the four nearby GPS (1345+125 and 1718–64).
In Table 6 we list the undetected sources with their $`S`$, $`\sigma _\mathrm{n}`$ and $`\sigma _\mathrm{T}`$, in mJy.
### 5.2 Comparison with IRAS data
As stated in the IRAS Explanatory Supplement (1985) the sensitivity limits (3$`\sigma `$) of the all–sky survey for point sources (PSC) are quite high ($``$0.5 Jy at 60 $`\mu `$m and 1.5 Jy at 100$`\mu `$m) but can be sensibly reduced by co–adding several scans, as done, for instance by Heckman et al. (heckman2 (1992); heckman3 (1994)), who obtained 3$`\sigma `$ detections on individual objects of $``$100 and $``$400 mJy at 60 $`\mu `$m and 100$`\mu `$m respectively. The most sensitive IRAS observations were obtained by pointed observations (see e.g. Neugebauer et al. neugebauer (1986); Golombek et al. golombek (1988)). The resulting noise varies from field to field and is typically in the range of 50–100 mJy ($`3\sigma `$) at 60$`\mu `$m and 100–300 mJy ($`3\sigma `$) at 100$`\mu `$m. These values have to be compared with our typical 3$`\sigma `$ errors, given in Tables 5 and 6, of $``$90 mJy at 60 $`\mu `$m and $``$75 mJy at 90$`\mu `$m (median values). Therefore, while at $`\lambda =60\mu `$m ISO is not much more sensitive than IRAS, at $`\lambda =90\mu `$m the situation is improved.
We searched the literature to see whether any of the sources in our sample has been detected by IRAS at any wavelength. The galaxies from our lists detected by IRAS at 60 and/or 100 $`\mu `$m are reported in Table 7.
On the strongest source (1345+125) about 30% of the IRAS flux density is missing in our observations. This is the case where chopper frequency corrections, that we have not applied (Sect. 4.3), are expected to be larger. For 3C459, which is not as strong, the agreement with the IRAS data is rather good. We are therefore confident that the missing correction is definitely minor for the fainter sources and negligible for the median flux densities that we report. For the fainter sources, we note that 3C79 is possibly detected in our observations only at wavelengths longer than 100$`\mu `$m and that 3C318 at $`90\mu `$m is fainter than in IRAS, although the two flux densities are still within $``$ 2.5 of the combined errors.
Conversely 3C284 possibly detected (3.8 $`\sigma `$ level) at 60 $`\mu `$m in our ISO observations, has only an upper limit in IRAS (Impey & Gregorini impey (1993)) not too different from our measurement. The remaining sources detected (1819+39) or possibly detected at 60 and/or 90 $`\mu `$m by ISO, are weaker than the IRAS upper limits (Heckman et al. heckman3 (1994)).
### 5.3 Comments on some individual sources
1345+125.
This radio source is identified with a galaxy with “two nuclei”, of which the western one has an optical spectrum consistent with Narrow Line radio galaxies and Seyfert type II (Gilmore & Shaw gilmore (1986)). There is some confusion in the literature about which of the two hosts the radio source. Gilmore & Shaw (gilmore (1986)) and Baum et al. (baum (1988)) indicated the eastern (non Seyfert nucleus). More recently, however, Capetti (1998, private communication), Stanghellini et al. (stanghellini2 (1998)) and Evans et al. (evans (1999)) presented convincing evidence for the association of the GPS source with the western Seyfert nucleus. This double nucleus galaxy is considered a merger which has triggered the radio activity (e.g. Heckman et al. heckman (1986); Sanders et al. sanders (1988); Surace et al. surace (1998)). Although the optical spectrum is characterized by Narrow Lines, broad Pa$`\alpha `$ (FWHM $`2600`$ km s<sup>-1</sup>) is seen in the near-IR (Veilleux et al. veilleux (1997)). In addition, Hurt et al. (hurt (1999)) report HST/FOC ultraviolet observations which show extended polarized UV continuum light. Both the UV polarization and the broad Pa$`\alpha `$ indicate that 1345+125 has a hidden (but not too deeply hidden) quasar. This is the first GPS radio galaxy detected in X-rays ($`L_{\mathrm{X}(210\mathrm{K}\mathrm{e}\mathrm{v}})2\times 10^{43}`$erg s<sup>-1</sup>, O’Dea et al. odea4 (2000)).
We estimate a temperature of $`30`$ K and a dust mass of $`3.5\times 10^8`$ M for the component which makes the dominant contribution to our ISO data (Table 10). Although our ISO observations are not sensitive to gas at temperatures below about 25 K, the CO observations are sensitive to gas with kinetic temperatures to within a factor of a few of the microwave background temperature (e.g. O’Dea et al. odea (1994); Allen et al. allen (1995)). For a standard Galactic gas/dust ratio of 100 this corresponds to a mass of molecular gas $`3.5\times 10^{10}`$ M which is in good agreement with the estimate derived from CO observations ($`3\times 10^{10}`$ M, Mirabel et al. mirabel (1989); Evans et al. evans (1999)). This agreement between the ISO and CO estimates of the molecular gas mass may be fortuitous since the gas-to-dust ratio may vary between galaxies and in general is not well determined (e.g. Goudfrooij et al. goudfrooij (1994); Falco et al. falco (1999)).
3C 318
This object was classified as an N–galaxy at a redshift 0.75 by Spinrad & Smith (spinrad (1976)) and considered to be a Narrow Line (NL) radio galaxy by Hes et al. hes (1995). Recently Willott et al. (willott (2000)), from UKIRT spectroscopic observations, detected broad H<sub>α</sub> and revised the redshift at 1.574. The object is now classified as a quasar. It is a strong X–ray emitter (Taylor et al., taylor (1992)) with a luminosity, updated for the new redshift, of $`L_{\mathrm{X}(0.54.5\mathrm{Kev}})8\times 10^{44}`$erg s<sup>-1</sup>,
1622+663:
This object has a broad a H<sub>α</sub> line. It can be classified as a Seyfert 1.5 (Snellen et al. snellen2 (1999)). Its broad–line region seems to be heavily obscured. It is marginally detected at 90 $`\mu `$m.
1718–649
The source is identified with the 12.6 blue magnitude galaxy NGC 6328 (Savage savage (1976)), whose nucleus has a LINER type optical spectrum (Fosbury et al. fosbury (1977)). The galaxy is quite peculiar, having the appearance of a high luminosity elliptical with faint outer spiral structure and containing a large amount of atomic hydrogen (Veron-Cetty et al. veron (1995)). The object is regarded as a merger of two galaxies, at least one of which is a gas-rich spiral, in the process of forming an elliptical. A very strong source (a star?) is listed in the IRAS PSC, at 3.7 arcmin from 1718–649. Inspection of IRAS maps at 60 and 100 $`\mu `$m shows a fainter extension of this bright source to the North, at the position of 1718–649.
The dominant component of dust contributing to our ISO detection has an estimated temperature of 20 K and a mass of $`0.1\times 10^8`$ M (Table 10) which would correspond to a gas mass of $`10^9`$ M. This is at least an order of magnitude less than the mass of HI ($`3.1\times 10^{10}`$ M, Veron-Cetty et al. veron (1995)). The HI is distributed over the entire galaxy and much of it lies in a partial ring with a diameter of $`95\mathrm{"}`$ (19 kpc). Possibly, much of the dust in this object lies at large distances from the nucleus and is relatively cold, thus escaping detection in our ISO observations. CO observations would be useful.
1819+39
Its redshift has been re–measured by Vermeulen et al. (vermeulen (1996)) who also show that this object has broad H<sub>β</sub>. Thus, this object appears to be Broad Line radio galaxy.
1934–638
The radio source is associated with the brighter galaxy of a pair of compact galaxies ($``$ 8 kpc separation) sharing a common envelope (Jauncey et al. jauncey (1986)). Fosbury et al. (fosbury2 (1987)) suggest larger amounts of gas and dust than generally found in typical Narrow Line radio galaxies. Tadhunter et al. (tadhunter (1994)) found presence of polarized light, suggesting an anomalous environment (see also Morganti et al. morganti (1997)). It is undetected at the only observed wavelength (200 $`\mu `$m).
1946+70
This is one of the three GPS sources from the sample of Snellen et al. (snellen (1998)), and it is much weaker than those from the 3CR and PW samples. It is “formally” detected at 200$`\mu `$m.
3C459
The optical object has been classified as N–galaxy by Spinrad et al. (spinrad2 (1985)). It was studied in detail at radio and optical frequencies by Ulvestad (ulvestad (1985)), who concluded that this galaxy may have undergone a large amount of fairly recent star formation. Hes et al. (hes (1995)) classified it as Narrow Line radio galaxy and suggest that a fraction of the FIR emission may be due to heating of the dust by the moderately young blue stars.
### 5.4 Average Detections
We have looked for “statistical” emission from both the representative CSS/GPS sample (i.e. for $`0.2z0.8`$) and the control sample. We have computed both the mean and the median flux density, although in the following we use the second one, which is considered to be more robust since it is less affected by extreme flux density values. In computing the mean flux density we excluded 3C459 for C100, since its large detected flux density would affect the results too much (the source was used, instead, in computing the median).
Values are given in Table 8. The uncertainty on the mean and median are derived from the flux density distribution itself. Note that, as the error distribution of the median is not Gaussian, we choose to give the 95% confidence uncertainty, equivalent to a 2 $`\sigma `$ limit for a Gaussian distribution. To make the comparison easier we then give 2$`\sigma `$ errors also on the mean.
At 60 and 90 $`\mu `$m there is clearly a 3 $`\sigma `$ detection for the CSS/GPS sample. In the comparison sample, where the statistics is poorer, as we do not yet have data for $`1/3`$ of the objects (see Table 2), both mean and median flux densities, compared with the errors, do not indicate a significant detection. They are, however, largely consistent with the CSS/GPS results. This is reinforced by a comparison done with Heckman et al. (heckman2 (1992); heckman3 (1994)). These authors give co–added flux densities for a sample of radio galaxies. In their “near sub–sample” ($`0.3<z<0.85`$, median $`z=0.48`$), composed of 5 CSS/GPS and 36 large size sources, and therefore similar to our comparison sample, the median flux densities (S$`{}_{60}{}^{}25\pm `$5 mJy, S$`{}_{100}{}^{}40\pm `$12) are consistent with ours, but have smaller errors, due to the much larger number of sources in their sample.
At 174$`\mu `$m and 200$`\mu `$m radio galaxies in the control sample seem stronger FIR emitters, although the difference is not significant at either wavelength. Averaging the two wavelengths the difference is around 2$`\sigma `$.
We have also applied a Kolmogoroff–Smirnov test, which shows that the hypothesis that the two galaxy samples are extracted from the same population is acceptable at a 99% confidence level at all wavelengths.
Considering also the results from Heckman et al. (heckman2 (1992); heckman3 (1994)), we estimate that any IR excess for CSS/GPS sources does not exceed (at a 95% confidence level) $``$ 25 mJy in the range 60–90 $`\mu `$m and $``$ 100 mJy in the range 170–200 $`\mu `$m.
Our first conclusion then is that we do not have any statistical evidence that the MFIR properties of the galaxies associated with CSS/GPS sources are much different from those of the galaxies associated with large size radio sources (see also Sect. 6.2) consistent with the results of Heckman et al. (heckman3 (1994)). The small and large sources we observed seem to have similar average luminosities at all four wavelengths, and hence a similar spectrum between 60 and 200 $`\mu `$m (observer frame). This will allow us to combine the two sets of data in the discussion (Sect. 6.2).
## 6 Discussion
### 6.1 FIR tail of synchrotron emission
We have collected all the flux densities available in the literature at radio wavelengths for every source in the sample. No evidence is seen for bright flat spectrum cores at the highest available frequencies (often up to 230 GHz, 1300 $`\mu `$m). The spectra can be well fitted either with a power law or with a model which includes spectral curvature due to synchrotron losses (see Murgia et al. murgia (1999), for details). For each source, using the fitted spectrum, we have extrapolated the flux densities at 60, 90, 174 and 200 $`\mu `$m. The extrapolated average synchrotron flux density is very small ($`<`$ 2 mJy at 200 $`\mu `$m and less at the other wavelengths) and does not contribute significantly to the FIR emission. Therefore we attribute the emission we find to genuine thermal radiation.
### 6.2 FIR luminosities of the combined CSS/GPS and comparison sample
As discussed in Sect. 5.4, our data do not show any statistically significant evidence for differences in the FIR luminosities of CSS/GPS and large size radio galaxies at 60 and 90 $`\mu `$m, the situation being somewhat more uncertain in the 174–200 $`\mu `$m interval.
Now we combine the two samples together into a single sample of powerful radio galaxies at intermediate redshift to be studied on its own, independent of the type of associated radio source. In Table 9, we report the mean and median FIR flux densities of the combined sample and the average FIR luminosities in the 60–90$`\mu `$m and 174–200 $`\mu `$m ranges.
The FIR luminosities from the median values of the whole sample are calculated as:
$$L_{\mathrm{FIR}}=4\pi D_\mathrm{L}^2\nu S_\nu $$
where $`S_\nu `$ is the median flux density at the frequency $`\nu `$ and $`D_\mathrm{L}`$ is the luminosity distance.
The average $`L_{\mathrm{FIR}}`$ are in the range $`(0.61.0)\times 10^{11}`$ $`L_{}`$, or (2–5)$`\times 10^{44}`$ erg s<sup>-1</sup>. The brightest objects, 1819+39 and 3C459 , which are several times more luminous than average (Table 10), have peculiarities, as commented in Sect. 5.3, which may well account for their larger FIR luminosities.
These luminosities are much higher than those found in nearby radio quiet elliptical galaxies, which are in the range of up to $`10^8L_{}`$ (Goudfrooij & de Jong goudfrooij2 (1995); Bregman et al. bregman (1998)).
We point out that the median FIR luminosity we find follows the extrapolation of the $`L_{\mathrm{FIR}}`$ vs $`L_{\mathrm{radio}}`$ plot for nearby radio galaxies presented in Knapp et al. (knapp (1990)) (see Fig. 1).
### 6.3 Temperature and Dust masses
On the hypothesis that the dust is transparent at these wavelengths, the MFIR radiation due to dust at a uniform temperature $`T`$ follows a modified Planck law given by
$$S(\nu _0)=\frac{\mu (\nu _e)B_{\mathrm{b}.\mathrm{b}.}(T,\nu _e)M_{\mathrm{dust}}(1+z)}{D_\mathrm{L}^2}$$
(3)
where $`\nu _0`$ and $`\nu _e`$ are the observed and emitted frequencies, $`\mu `$ is the absorption coefficient, $`B_{\mathrm{b}.\mathrm{b}.}`$ is the brightness of a black body at temperature $`T`$, $`M_{\mathrm{dust}}`$ is the dust mass, $`D_\mathrm{L}`$ is the luminosity distance and $`z`$ is the source redshift. The absorption coefficient is rather uncertain. In order to keep consistency with the values generally assumed in the literature (e.g.: Hildebrand hildebrand (1983); Knapp et al. knapp (1990)), we take $`\mu (\lambda )=10\times (250/\lambda (\mu m))`$ for $`\lambda 250\mu `$m. With respect to other assumed values, this tends to overestimate the dust mass (e.g. by a factor of two as compared to the $`\mu (\lambda )`$ assumed by Hughes et al., hughes (1997)).
The MFIR average spectrum for our sample is shown in Fig. 2. In order to describe it better, we used also the following additional upper limits or median flux densities: $`<`$ 5 mJy at 1.3 mm (230 GHz), as maximum flux density excess over the extrapolated synchrotron spectrum (Murgia et al. murgia (1999)); $`<`$ 9 mJy at 12$`\mu `$m and $`(12\pm 4)`$ mJy at 25 $`\mu `$m (Heckman et al. heckman2 (1992), near sub–sample).
The dependence on frequency of the FIR flux density in the observed wavelength range 12–200 $`\mu `$m ($`\nu _{\mathrm{rest}}\mathrm{36\; 000}2000`$ GHz), can be described by a power law:
$$S(\nu )\nu ^{1\pm 0.2}$$
At lower frequencies, however, the spectrum of the FIR component must turn over to meet the 230 GHz upper limit.
This spectrum is clearly not consistent with a single temperature. A multi–temperature fit (as in models by Sanders et al. sanders2 (1989)) would give a plausible range of temperatures between $`25`$ K and $``$ 120 K in the wavelength range we have considered here. However, as the data quality is not sufficient to allow a meaningful multi–temperature fit, we have taken a simple two temperature approximation, with $`T_1=80`$ K and $`T_2=25`$ K. With corresponding dust masses $`M_15\times 10^5M_{}`$ and $`M_22\times 10^8M_{}`$, we can reproduce the FIR flux densities in the range from 60 $`\mu `$m to 200 $`\mu `$m.
Any dust component colder than 25 K is poorly constrained by our data. For instance, up to $`10^9M_{}`$ would be consistent with the 174/200 $`\mu `$m median flux densities and the 1.3 mm upper limit if the temperature were $`20`$ K. In order to probe such low temperatures, measures at longer wavelengths would be required, as with SCUBA (Holland et al., holl (1999)). To our knowledge, such data do not exist for our sources.
A comparison of our average mass estimates with those for nearby radio quiet ellipticals (Bregman et al. bregman (1998)) and nearby radio galaxies (Knapp et al. knapp (1990)) is not straightforward, as these estimates depend on the wavelengths available and on the assumed temperatures. Bregman et al. (bregman (1998)) evaluate the temperatures from the 60 $`\mu `$m to 100$`\mu `$m flux density ratio ($`30`$ K) and obtain $`10^5M_{\mathrm{dust}}/M_{}10^7`$. Knapp et al. (knapp (1990)) assume $`T19`$ K and derive $`10^6M_{\mathrm{dust}}/M_{}10^8`$.
In our opinion, it is far from being excluded that the range of dust masses in these samples and in ours are similar, and the different estimates may be due to either the different assumed temperatures and/or to the different wavelength ranges explored.
Using the two temperature approximation, we have also estimated dust masses for the brighter sources detected, or possibly detected, at more than one wavelength.
The results of this calculations are presented in Table 10. Compared to the median values of the total sample, the sources 1345+145, 3C318, 1819+39 and 3C459 seem to have a higher mass of the warm dust component, while the cold dust masses seem to be similar. As the first three objects are a Seyfert type galaxy, a quasar and a Broad Line radio galaxy, we may expect that a favorable orientation of the objects with respect to the line of sight allows us to see the warmer dust associated with the accretion disk.
### 6.4 What heats the dust?
According to the unification models, the radio galaxies in our sample, which are very powerful in the radio band and have large luminosities in the Narrow Emission Lines ($`L_{\mathrm{OIII}}10^{43}`$erg s<sup>-1</sup>, Gelderman & Whittle, gelderman (1994)), are expected to harbor an obscured quasar (see, e.g., Falcke et al. falcke (1995), in the context of ”jet/disk–symbiosis”). A large fraction of the MFIR emission would then originate from the dust in the absorbing torus itself, heated by the UV/X radiation emitted at its center, with some contribution from the interstellar medium, still heated by the central AGN but now distributed over a scale of several kpc, and therefore with temperatures decreasing outwards. We expect this interstellar dust to emit its radiation at the longer wavelengths.
If an obscured AGN is present, adopting a heating model as in Sanders et al. (sanders2 (1989)), we expect dust temperatures in a range similar to that estimated in Sect. 6.3.
If, instead, no AGN is present and heating is due to the stellar radiation field only, the dust temperature could be $`<`$ 30 K, as in nearby radio quiet elliptical galaxies (Bregman et al. bregman (1998)). The emission would then peak at $`\lambda 200\mu `$m. An additional contribution to heating could arise from a burst of star formation, as may be the case in 3C459.
The interpretation based on heating by an obscured quasar is supported by the correlation between radio emission and MFIR emission (Knapp et al. knapp (1990); Heckman et al. heckman3 (1994)), with which our data are also consistent (Fig. 1). Also the strong luminosity in Narrow Lines, which is often interpreted as powered by photo–ionization, gives support to this interpretation. In addition we note that the MFIR average spectrum (for $`\lambda _{\mathrm{rest}}20\mu `$m) in nearby radio quiet ellipticals (Bregman et al. bregman (1998)) is definitely steeper than that of our radio galaxies and of those in Knapp et al. (knapp (1990)). The spectral index of the former is $`\alpha _{\mathrm{RQE}}2.3\pm 0.2`$, as compared to $`\alpha _{\mathrm{us}}1.0\pm 0.2`$ in our sample (Sect. 6.2) and $`\alpha _{\mathrm{Knapp}}1.2\pm 0.2`$ in Knapp et al. (knapp (1990)). This suggests that the heating processes are different, with the one in radio galaxies being more effective.
We conclude that our observations are consistent with the MFIR luminosity in these radio galaxies being powered primarily by the AGN, and assume that most of the radiation comes from the “obscuring torus”.
### 6.5 Implications for the circumnuclear torus
Our average FIR luminosities, consistent with the results of Heckman et al. (heckman2 (1992)), are about a factor 4 – 5 lower than the FIR luminosities of radio quasars of comparable radio luminosity (see Fig. 2).
This causes problems for the unification models (see Heckman et al. heckman2 (1992)), since, if the torus dust is transparent at FIR wavelengths, its emission should be orientation independent and the quasars should be as luminous in the FIR as the radio galaxies. On the other hand, if the disk is optically thick in some wavelength range, one should justify the constant flux density ratio between radio galaxies and quasars over a broad wavelength range (from 12 to 100 $`\mu `$m).
Hes et al. (hes (1995)) discuss the possibility that a beamed non–thermal component is present in quasars and contributes at the FIR wavelengths.
### 6.6 Back to the frustration scenario for CSS/GPS sources
The original aim of this work was to study the cold phase of the interstellar medium in radio galaxies, in order to see whether its density is high enough to support the “frustration scenario” for CSS/GPS sources. In this scenario, CSS/GPS and large size radio galaxies have similar lifetimes, so that the difference in size (more than a factor of 100, on average, in our two samples, Tables 1 and 2) requires an environment significantly denser for the former (De Young deyoung (1993); Fanti et al. fanti2 (1995)).
The hot component of the ISM is too tenuous and has insufficient pressure to confine the radio source (e.g. O’Dea et al. odea2 (1996); Readhead et al. readhead (1996)). Thus the radio source must be confined via interaction with relatively cold dense material along the path of propagation. However, as anticipated in the introduction, even if a lot of dense cold gas is present in the host galaxy, distributed in the disk/torus of the “unified models” (which is perpendicular to the radio axis) it will have no effect on the radio source propagation. “Frustration”, if present, must be due to the diffuse cold phase of the interstellar gas.
¿From ram pressure computations, the estimates of the density of the interstellar gas required to frustrate CSS radio galaxies imply masses $`2\times 10^{10}M_{}`$ within a volume of $`10`$ kpc in radius (Fanti et al. fanti2 (1995)), corresponding, for a gas to dust ratio as in our own Galaxy, to an excess dust mass $`2\times 10^8M_{}`$. This would produce an excess in the FIR emission of the CSS/GPS, as compared to the large size radio galaxies, the exact amount depending on the dust temperature. The upper limits to any IR excess derived in Sect. 5.4 convert into an excess dust mass of $`<\mathrm{0.6\; 10}^8M_{}`$ for $`T>30`$ K, (if heating is due to a powerful AGN), which is not enough for the frustration model. Even if we attribute the whole MFIR emission we detect from CSS/GPS to a frustrating medium only, and not to the dusty disk/torus, the implied mass (for $`T>`$ 30 K), would be still not enough, the upper limits to the differences in the MFIR emission being similar to the actual flux densities of CSS/GPS sources.
As a consequence, we may be confident that there is no evidence for a component of the interstellar medium, with temperatures $`>`$30 K, significantly denser or more massive in CSS/GPS galaxies to cause frustration.
For lower temperatures, our FIR observations are not sensitive enough to firmly constrain larger masses, since the emission would peak at $`\lambda >200\mu `$m, and the uncertainties we have on the flux density limits at long wavelengths are too large to provide safe constraints. For instance, if the temperature were $``$20 K, up to 5$`\times 10^8M_{}`$ of dust would be permitted. This should be probed by observations at longer IR wavelengths (e.g. with SCUBA). An additional strong constraint on such large dust masses is obtained from the color properties of the host galaxies. Dust masses as required by frustration, spread over $``$ 10 kpc, would produce absorption in the optical band. The implied hydrogen column density, $`N_\mathrm{H}10^{22}`$ g cm<sup>-2</sup>, corresponds to an optical depth in the visual $`\tau _\mathrm{v}N_\mathrm{H}/(\mathrm{2\; 10}^{21})`$. Applying a model of homogeneously mixed dust and stars (as in Goudfrooij & de Jong goudfrooij2 (1995)), the above dust mass implies $`A_\mathrm{v}2`$ mag., and color reddening E(B–V) $`0.7`$, which is inconsistent with the data of de Vries et al. (devries (1998)).
If “frustration” occurs only for GPS (sizes $`1`$ kpc), the implied masses are lower, but their temperature would be larger ($`50`$ K, for a model as in Sanders et al. sanders2 (1989)), due to the closer proximity to the hidden quasar, so that the expected IR luminosity would be even more discrepant with what we observe.
The present conclusion is therefore that the FIR radiation we detect with ISO does not imply an amount of gas (heated by the AGN) large enough to support the frustration scenario of CSS/GPS radio sources. This is consistent with the results of Owsianik et al. (owsianik (1998)) and Owsianik & Conway (owsianik2 (1998)), who found fast proper motions of the outer lobes in some small symmetric GPS, and with the radio spectra analysis of CSS/GPS by Murgia et al. (murgia (1999)).
We finally note that recent observations with SCUBA (Archibald et al. archi (2000)) of high redshift radio galaxies, both CSS and large size ones, do not show any significant difference between the two classes of sources.
The MFIR properties of CSS/GPS, however, point out a new problem for the alternative interpretation. An important ingredient of the “youth scenario” is that the radio luminosity is required to decrease with increasing source size, in order to explain the numerical proportion of CSS/GPS to large size sources. In the current models, this luminosity evolution is essentially due to expansion of the lobe volumes in an external medium of decreasing density and not to a decline of the power carried by the jet with time. CSS/GPS sources would evolve, with increasing size, toward lower luminosity larger size radio sources. This evolution in luminosity may also be described in terms of a higher efficiency for converting the beam power into radio band radiation (see also Gopal Krishna & Wiita gopal (1991)) in the early phases of the source evolution.
According to this idea, our two samples (CSS/GPS and comparison), which have very different sizes but similar radio luminosities, would have “radio engines” of power different by a factor 10 or so (CSS/GPS being less powerful). As in the past there have been strong suggestions that the jet power is proportional to the bolometric luminosity of the AGN (e.g. Baum & Heckman baum2 (1989); Rawlings & Saunders rawlings (1991); Falcke et al. falcke (1995)), one would expect the CSS/GPS to be definitely less powerful in the MFIR (and in the NL emission). This is not seen in our observations, at least at 60 and 90 $`\mu `$m.
Although we have no obvious solution to this problem, we speculate here about a couple of possibilities. First there may be an additional source of heating arising from conversion of a fraction of the jet power where the jet interacts with the interstellar medium. This may be more effective at distances close to the nucleus and less so further out as the source ages. The existence of emission line gas aligned with the radio source in low redshift CSS sources is consistent with strong interaction between GPS/CSS sources and the dense clouds in the ISM (de Vries et al. devries2 (1999)). Such an effect should roughly compensate for the lower radiation power from the central engine in CSS/GPS.
The second possibility is related to the scenario where the radio emission is triggered by a merger event. We speculate that in the early phase of the radio source life a large fraction of gas/dust is not fully settled in a disk/torus but is still distributed over a much larger solid angle, as seen from the central continuum source. In this scenario the fraction of intercepted and reprocessed UV radiation could be much higher in the younger/smaller sources, increasing their MFIR emission. We might also expect the additional obscuration in the smaller sources to result in lower emission line luminosities and/or redder colors. This is not generally seen in GPS and CSS sources, though there are a few cases which are consistent with this picture (O’Dea odea3 (1998); de Vries et al. devries (1998)). In addition, this hypothesis would require the gas/dust to settle into a disk on a timescale much less than the age of a radio source ($`10^{78}`$ yr) in order to remove the extra obscuration before the sources propagate to scales larger than tens of kpc.
At present, both of these explanations appear rather “ad hoc”. Thus, this issue remains a problem for the current evolution models.
## 7 Conclusions
a) We have presented ISOPHOT observations at $`\lambda =`$ 60, 90, 174 and 200 $`\mu `$m of CSS/GPS radio galaxies and of a matched comparison sample of extended radio galaxies. A minority of objects are detected individually at one or more wavelengths in both samples.
b) We have co–added the data for each sample to obtain mean and median flux densities. The extrapolated radio spectrum under–predicts the observed MFIR flux densities, arguing that the MFIR is due to dust rather than to synchrotron emission.
c) We find no significant differences in the MFIR flux densities of the two samples. Our results are then consistent with the CSS/GPS and the extended radio galaxies having similar MFIR luminosities. For dust temperatures $`>`$30 K, the deduced masses of the interstellar gas are lower than required by the frustration scenario. For lower temperatures larger masses would be allowed for by our data, but they would produce a large amount of obscuration and reddening in the optical, which is not seen in the existing data.
All this argues against the CSS/GPS sources being “frustrated” by a dense ambient medium.
d) Since no significant difference is seen between the two samples, we have combined them for further analysis. The average $`L_{\mathrm{FIR}}`$ is in the range $`(0.61.0)\times 10^{11}`$ L or $`(25)\times 10^{44}`$ erg s<sup>-1</sup>. Over the wavelength range of our observations, the spectrum can be fitted by a single power law with $`\alpha 1.0\pm 0.2`$.
e) We have fitted simple two temperature ($`T_1=80`$ K and $`T_2=25`$ K) models to the IR spectrum and have derived dust masses of $`M_15\times 10^5M_{}`$ and $`M_22\times 10^8M_{}`$. The mass of the cold dust appears higher than found in radio quiet elliptical galaxies, although this difference may be due to the differences in the assumed temperatures and the sampled rest-frame wavelength coverage.
f) Our observations are consistent with the MFIR luminosity in these powerful radio galaxies being mostly powered by an obscured AGN, although in some objects a contribution from star formation is possible.
###### Acknowledgements.
We are grateful to Dr. M. Haas of MPIA (Heidelberg) for his numerous suggestions during the data analysis stage and for providing us with specialized software to perform some non standard data reduction. CF and FP wish also to thank MPIA staff members for the help obtained during a short visit at the ISOPHOT group in Heidelberg. This work was partly supported by the Italian Ministry for University and Research (MURST) under grant cofin99-02-32
|
warning/0005/hep-th0005179.html
|
ar5iv
|
text
|
# Two-dimensional black holes as open strings: A new realization of the ADS/CFT correspondenceThis work is supported in part by funds provided by the U.S. Department of Energy (D.O.E.) under cooperative research agreement DE-FC02-94ER40818 and by a FCT grant, contract number BPD/20166/99.
## Abstract
We show that weak-coupled two-dimensional dilaton gravity on Anti-de Sitter space can be described by the dynamics of an open string. Neumann and Dirichlet boundary conditions for the string lead to two different realizations of the Anti-de Sitter/Conformal Field Theory correspondence. In particular, in the Dirichlet case the thermodynamical entropy of two-dimensional black holes can be exactly reproduced by counting the string states.
preprint: INFNCA-TH0009
The realization of the holographic principle in two spacetime dimensions is a subject that has recently attracted much attention in the literature, where it has been mainly investigated in the context of the Anti-de Sitter/Conformal Field Theory (AdS<sub>d</sub>/CFT<sub>d-1</sub>) correspondence . For $`d=2`$ it states that gravity on $`\mathrm{AdS}_2`$ is dual to a one-dimensional conformal field theory living on the boundary of $`\mathrm{AdS}_2`$ .
In spite of the efforts that have been devoted to clarify the AdS<sub>2</sub>/CFT<sub>1</sub> duality , the latter remains puzzling and mysterious. Since the conformal symmetry involved in the duality is infinite dimensional, the dynamics is expected to be highly constrained. However, the realization of the symmetry in terms of boundary states is far from trivial . This difficulty seems related to the topology of the boundary of $`\mathrm{AdS}_2`$ , which is one-dimensional and disconnected.
The lack in understanding of the AdS<sub>2</sub>/CFT<sub>1</sub> duality has prevented real progress in what is considered its main application: the study of two-dimensional (2D) gravity structures (e.g black holes) using conformal field theory techniques. This application is of fundamental relevance for black holes physics because it can be used to give statistical meaning to the entropy of both 2D black holes and higher dimensional black holes that reduce to 2D models upon compactification. Attempts to calculate the statistical entropy of 2D $`\mathrm{AdS}_2`$ black holes met only partial success . A mismatch of a factor $`\sqrt{2}`$ between the thermodynamical and statistical entropy was found.
In this letter we clarify the meaning of the AdS/CFT correspondence in two dimensions by showing that it can be realized in two different ways both steming from a more fundamental AdS<sub>2</sub>/CFT<sub>2</sub> correspondence. Using the nonlinear sigma model formulation of 2D dilaton gravity we show that weak-coupled dilaton gravity on $`\mathrm{AdS}_2`$ can be described by the dynamics of an open string. Using Neumann boundary conditions we retrieve the AdS<sub>2</sub>/CFT<sub>1</sub> correspondence that has been analyzed in Ref. . Dirichlet boundary conditions lead to a new realization of the AdS/CFT correspondence. In this case the properties of 2D black holes have a direct interpretation in terms of string dynamics. In particular, the entropy of the black hole can be exactly computed in terms of the degeneracy of the open string spectrum.
The simplest 2D gravity model admitting $`\mathrm{AdS}_2`$ as solution is the Jackiw-Teitelboim (JT) model
$$A=\frac{1}{2}\sqrt{g}d^2x\varphi \left(R+2\lambda ^2\right).$$
(1)
The classical solutions of the model,
$$ds^2=\left(\lambda ^2r^2\frac{2m}{\lambda \varphi _0}\right)dt^2+\left(\lambda ^2r^2\frac{2m}{\lambda \varphi _0}\right)^1dr^2,\varphi =\varphi _0\lambda r,m0,$$
(2)
can be interpreted as $`\mathrm{AdS}_2`$ -black holes . The black hole mass $`m`$ appearing in Eq. (2) is defined by the mass functional
$$M=\lambda ^2\varphi ^2_\rho \varphi ^\rho \varphi .$$
(3)
On-shell $`M`$ is constant and equal to $`2\varphi _0\lambda m`$.
The 2D gravity model (1) is pure gauge, i.e., it has no physical local degrees of freedom. Moreover, solutions (2) with different values of $`m`$ represent different, locally equivalent, parametrization of $`\mathrm{AdS}_2`$ . However, the presence of the scalar $`\varphi `$ makes them globally nonequivalent . Following the notation of Ref. we will denote with $`\mathrm{AdS}_2^+`$ and $`\mathrm{AdS}_2^0`$ the black hole solutions ($`m>0`$) and the ground state ($`m=0`$), respectively.
The link between 2D AdS-gravity and CFT can be established using the asymptotic symmetries of $`\mathrm{AdS}_2`$ . It has been shown in Refs. that the asymptotic symmetries of $`\mathrm{AdS}_2`$ are generated by a Virasoro algebra and that the deformations of the timelike boundary of $`\mathrm{AdS}_2`$ give a realization of the conformal symmetry. In Refs. the $`(r,t)`$ coordinates of Eq. (2) have been used to discuss the asymptotic symmetries of $`\mathrm{AdS}_2`$ , yet for our purposes it is convenient to use light-cone coordinates $`(u,v)`$. In the $`(u,v)`$-frame the $`\mathrm{AdS}_2^0`$ solution is $`g_{uv}=2/\lambda ^2(u+v)^2`$, $`g_{uu}=g_{vv}=0`$, $`\varphi =\varphi _0/\lambda (u+v)`$, so the boundary conditions to be imposed on the metric and on the dilaton are $`g_{uv}=2/\lambda ^2(u+v)^2+O\left(1\right)`$, $`g_{uu}=O\left(1\right)`$, $`g_{vv}=O\left(1\right)`$, and $`\varphi =O\left((u+v)^1\right)`$, respectively. The metric and the dilaton have the asymptotic, $`uv`$, form
$`g_{uu}`$ $`=`$ $`U_0+\mathrm{}+U_n(u+v)^n+\mathrm{},`$ (4)
$`g_{uv}`$ $`=`$ $`{\displaystyle \frac{2}{\lambda ^2(u+v)^2}}+Y_0+\mathrm{}+Y_n(u+v)^n+\mathrm{},`$ (5)
$`g_{vv}`$ $`=`$ $`V_0+\mathrm{}+V_n(u+v)^n+\mathrm{},`$ (6)
$`\varphi `$ $`=`$ $`\varphi _0\left[{\displaystyle \frac{\omega _1}{\lambda (u+v)}}+\omega _1\lambda (u+v)+\mathrm{}+\omega _n\lambda ^n(u+v)^n+\mathrm{}\right],`$ (7)
where the coefficients $`U_k,U_k,\omega _k`$ are functions of $`uv`$ only. The transformations generated by the asymptotic symmetry group leave unchanged the leading terms in Eq. (4) and act on the remaining functions $`U_k,V_k,Y_k`$ and $`\omega _k`$. These can be thought as characterizing the deformations of the $`u=v`$ boundary of $`\mathrm{AdS}_2`$ .
The asymptotic symmetry group is generated by the Killing vectors
$`\chi ^u`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left[ϵ+ϵ^{}(u+v)+{\displaystyle \frac{1}{2}}ϵ^{\prime \prime }(u+v)^2\right]+\alpha ^u,`$ (8)
$`\chi ^v`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left[ϵ+ϵ^{}(u+v){\displaystyle \frac{1}{2}}ϵ^{\prime \prime }(u+v)^2\right]+\alpha ^v,`$ (9)
where $`ϵϵ(uv)`$ is an arbitrary function, $`\alpha ^{u,v}=_{k=3}^{\mathrm{}}\alpha _k^{u,v}(uv)(u+v)^k`$, and $`{}_{}{}^{}=d/d(uv)`$. $`\alpha ^{u,v}`$ represent “pure gauge” diffeomorphisms of the 2D gravity theory that fall off rapidly as $`uv`$. The Killing vectors (8) define a conformal group generated by a Virasoro algebra and the boundary fields $`\mathrm{\Theta }_k=(U_k,V_k,Y_k,\omega _k)`$ span a representation of this symmetry. Their transformation law has the form
$$\delta _ϵ\mathrm{\Theta }_k=ϵ\mathrm{\Theta }_k^{}+(h+k)ϵ^{}\mathrm{\Theta }_k+\mathrm{},$$
(10)
where $`h=2`$ for the fields $`U,V,Y`$ and $`h=0`$ for the fields $`\omega `$, and dots denote terms that depend on higher derivatives of $`ϵ`$ and on the “pure gauge” diffeomorphisms. It is important to notice that although “pure gauge” transformations affect the boundary fields, the charges that are associated with the asymptotic symmetry are invariant under pure gauge transformations. The mass functional $`M`$ can be likewise expanded near the $`u=v`$ boundary
$$M=\underset{k=0}{\overset{\mathrm{}}{}}M_k(uv)(u+v)^k.$$
(11)
Using Eqs. (3) and (4) the $`M_k`$ can be expressed in terms of the boundary fields. They follow the transformation law (10) with $`h=0`$.
The action (1) can be cast in the form of a 2D conformal nonlinear sigma model with Lagrangian
$$=\sqrt{g}_\mu M^\mu \psi \frac{1}{14\lambda ^2\psi ^2M},$$
(12)
where $`\varphi =(2\lambda ^2\psi )^1`$ and we have neglected irrelevant surface terms. The Lagrangian (12) can be expanded around $`\psi =0`$
$$=\sqrt{g}\underset{k=0}{\overset{\mathrm{}}{}}_\mu M_k^\mu \psi _k,$$
(13)
where $`M_k=M^{k+1}/(k+1)`$, $`\psi _k=(2\lambda \psi )^{2k+1}/2\lambda (2k+1)`$. Equation (13) is both a perturbative expansion in terms of the (coordinate-dependent) gravitational coupling of the model (1) ($`\varphi ^1`$) and an expansion near the boundary of $`\mathrm{AdS}_2`$ . Each term in Eq. (13) has the form of a free-field conformal theory and transforms according to Eq. (10) with $`h=2`$, although the sum (full theory) does not. In the weak-coupling regime, $`\psi <<1`$, the theory can be treated perturbatively. In particular, if we restrict ourselves to the first order in the perturbative expansion the theory reduces to a free CFT and the usual machinery of CFTs can be applied. In this sense we speak of “duality” between (weak-coupled) $`\mathrm{AdS}_2`$ gravity and CFT. Let us stress that the identification of a weak-coupling regime, where a consistent perturbative expansion can be constructed, is a fundamental feature of the sigma model representation that does not have counterpart in the original formulation based on Eq. (1) and is essential for the identification of the fundamental microscopic degrees of freedom of the theory. In this paper we shall focus attention on the first term in the perturbative expansion (13) leaving the discussion of higher terms to further investigations. This amounts to neglecting, in first approximation, higher order perturbative corrections generated by the running gravitational coupling. Note that the latter becomes strong near $`r=0`$, i.e., precisely in the “opposite” region of $`AdS_2`$ around which we are expanding. In this approach higher order corrections are described by interacting terms for the free CFT.
It is convenient to define the new fields
$$\begin{array}{ccccc}\sqrt{\pi \alpha ^{}}M_0\hfill & =\hfill & \frac{1}{2}(X^1+iX^2)\hfill & =\hfill & \frac{1}{2}(X^0+X^1),\hfill \\ \sqrt{\pi \alpha ^{}}\psi _0\hfill & =\hfill & \frac{1}{2}(X^1iX^2)\hfill & =\hfill & \frac{1}{2}(X^1X^0),\hfill \end{array}$$
(14)
where $`\alpha ^{}`$ is a constant with dimension of $`lenght^2`$. Using complex coordinates $`zu=(t+x)/2=(\sigma ^1+i\sigma ^2)/2`$ and $`\overline{z}v=(xt)/2=(\sigma ^1i\sigma ^2)/2`$, the leading term in the expansion (13) can be cast in the usual bosonic string form. \[See Ref. for notations\]. Since $`\mathrm{AdS}_2`$ has a timelike boundary at $`x=0`$, we are dealing with open strings and the expansion (13) defines a AdS<sub>2</sub>/CFT<sub>2</sub> correspondence between open string theory and dilaton gravity on AdS<sub>2</sub>. Boundary conditions are restricted to Dirichlet ($`X^\mu (x=0)=const`$) or Neumann ($`n^a_aX^\mu (x=0)=0`$) type, respectively. Mixed boundary conditions are not allowed because from the boundary expansion (4) for the field $`\varphi `$ it follows $`_tX^0(x=0)=_tX^1(x=0)`$. The choice of boundary conditions determines the realization of the AdS/CFT correspondence. The AdS<sub>2</sub>/CFT<sub>1</sub> correspondence that has been proposed in Ref. is obtained by imposing Neumann boundary conditions, which allow for excitations on the boundary. In this case we have $`X^\mu (x=0)=F(t)=M_0`$ and the conformal symmetry can be realized on the boundary by the charges that are associated with the asymptotic symmetries of $`\mathrm{AdS}_2`$ . Dirichlet boundary conditions break translation invariance in the $`x`$ direction and no dynamical degrees of freedom are allowed on the boundary, the string endpoint being fixed. In this case we are naturally lead to a new realization of the AdS/CFT correspondence. It is shown below that the correspondence is realized in terms of pure deformations of the boundary of AdS<sub>2</sub>.
In addition to the timelike boundary at $`x=0`$, AdS$`{}_{}{}^{0}{}_{2}{}^{}`$ has an inner null boundary. However, the presence of the latter does not influence the dynamics of the open string. Writing the solution, Eq. (2), in the conformal coordinate frame $`(t,x)`$, one finds that AdS$`{}_{}{}^{0}{}_{2}{}^{}`$ is conformal to Minkowski space and that the presence of the dilaton requires
$$\mathrm{}<t<\mathrm{},0x<\mathrm{}.$$
(15)
In this coordinate frame the inner null boundary is located at $`x=\mathrm{}`$. Hence, because of conformal invariance, open strings on $`\mathrm{AdS}_2^0`$ are equivalent to open strings on the region of the Minkowski spacetime defined by Eq. (15).
The AdS<sub>2</sub>/CFT<sub>2</sub> correspondence is expressed in a exact form by putting in a one-to-one correspondence the symmetries and local degrees of freedom of the open string and the asymptotic symmetries and excitations of $`\mathrm{AdS}_2`$ . The conformal symmetry in two spacetime dimensions is generated by the Killing vectors, $`\chi =\chi (z)+\overline{\chi }(\overline{z})\overline{}`$. A generic CFT<sub>2</sub> field $`\psi (z,\overline{z})`$ of weights $`(h,\overline{h})`$ transforms as
$$\delta _{\chi ,\overline{\chi }}\psi =(\chi +h\chi )\psi +(\overline{\chi }\overline{}+\overline{h}\overline{}\overline{\chi })\psi .$$
(16)
Dirichlet boundary conditions require that $`\chi `$ and $`\overline{\chi }`$ are related by the condition
$$\chi (z)+\overline{\chi }(\overline{z})=0.$$
(17)
This equation implies that the conformal symmetry is generated by a single copy of the Virasoro algebra. By expanding the Killing vectors on the boundary we obtain Eq. (8) with $`\chi ((z\overline{z})/2)=\overline{\chi }((z\overline{z})/2)=ϵ(uv)/2`$, where the pure gauge diffeomorphisms have been fixed as $`\alpha _k^u=(1)^{k+1}\alpha _k^v=(1/2k!)d^kϵ/d(uv)^k`$. Thus, by fixing the pure gauge diffeomorphisms appropriately, the symmetry group of the Dirichlet open string and the asymptotic symmetry group of AdS<sub>2</sub> coincide. Each $`\mathrm{AdS}_2`$ field living near $`x=0`$ can be interpreted as the coefficient of the expansion of the CFT<sub>2</sub> field around the boundary with given weight $`h+\overline{h}`$ and pole of order $`p`$. Moreover, the above correspondence allows to determine from CFT<sub>2</sub> the Virasoro generators of the asymptotic symmetry group of AdS<sub>2</sub>. Using Eq. (17) and expressing the CFT<sub>2</sub> Virasoro generators $`L_m^{CFT}=z^{m+1}`$ and $`\stackrel{~}{L}_m^{CFT}=\overline{z}^{m+1}\overline{}`$ as functions of the $`x,t`$ coordinates we find
$$L_m^{AdS}=2^{m1}\left\{\left[(t+x)^{m+1}+(tx)^{m+1}\right]_t+\left[(t+x)^{m+1}(tx)^{m+1}\right]_x\right\}.$$
(18)
The AdS Virasoro generators (18) are valid both on the boundary and outside the boundary, where they generate the full symmetry group of the open string with Dirichlet boundary conditions. Equation (18) leads to the asymptotic AdS Killing vectors (8) with fixed gauge diffeomorphisms. By fixing the pure gauge diffeomorphisms of the AdS asymptotic symmetries we can reconstruct the full symmetry group of the Dirichlet open string. According to this picture the Virasoro generators $`L_m^{AdS}`$ cannot be interpreted as generating the symmetries of a 1D conformal field theory living on the boundary of $`\mathrm{AdS}_2`$ , the latter being frozen by the Dirichlet boundary conditions.
The AdS<sub>2</sub>/CFT<sub>2</sub> correspondence can also be realized using local oscillator degrees of freedom. Let us expand the string field in normal modes
$$X^\mu =x^\mu ip^\mu \mathrm{log}|z|^2+i\left(\frac{\alpha ^{}}{2}\right)^{1/2}\underset{m=\mathrm{}}{\overset{\mathrm{}}{}}\frac{1}{m}\left(\alpha _m^\mu z^m+\stackrel{~}{\alpha }_m^\mu \overline{z}^m\right).$$
(19)
Comparing Eq. (19) to the asymptotic expansions of $`M_0`$ and $`\psi _0`$
$$M_0=\underset{k=0}{\overset{\mathrm{}}{}}\underset{m=\mathrm{}}{\overset{\mathrm{}}{}}M_{km}x^kt^m,\psi _0=\underset{k=1}{\overset{\mathrm{}}{}}\underset{m=\mathrm{}}{\overset{\mathrm{}}{}}\mathrm{\Psi }_{km}x^kt^m,$$
(20)
we find \[we assume $`t>0`$ for simplicity\]
$$\alpha _m^\mu =(1)^{m+1}\stackrel{~}{\alpha }_m^\mu =i\sqrt{\pi }2^{1/2m}\left[M_{1,1m}\mathrm{\Psi }_{1,1m}\right],$$
(21)
where we have imposed the Dirichlet boundary conditions that imply $`p^\mu =0`$, $`M_{00}=const`$, and $`M_{0m}=0`$ for $`m0`$. The asymptotic excitations of the gravity theory are in a one-to-one correspondence with the open string modes. Moreover, the lower terms in the asymptotic expansion are sufficient to determine the whole CFT<sub>2</sub> theory. The fields $`M_1`$, $`\mathrm{\Psi }_1`$ and $`M_0`$ are invariant under pure gauge bulk transformations and transform conformally with weight $`h=1`$ ($`M_1`$ and $`\mathrm{\Psi }_1`$) and $`h=0`$ ($`M_0`$). Therefore, Eq. (21) provides a realization of the AdS<sub>2</sub>/CFT<sub>2</sub> correspondence: Asymptotic 2D gravity modes around the boundary that describe boundary deformations determine completely CFT<sub>2</sub> (the open string theory) and vice versa. Finally, using Eq. (21) the CFT<sub>2</sub> Virasoro generators can be expressed in terms of the asymptotic modes
$$L_m^{CFT}=\frac{1}{2}\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\alpha _{mn}^\mu \alpha _{\mu n}=\pi 2^m\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}M_{1,1n}\mathrm{\Psi }_{1,1m+n}.$$
(22)
By imposing Neumann boundary conditions on the open string, the string modes are determined by the gravitational modes $`M_{0,m}`$. In this case the Virasoro generators (22) and the CFT<sub>2</sub> action are zero at any order in the expansion and the conformal symmetry cannot be realized in terms of local string oscillators. Rather, we are dealing with a topological theory which has no physical local degrees of freedom and the conformal symmetry is realized by the charges associated with the asymptotic symmetries .
The AdS<sub>2</sub>/CFT<sub>2</sub> correspondence leads to a natural interpretation of the Hawking evaporation of the $`\mathrm{AdS}_2`$ black hole . From Eq. (18) it follows that the invariant $`SL(2,R)`$ algebra is generated by
$$L_0^{AdS}=t_t+x_x,L_1^{AdS}=2_t,L_1^{AdS}=\frac{1}{2}(t^2+x^2)_t+xt_x.$$
(23)
In the representation above $`L_0^{AdS}`$ does not generate translations in $`t`$ but dilatations. $`L_0^{AdS}`$ generates time translations in the $`T,X`$ coordinates
$$\lambda t=e^{\lambda T}\mathrm{cosh}(\lambda X),\lambda x=e^{\lambda T}\mathrm{sinh}(\lambda X).$$
(24)
Using Eq. (24) the metric of the $`\mathrm{AdS}_2^0`$ ground state is
$$ds^2=\frac{1}{\mathrm{sinh}^2(\lambda X)}(dT^2+dX^2).$$
(25)
Equation (25) describes a 2D black hole . Hence, the Hawking evaporation process can be explained in the CFT<sub>2</sub> context using the same arguments of Ref. . Positive frequency modes of a quantum field with respect to Killing vector $`_t`$ are not positive frequency modes with respect to Killing vector $`_T`$, i.e., the vacuum state for an observer in the $`(X,T)`$ reference frame appears filled with thermal radiation to an observer in the $`(x,t)`$ frame. The value of the Hawking flux has been calculated in Ref. .
The correspondence between the open string with Dirichlet boundary conditions and the 2D $`\mathrm{AdS}_2`$ black hole can be used to calculate the statistical entropy of the latter. Since local oscillators of the Dirichlet string are in one-to-one correspondence with excitations of $`\mathrm{AdS}_2^0`$ we can count black hole states by counting states of CFT<sub>2</sub>. To this purpose, we calculate the central charge $`c`$ associated with the central extension of the Virasoro algebra generated by $`L_m^{CFT}`$. Keeping in mind the interpretation of $`c`$ as a Casimir energy (see for instance Ref. ), the transformation law of the stress-energy tensor under the change of coordinates (24) ($`w=T+X`$) is
$$T_{ww}^{(2)}=(_wz)^2T_{zz}^{(2)}\frac{c}{12}\{w,z\}(_wz)^2.$$
(26)
The vacuum energy is shifted by $`l_0^{(2)}l_0^{(2)}\frac{c}{24}`$, where $`l_0^{(2)}`$ is the eigenvalue of $`L_0^{CFT}`$ which is associated to the vacuum. This shift corresponds to a Casimir energy $`E=\frac{c}{24}\lambda `$.
The coordinate transformation (24) maps the $`\mathrm{AdS}_2^0`$ ground state solution of the 2D dilaton gravity theory into the $`\mathrm{AdS}_2^+`$ black hole solution (25) with mass $`m=\frac{\varphi _0}{2}\lambda `$ (see Ref. ). Because of the correspondence between the gravitational theory and the Dirichlet string we can interpret the previous map as the gravity theory counterpart of the shift of $`L_0^{CFT}`$ in CFT<sub>2</sub> and equate the Casimir energy $`E`$ with $`m`$. There is a subtlety concerning the sign to be used in the equation. The coordinate transformation (24) is analogous to the coordinate transformation that maps the Rindler spacetime into the Minkowski spacetime, i.e., it maps observers. So an observer in the $`\mathrm{AdS}_2^+`$ vacuum sees the $`\mathrm{AdS}_2^0`$ vacuum as filled with thermal radiation with negative flux . Since the Casimir energy $`E`$ is the energy of the $`z`$-vacuum as seen in the $`w`$-frame, we must use the equation $`E=m`$ which leads to $`c=12\varphi _0`$. Finally, the eigenvalue of $`L_0^{CFT}`$ can be expressed in terms of the black hole mass. Using the Cardy formula the statistical black hole entropy is
$$S=2\pi \sqrt{\frac{cL_0^{CFT}}{6}}=4\pi \sqrt{\frac{\varphi _0m}{2\lambda }},$$
(27)
in complete agreement with the thermodynamical result.
In this letter we have proved that the correspondence between 2D gravity and open strings allows for two distinct realization of the AdS/CFT correspondence. The first realization, which is obtained by imposing Neumann boundary conditions to the open string, implies the existence of a genuine one-dimensional CFT living on the boundary of $`\mathrm{AdS}_2`$ . This realization is, however, problematic from different points of view . The realization which is obtained by imposing Dirichlet boundary conditions supports the viewpoint of Ref. , where, by quite a different argument, the authors conclude that the correspondence should be realized as $`\mathrm{AdS}_2`$ /CFT<sub>2</sub>, rather than $`\mathrm{AdS}_2`$ /CFT<sub>1</sub>.
|
warning/0005/astro-ph0005338.html
|
ar5iv
|
text
|
# The likely detection of pulsed high-energy 𝛾-ray emission from millisecond pulsar PSR J0218+4232
## 1 Introduction
Pulsed high-energy emission from millisecond (ms) pulsars has so far been detected at X-ray energies below $``$ 10 keV for only five pulsars: PSR J0437-4715 (Becker & Trümper (1993)), PSR J2124-3358 (Becker & Trümper (1999)), PSR B1821-24 (Saito et al. (1997)), PSR J0218+4232 (Kuiper et al. (1998)) and PSR B1937+21 (Takahashi et al. (1999)). The first two exhibit broad X-ray pulses, have soft spectra and relatively low luminosities in the X-ray window, about 3 orders of magnitude lower than derived for the latter three ($`L_\text{X}^{110\text{keV}}10^{32}\text{erg\hspace{0.17em} s}^1`$ assuming emission in a 1 sr beam). In addition to the higher luminosity, these have very narrow X-ray pulses (intrinsically $`100\mu s`$ or narrower) and hard power-law shape spectra measured up to $``$ 10 keV (Saito et al. (1997), Mineo et al. (2000), Takahashi et al. (1999), respectively), the two hardest spectra having indices as hard as $``$ -0.65. This short observational summary suggests that this small sample can de devided in two distinct classes of ms pulsars: Class I, ms pulsars with soft, low-luminosity X-ray emission in broad pulses; Class II, with highly non-thermal, high-luminosity X-ray emission in narrow pulses.
Millisecond pulsars not only differ from normal radio pulsars in that their spin periods are 1 to 2 orders of magnitude shorter, reducing correspondingly their light cylinder radii, but particularly their surface magnetic field strengths are 3 to 4 orders of magnitude weaker. Nevertheless, Bhattacharya & Srinivasan (1991) and Sturner & Dermer (1994) showed that both of the competing classes of models for the production of $`\gamma `$-rays (polar cap and outer gap models) predict the production of detectable non-thermal emission up to the high-energy $`\gamma `$-rays for a sizable number of ms pulsars. An early systematic search for pulsed high-energy $`\gamma `$-ray emission from ms pulsars rendered, however, only upper limits (Fierro (1995)). In this paper we will present circumstantial evidence for the first detection of pulsed high-energy gamma-ray emission from a Class II ms pulsar: PSR J0218+4232.
PSR J0218$`+`$4232 is a 2.3 ms pulsar in a two day orbit around a low mass ($``$ 0.2 M) white dwarf companion (Navarro et al. (1995); van Kerkwijk (1997)). The dipolar perpendicular magnetic field strength ($`B_{}`$) at the surface of the neutron star is estimated to be $`4.3\times 10^8`$ G and the spin-down age is $`4.6\times 10^8`$ years. The spin-down energy loss $`L_{\text{sd}}`$ of the pulsar amounts $`2.5\times 10^{35}`$ erg s<sup>-1</sup>. The pulsar distance inferred from its dispersion measure and from the electron density model of Taylor & Cordes (1993) is $`5.7`$ kpc.
Soft X-ray emission from the pulsar was first detected by Verbunt et al. (1996) in a 20 ks ROSAT HRI observation. In a 100 ks follow-up observation X-ray pulsations were discovered at a significance of about 5 $`\sigma `$ (Kuiper et al. (1998)). The X-ray pulse profile is characterized by a sharp main pulse with an indication for a second peak at a phase separation of $`\mathrm{\Delta }\varphi 0.47`$. The pulsed fraction inferred from the ROSAT HRI (0.1-2.4 keV) data is $`37\pm 13`$ %. It is interesting to note that also in the radio domain the source exhibits an unusually high unpulsed component of $``$ 50 % (Navarro et al. (1995)).
The ROSAT HRI provides no spectral information and the number of counts recorded in a far off–axis PSPC observation does not allow spectral modeling in the soft X-ray regime (0.1-2.4 keV). Also ASCA detected this source, however, the observation was too short: no pulsation could be detected, and a spectral fit to the weak total excess resulted in a power–law photon index of $`1.6\pm 0.6`$ (Kawai & Saito 1999).
The spectral information for PSR J0218$`+`$4232 improved enormously analyzing the data from a 83 ks BeppoSAX MECS (1.6-10 keV) observation performed early 1999 (Mineo et al. (2000)). Pulsed emission was detected up to 10 keV, the pulse profile clearly showing two peaks at the same phase separation of 0.47 which we reported earlier combining ROSAT HRI and PSPC observations (Kuiper et al. (1998)). The BeppoSAX MECS observation reveals that PSR J0218$`+`$4232 exhibits the hardest pulsar X-ray spectra measured so far: Between 1.6 and 10 keV one peak has a spectrum consistent with a power-law photon index of $`0.84`$ and the other with an index of $`0.42`$. The total pulsed spectrum can be described with an index $`0.61`$ (Mineo et al. (2000)).
At high-energy $`\gamma `$-rays, Verbunt et al. (1996) noticed the positional coincidence of PSR J0218$`+`$4232 with the second EGRET catalog source 2EG J0220$`+`$4228 (Thompson et al. (1995)), which was identified in the catalog and other publications with the BL Lac 3C 66A (Dingus et al. (1996); Mukherjee et al. (1997); Lamb & Macomb (1997)). Using some additional EGRET observations, and applying a combination of spatial and timing analyses, Kuiper et al. (1999a) conclude that 2EG J0220$`+`$4228 is probably multiple: between 100 and 1000 MeV PSR J0218$`+`$4232 is the most likely counterpart, and above 1000 MeV 3C 66A is the best candidate counterpart. The third EGRET catalog (Hartman et al. (1999)), which is based on more viewing periods than the 2EG catalog, also identifies 3EG J0222$`+`$4253 (2EG J0220$`+`$4228) with 3C 66A, rather than with the ms-pulsar. However, in a note on this source, they indicate that the identification with 3C 66A stems from the catalog position based on the $`>`$ 1 GeV map. Furthermore, they confirm that for lower energies (100-300 MeV) the EGRET map is consistent with all the source flux coming from the pulsar, 3C 66A being statistically excluded.
In this paper we present the results of spatial, timing and pulse-phase resolved spatial analyses using all available EGRET (30 MeV - 30 GeV) data collected between November 1991 and November 1998 in 5 observations with PSR J0218$`+`$4232 within 25$`\mathrm{°}`$ of the pointing axis. Analysis of radio monitoring data of this pulsar provided us with an ephemeris valid over the total period of 7 years covering the EGRET observations, allowing phase folding of all selected EGRET events in a single trial. The resulting high-energy $`\gamma `$-ray pulse profile is compared with pulse profiles detected at X-ray energies up to 10 keV, and in absolute phase with the radio profile at 610 MHz. The results are finaly discussed in relation to the Class II ms pulsars and the Crab, as well as with recent theoretical predictions for the production of X-ray and $`\gamma `$-ray emission in the magnetospheres of ms pulsars.
## 2 Instrument description and observations
EGRET (the Energetic Gamma Ray Experiment Telescope) aboard the Compton Gamma Ray Observatory (CGRO) has a (gas-filled) sparkchamber and is sensitive to gamma-rays with energies in the range 30 MeV to 30 GeV. In the mode used for most of the observations the field of view is approximately $`80\mathrm{°}`$ in diameter, although the instrument point-spread function (PSF) and the effective area degrade considerably beyond $`30\mathrm{°}`$ off-axis. Its effective area is approximately $`1500cm^2`$ between 200 and 1000 MeV, falling off at lower and higher energies. The angular resolution is strongly energy dependent: the $`67\%`$ confinement angle at 35 MeV, 500 MeV and 3 GeV are $`10\stackrel{}{.}9`$, $`1\stackrel{}{.}9`$ and $`0\stackrel{}{.}5`$ respectively. The energy resolution $`\mathrm{\Delta }E/E`$ is $`20\%`$ (FWHM) over the central part of the energy range. Each registered event is time tagged by the on-board clock, serving also the other 3 CGRO instruments BATSE, OSSE and COMPTEL. The on-board time is converted to Coordinated Universal Time (UTC) with an absolute accuracy better than $`100\mu s`$, and a relative accuracy of $`8\mu s`$. For a continued proper sparkchamber performance regular gas replenishments of the sparkchamber are required in order to restore the efficiency after the gas has aged. The sparkchamber efficiency is therefore a function of time and energy. For a detailed overview of the EGRET detection principle and instrument characteristics, see Thompson et al. (1993). The inflight calibration and performance are presented in detail by Esposito et al. (1999). In this work we selected those CGRO Cycle I-VII Viewing Periods (VP) in which PSR J0218$`+`$4232 , located at (l,b) = (139.508,$``$17.527), was less than $`25\mathrm{°}`$ off-axis. In Table 1 the details of each selected VP are given.
COMPTEL, the imaging Compton Telescope aboard CGRO, is co-aligned with EGRET and had PSR J0218$`+`$4232 in its field of view during the same VP’s as EGRET. COMPTEL operates in the 0.75-30 MeV energy window and has an energy resolution of about 5-10 % FWHM, a large field of view ($`1`$ steradian) and a location accuracy of $`1\mathrm{°}`$ (see Schönfelder et al. (1993)). Events are time-tagged with a $`0.125`$ ms resolution. A timing analysis of PSR J0218+4232 in the COMPTEL energy window did not yield a significant timing signal and subsequent imaging studies of the sky region containing PSR J0218$`+`$4232 did not show a source detection at the pulsar position. Therefore, only the flux upper limits are presented in this paper (see Sect. 8).
OSSE, the Oriented Scintillation Spectrometer Experiment aboard CGRO, is a non-imaging detector system consisting of 4 independent actively shielded NaI(Tl)-Cs(Na) phoswich detectors operating in the 0.05 to 10 MeV energy range, each having a $`3\stackrel{}{.}8\times 11\stackrel{}{.}4`$ (FWHM) field of view (see Johnson et al. (1993)). PSR J0218+4232 was the primary target of OSSE during VP 728.7 and VP 728.9 and was observed in event-by-event mode in the 50-150 keV energy band with a timing accuracy of 0.125 ms. Like in the case for COMPTEL, also OSSE did not detect a timing signature. Flux upper limits are given in Sect. 8.
## 3 Spatial analysis
Events arriving from within $`30\mathrm{°}`$ off axis when EGRET is in full FoV mode and $`19\mathrm{°}`$ in narrow field mode, are sorted in a 3 dimensional data cube with galactic longitude, latitude and energy as axes. The longitude and latitude bin widths are $`0\stackrel{}{.}5`$, and 10 narrow “standard” energy ranges are selected: 30-50, 50-70, 70-100, 100-150, 150-300, 300-500, 500-1000 MeV, 1-2, 2-4 and 4-10 GeV.
Because the Earth atmosphere is the largest source of non-celestial $`\gamma `$-rays the events are subjected to an energy dependent zenith angle cut. We used the “standard” values for the 10 selected energy windows. The corresponding energy dependent exposure maps are calculated using the “exposure history” files taking into account the instrument calibration characteristics, the instantaneous timeline, the operation mode of the instrument and the time dependent spark chamber sensitivity factors (see Esposito et al. (1999) and Table 1).
To be consistent with the selection criteria used in the generation of the exposure matrices we demand that the energy deposit in the TASC (Total Absorption Shower Counter) measured by at least one of its PHA’s is above a threshold of $`6.5`$ MeV.
The imaging method employed here is based on our Maximum Likelihood Ratio (MLR) program, part of the COMPTEL analysis software package COMPASS (de Vries (1994)). In this program point sources are searched for on top of a diffuse background model which describes the galactic and extra-galactic $`\gamma `$-ray emission separately. The galactic component originating in cosmic-ray interactions with the protons of the atomic and molecular Hydrogen gas, as well as inverse Compton interactions of cosmic-ray electrons with the ambient photon field, is described by a combination of 2 different models: one results from the convolution of the EGRET PSF with the spatial distribution of the atomic Hydrogen column density and the second from the convolution with the spatial distribution of CO used as tracer for the molecular Hydrogen gas in the galaxy. The extra-galactic component is assumed to be isotropic.
The image resulting from the Maximum Likelihood Ratio program is based on likelihood ratio tests at user defined grid points in a skyfield containing the object of interest. At each grid point ($`x_{sky},y_{sky}`$) we determine the Maximum Likelihood under two hypotheses: 1) a description of the data in terms of the diffuse background models only ($`_0`$) and 2) a description in terms of the diffuse background models and a point source at the ($`x_{sky},y_{sky}`$) position ($`_1`$). Under the $`_1`$ hypothesis the number of counts ($`\mu `$) expected in a measured sky pixel $`(i,j)`$ is given by :
$$\mu _{ij}=\sigma PSF_{ij}+\alpha ^{HI}M_{ij}^{HI}+\alpha ^{CO}M_{ij}^{CO}+\alpha ^{Iso}M_{ij}^{Iso}$$
(1)
where $`M^{HI},M^{CO}`$ and $`M^{Iso}`$ represent the convolved diffuse galactic and extra- galactic models.
Because our mosaic of observations is composed of viewing periods with pointing directions concentrated in a narrow band at low galactic latitudes (see Table 1) the $`\alpha ^{Iso}`$ scale factors are poorly constrained in the optimization process due to the dominating galactic components. So, we keep them fixed at values derived from a study of the extra-galactic $`\gamma `$-ray emission using a much larger database containing all EGRET Cycle I,II and III observations (Sreekumar et al. (1998)).
By optimizing the Likelihood under $`_1`$ with respect to its free scale parameters, $`\sigma ,\alpha ^{HI},\alpha ^{CO}`$ we can derived the flux and flux uncertainty from $`\sigma `$ and its error for a putative source at position ($`x_{sky},y_{sky}`$). From optimizations under $`_1`$ and $`_0`$ we can determine the Likelihood ratio $`\lambda `$ defined as $`2\mathrm{ln}(__0/__1)`$. This quantity is distributed as a $`\chi ^2`$ for 1 degree of freedom for a known source position and yields the source detection significance.
The MLR map for energies $`>100`$ MeV (Fig. 1) confirms the detection of the EGRET source 2EG J0220+4228 / 3EG J0222+4253 (Thompson et al. 1995; Hartman et al. 1999) at a $`10\sigma `$ significance level for 1 degree of freedom, i.e. the source position is known. The $`_1`$ and $`_0`$ hypotheses include also the contributions from well-established $`\gamma `$-ray sources (Hartman et al. (1999)) within a $`30\mathrm{°}`$ radius around our target in order to describe the $`\gamma `$-ray sky near our target optimally. The binned event matrix for this integral energy window is a combination of the matrices for the differential energy windows above 100 MeV, each with a different Earth zenith cut angle. The $`>100`$ MeV exposure matrix is in this case a power-law weighted composition (index $``$2.1) of the differential exposure matrices. This forms a consistent event/exposure set with respect to the applied selection criteria.
We compared the derived optimum scale factors for the Galactic diffuse emission components with the findings from more detailed studies on this diffuse emission (Strong & Mattox (1996), Hunter et al. (1997)) and found that our values are in all cases consistent with the published results.
It is evident from Fig. 1 that the high-energy $`\gamma `$-ray source is positionally consistent with both PSR J0218+4232 and 3C 66A (located at (l,b) = (140.143,$``$16.767)). The total excess contains $`225\pm 27`$ counts. We analyzed this excess also in the differential energy windows: 100-300 MeV, 300-1000 MeV and 1-10 GeV. In each window the source was seen: 100-300 MeV $`7.0\sigma `$ detection significance and $`138\pm 24`$ counts, 300-1000 MeV $`7.0\sigma `$ and $`57\pm 12`$ counts and finally 1-10 GeV $`6.5\sigma `$ and $`22\pm 6`$ counts. The location confidence contours for the excesses in the 3 broad energy windows are shown in Fig. 2. This figure shows that 3C 66A is the evident counterpart for the 1-10 GeV window (consistent with the third EGRET catalogue results (Hartman et al. (1999)), whereas PSR J0218+4232 is the most likely counterpart for the 100-300 MeV window. Between 300 and 1000 MeV both sources contribute to the excess.
For energies below 100 MeV we see indications for an excess, but the EGRET sensitivity is becoming too low and the spatial response too wide to draw firm conclusions. We estimated a $`2\sigma `$ flux upper limit for the spectrum of PSR J0218$`+`$4232 (see Sect. 8).
## 4 Long-term time variability
Earlier studies of $`\gamma `$-ray emission from spin-down powered pulsars showed that they are steady $`\gamma `$-ray emitters (see e.g. the review by Thompson et al. (1997)). On the contrary, most Active Galactic Nuclei appeared to be highly variable at $`\gamma `$-ray energies (see e.g. Mukherjee et al. (1997)). Therefore, we investigated whether there is (absence of) evidence for time varability of 2EG J0220+4228 / 3EG J0222+4253, particularly for the 100-300 MeV and 1-10 GeV bands, in which PSR J0218+4232 and 3C 66A appear to be the most likely counterparts, respectively.
Using integration intervals of typically 2 or 3 weeks, the results are shown in Fig. 3. The 100-300 MeV flux measurements are fully consistent with being constant, as expected for $`\gamma `$-ray emission from spin-down powered pulsars. The 1-10 GeV flux points show indications for variability and deviate at a $`2\sigma `$ level (93%) from being constant. According to the variability criteria defined by McLaughlin et al. (1996) the 1-10 GeV variability index $`V`$ of 1.33 points to a variable nature of the 1-10 GeV emission. This type of variability is indeed reminiscent of the behaviour observed frequently for the $`\gamma `$-ray emission from AGN. The above supports the conclusion from the spatial analysis, namely, that 2EG J0220+4228 / 3EG J0222+4253 is multiple: above 1 GeV the BL Lac 3C 66A is the obvious counterpart, whereas below 300 MeV PSR J0218+4232 is the most likely counterpart.
## 5 Timing analysis
In the timing analysis similar event selections have been applied as in the spatial analysis, except we ignored the specific TASC (Thompson et al. (1993)) flags of the event triggers in the event selection process. Especially the selection on the TASC zero cross overflow bit (set to 1 if less than 6.5 MeV is deposited in the TASC), which is only effective for the lower energy $`\gamma `$-ray photons ($`<150`$ MeV), is not taken into account. We verified this selection by a timing analysis of the Crab pulsar (combining many Cycle 0 $``$ VI VP’s) which showed that ignoring the TASC flags gives a significant improvement of the timing signal, particularly for energies below 100 MeV, with respect to the case in which we demand a TASC energy deposit of at least 6.5 MeV measured by one of its 2 PHA’s.
An additional difference in the selection procedure with the spatial analysis, where the spatial information of all events is used, is that we now have to specify an event extraction radius around the pulsar position. Contrary to what is commonly used in the timing analysis of EGRET data, namely, selecting events within an energy dependent extraction radius $`r_{ext}`$ of $`5\stackrel{}{.}85(E/100MeV)^{0.534}`$ containing approximately $`67\%`$ of the counts from a point-source, with $`E`$ the measured $`\gamma `$-ray energy, we optimized in each narrow energy window (e.g. 100-150 MeV) the signal-to-noise ratio $`S/N`$ as a function of extraction radius taking into account the modelled (2d) spatial distribution of the optimized diffuse models and neighbouring sources as obtained in the spatial analysis (see e.g. the thesis of Fierro (1995) p.49-50). This method provides the optimal extraction radius for a given energy window and a given sky-background structure. The values obtained from this study for the narrow energy windows between 100 and 1000 MeV are listed in Table 2.
From our timing observations of PSR J0218+4232 at radio wavelengths we obtained one single accurate ephemeris (rms error $`85\mu s`$), which is listed in Table 3. The validity interval of this ephemeris covers almost 5 years and in view of the stable rotation behaviour observed for millisecond pulsars its validity should extend far beyond the indicated range.
Phase folding the barycentered arrival times, taking into account the binary nature of the system, of the selected events with measured energies between 100-1000 MeV from all observations listed in Table 1 yields a $`3.5\sigma `$ modulation significance applying a $`Z_4^2`$ test (Buccheri et al. (1983)) on the unbinned sample of pulse phases. An H-test (de Jager et al. (1989)) in which the internal optimization of the number of harmonics is taken into account in the significance estimate yields a $`3.2\sigma `$ modulation significance at an optimum number of harmonics of 4. The 100-1000 MeV pulse profile is shown with 10 bins in Fig. 4 with superposed its Kernel Density Estimator (KDE; de Jager et al. (1986)) with the $`\pm 1\sigma `$ uncertainty interval. This KDE approaches the genuine underlying pulse profile (convolved with the instrumental time resolution) for an infinite number of events. The pulse profile shows one prominent narrow emission feature between phases $`0.6`$ and $`0.7`$ following a broad less prominent pulse with maximum at phase $`0.2`$. The phase separation of $`0.45`$ is remarkably similar to the value of $`0.47`$ observed at soft/medium energy X-rays by the ROSAT HRI (Kuiper et al. (1998)) and BeppoSAX MECS (Mineo et al. (2000); detailed comparisons will be presented below).
We also produced phase distributions in broader differential energy intervals. The pulse profiles for 100-300 MeV and 300-1000 MeV both showed consistently the same narrow and broad pulses ($`Z_4^2`$ probabilities $`2.5\sigma `$ and $`1.9\sigma `$, respectively). For 30-100 MeV and 1-10 GeV no hints for pulsation were found.
## 6 Pulse phase resolved spatial analysis
The pulse profile shown in Fig. 4 reaches a significance of $`3.5\sigma `$, indicating that the probability is low, only $`4.710^4`$, that this deviation from a flat distribution is caused by a random fluctuation. Given the importance of the discovery of high-energy $`\gamma `$-ray emission from a millisecond pulsar, we investigated further whether there is additional support in our data to claim this detection. As explained above, for the timing analysis the events were selected within an extraction radius around the position of PSR J0218+4232 using only $`56\%`$ of the events of a point source. In order to verify whether the source events outside the extraction radius ($`44\%`$) exhibit the same timing signature, we produced a pulse profile using all source events by performing a pulse phase resolved spatial analysis for energies between 100 and 1000 MeV.
The procedure is the following: Construct a pulse profile by repeating the spatial analysis for events selected in different pulse phase intervals. Contrary to the phase folding we need to select the events in relatively broad phase intervals to have sufficient statistics to do the spatial analysis: We selected 10 phase bins of width 0.1.
In order to estimate first the contribution of 3C 66A, which is obviously independent of the pulsar phase, to the total high-energy $`\gamma `$-ray excess in the 100-1000 MeV energy band we have fitted this excess for the full phase range in terms of point-sources at the positions of PSR J0218+4232 and 3C 66A. This yielded the following decomposition: the number of counts assigned to PSR J0218+4232 and 3C 66A are $`151\pm 52`$ and $`42\pm 51`$, respectively. The insignificant 3C 66A contribution, coming from events with energies $`>300`$ MeV, is nevertheless taken into account as a small correction in the pulse phase resolved spatial analysis (4.2 counts are assigned to 3C 66A for each 0.1 wide phase bin). Fitting then the measured 100-1000 MeV spatial event matrices for each pulse phase slice in terms of a PSR J0218+4232 model with a free scale factor atop the galactic diffuse models (both with free scale factors), the (fixed) isotropic extragalactic component and all (fixed) nearby-source models including 3C 66A, we obtain the total number of counts correlating with a point-source at the PSR J0218+4232 position for each phase slice.
The resulting 10 bin pulse profile is shown in Fig. 5. The total number of source counts in this light curve is 153 (the background level $``$ 0). Comparing Fig. 5 with the profile obtained from the timing analysis (Fig. 4), it is evident that the shape is statistically identical. For the phase folding we had selected only $`56\%`$ of the events for a real source (cf Table 2). Scaling from the number of 153 source counts measured in Fig. 5, to be consistent, the number of pulsar excess counts in Fig. 4 should be $`86`$, i.e. the backgound level should be at $`22`$. It is evident from this comparison that the two profiles are fully consistent in shape as well as in number of counts in the timing signature. Thus, the pulsed signal is also present outside the dataspace confined by the used extraction radius, as expected for a real signal, i.e. the timing and spatial signatures are consistent with the detection of PSR J0218+4232.
A more well-known display of the same conclusion are “ON”-“OFF” maps, or “pulsed”-“unpulsed” maps. Guided by the shape of the 100-1000 MeV pulse profile in a 20 bin representation (see Fig. 7e) we tentatively defined a “pulsed” phase interval as the combination of the phase ranges 0.05-0.40 and 0.55-0.70 and an “unpulsed” interval as its complement. We then produced MLR maps selecting the events now also on their phase location in either of the 2 pulse phase windows for the 100-300, 300-1000 and 100-1000 MeV energy ranges. The results are shown in Fig. 6. It is evident that the 100-300 MeV signal is confined within the “pulsed” interval, strengthening the conclusion that PSR J0218+4232 is the counterpart of 2EG J0220$`+`$4228 for energies between 100 and 300 MeV. In the 300-1000 MeV “unpulsed” MLR map $`4\sigma `$ residual emission is visible which can be explained by emission from 3C 66A and pulsed emission from PSR J0218+4232 emitted outside the defined “pulsed” interval (e.g. possible contribution from a weak pulse near phase 0.9 in Figs. 4 or 5). The overall picture for energies below 1000 MeV points to a very dominant PSR J0218+4232 and a minor 3C 66A contribution.
## 7 Multi-wavelengths profile comparisons
### 7.1 Comparison with radio profiles
The ephemeris of PSR J0218+4232 given in Table 3, and used for our $`\gamma `$-ray analysis, has been determined using Jodrell Bank observations at 610 MHz. The corresponding radio profile is shown in Fig. 7a (see also Stairs et al. 1999). It is remarkable that the pulsar is practically never “off”; three pulses seem to cover the entire phase range from 0 to 1.
Because the fiducial point in the 610 MHz radio profile defining the anchor point in the template used in the fitting process of the time of arrival of the radio pulses is known, its geocentric arrival time specified by the “Epoch of the period” in Table 3 can be translated to solar system barycentric arrival time. This timestamp is subsequently converted to a phase zero taking into account the binary nature of the system. This phase zero value, corresponding to the fiducial point, is finally subtracted from the $`\gamma `$-event phases, obtained by the same folding procedure, to align these with the radio profile. Thus, we can compare the 100-1000 MeV pulse profile in absolute phase with the 610 MHz radio profile. The aligned $`\gamma `$-ray pulse profile is shown in Fig. 7e, now in 20 bins to allow a more detailed comparison. The bin width of $`115\mu s`$ is comparable to the CGRO absolute timing accuracy of better than $`100\mu s`$. Also shown is the same KDE profile as shown in the 10 bin pulse profile in Fig. 4, to aid the comparison of the two $`\gamma `$-ray histograms, given the low counting statistics. In order to guide the eye, the pulsar phases of the three maxima in the 610 MHz radio profile are indicated by vertical lines.
In Fig. 7b is also indicated the 1410 MHz radio profile (Kramer et al. (2000)) which has been aligned by cross-correlation with the 610 MHz profile (phase uncertainty $`0.01`$ in alignment). It is clear from this figure that the 2 emission features in the $`\gamma `$-ray pulse profile coincide within the absolute timing uncertainties with 2 of the 3 pulses in the 610 MHz radio profile. Comparing the 610 and 1410 MHz radio profiles it is notable that one of these “radio/$`\gamma `$-ray” pulses (at phase 0.62) coincides with a dip in the 1410 MHz profile, followed and preceded by smaller pulses. Also between the two main emission features a shoulder is visible in the 1410 MHz profile which is absent in the 610 MHz one.
### 7.2 Comparison with X-ray profiles
We reported earlier significant detections of pulsed X-ray emission from PSR J0218+4232 analysing ROSAT HRI data ($`4.8\sigma `$ modulation significance in the 0.1-2.4 keV energy range; Kuiper et al. (1998)) and BeppoSAX MECS data ($`6.8\sigma `$, 1.6-10 keV energies; Mineo et al. (2000)). In the BeppoSAX MECS analysis we used the same ephemeris of Table 3 as in the present work. In the ROSAT HRI analysis, however, we used for the phase folding the extrapolated timing parameters from Navarro (1995). Given the availability of the new ephemeris which is valid over a nearly 5 year period and covers the ROSAT HRI observation, we decided for consistency reasons to reanalyze the 100 ks ROSAT HRI data. In addition, application of improved maximum likelihood algorithms in the spatial analysis to determine the centroid of emission in the X-ray map allowed for a better determination of the optimal extraction radius ($`8\mathrm{}`$). The result is shown in Fig. 7c. The modulation significance has increased to $`6\sigma `$ ($`Z_2^2`$ test), particularly the prominence of the second weaker pulse near phase 0.6 has improved in comparison with the result shown in Kuiper et al. (1998).
The new ROSAT HRI profile can be compared with the BeppoSAX MECS profile (Fig. 7d; Mineo et al. (2000)), which just overlaps in energy window. The alignment of the profiles was done by cross correlation, like in Mineo et al. (2000), since the uncertainties in the ROSAT and BeppoSAX absolute timing are too large to allow an absolute comparison. The identical peak separations of $`0.47`$ and the consistent difference in the spectra of the two peaks (Mineo et al. (2000)), make us confident that the alignment is accurate.
The next step is the alignment of the X-ray profiles with the absolute timing of the $`\gamma `$-ray and radio profiles. We cross correlated the most significant X-ray profile (from BeppoSAX MECS) with the EGRET profile, and applied the phase shift which corresponds to the highest probability in the correlation analysis to the aligned ROSAT HRI and BeppoSAX MECS profiles. These aligned profiles are shown in Fig. 7. in which the BeppoSAX MECS and EGRET profiles are both displayed in 20 bins. It is obvious that all three high-energy profiles exhibit two pulses with the same phase separation of about $`0.47`$. Fine structure in the gamma-ray profile, like the local maximum at phase $`0.9`$, is not significant, even though the strong radio pulse at phase $`0.9`$ makes that phase “special”.
## 8 Multi-wavelength spectrum
The X-ray spectrum of the pulsed emission from PSR J0218+4232 between 1.6 and 10 keV is the hardest measured so far for any (millisecond) radio pulsar. The best power-law fit to the BeppoSAX MECS pulsed spectrum has an index $`0.61\pm 0.32`$. The spectrum becomes somewhat softer (index $`0.94\pm 0.22`$) when a 27% DC component is included (Mineo et al. (2000)). This DC component is visible in the ROSAT data (Kuiper et al. (1998)) and in the BeppoSAX data up to 4 keV. Above 4 keV there is no sign of a DC component.
In the EGRET $`\gamma `$-ray data above 100 MeV, the signal seen from PSR J0218+4232 is also consistent with being 100% pulsed. However, the detailed structure of the pulse profile is not clear, i.e., is there a phase interval in which the $`\gamma `$-ray signal is clearly off, or how wide are the wings of the pulses? Possibly, the pulsed $`\gamma `$-ray signal extends over the total phase range with only one or two very narrow dips, just like in the radio profile. Therefore, it is difficult to determine a background region in the $`\gamma `$-ray pulse profile for the construction of a pulsed spectrum. We decided to determine the $`\gamma `$-ray spectrum using again the spatial maximum likelihood analysis, estimating the number of source counts (and then flux) on top of the diffuse background models and all relevant nearby sources, for the following energy intervals: 30-100 MeV, 100-300 MeV, 300-1000 MeV, 1-10 GeV. The resulting flux values and upper limits are given in Table 4 for PSR J0218$`+`$4232 and the simultaneously derived values for 3C 66A in Table 5 (power-law photon index $`1.5`$). Table 4 also lists the upper limits derived for the simultaneous COMPTEL observations and the OSSE observation during VP 728.7/9 (see Table 1). The COMPTEL $`2\sigma `$ -upper limits are derived in a spatial analysis analoguous to the EGRET approach. The OSSE $`2\sigma `$ \- upper limits are estimated from the statistically flat phase histograms according to the description presented in Ulmer et al. (1991) assuming a duty cycle of 0.5.
In Fig. 8 we have collected all available data for a total spectrum from radio up to high-energy $`\gamma `$-rays in the format $`E{}_{}{}^{2}\times flux`$, showing the observed power per logarithmic energy interval. The very high luminosity at $`\gamma `$-ray energies between 100 MeV and 1 GeV is striking and a large fraction of the total spin-down luminosity $`L_{\text{sd}}`$ will be emitted in high-energy $`\gamma `$-rays. This fraction $`\eta _{\text{obs}}`$ can be estimated as follows:
$$\eta _{\text{obs}}=L_\gamma /L_{\text{sd}}=\frac{1.6410^{34}(\mathrm{\Delta }\mathrm{\Omega }/1\text{sr})(d/5.7\text{kpc})^2}{2.3610^{35}(I/10^{45}\text{gcm}^2)}$$
with $`\mathrm{\Delta }\mathrm{\Omega }`$ the $`\gamma `$-ray beam size, $`d`$ the distance to the pulsar and $`I`$ the moment of inertia of the neutron star. Assuming $`\mathrm{\Delta }\mathrm{\Omega }=1\text{sr}`$, $`d=5.7\text{kpc}`$ and $`I=10^{45}\text{gcm}^2`$ we obtain an efficiency of $`7\%`$ for PSR J0218+4232. Over the 100-1000 MeV range the $`\gamma `$-ray spectrum is soft and consistent with a photon power-law index of $`2.6`$. The extrapolation of the very hard spectrum between 0.1 and 10 keV is just in agreement with the OSSE upper limit(s). Fig. 8 suggests that the maximum luminosity is reached in the COMPTEL MeV range just below the COMPTEL upper limits.
## 9 Summary and discussion
In this study we performed detailed spatial and timing analyses on PSR J0218$`+`$4232 using the high-energy $`\gamma `$-ray data from CGRO EGRET and found that we have good circumstantial evidence for the first detection of pulsed high-energy $`\gamma `$-rays from a Class II ms-pulsar, PSR J0218$`+`$4232 , namely:
* The spatial distribution is consistent with the pulsar being detected: Between 100 and 300 MeV the EGRET source position is consistent with that of PSR J0218$`+`$4232 with the total signal concentrated in 2 pulses. The 100-300 MeV flux does not show time variability at a 2/3 weeks time scale, indicative for a steady $`\gamma `$-ray emitter like spin-down powered pulsars. Above 1 GeV the nearby (angular separation $`1\mathrm{°}`$) BL Lac, 3C 66A, is the evident counterpart for the $`\gamma `$-ray excess. For energies between 300 MeV and 1 GeV the pulsar and the BL Lac contribute to the excess.
* Timing analysis (phase folding, using the timing parameters measured at radio wavelengths) in the 100-1000 MeV energy interval, selecting roughly $`56\%`$ of the source photons, yields a double-peaked pulse profile with a $`3.5\sigma `$ modulation significance. The same pulsed signature is also present in the data outside the extraction radius used in the timing analysis, containing the remaining $`44\%`$ of the source photons.
* The phase separation of $`0.45`$ of the two $`\gamma `$-ray pulses is similar to that measured between the two pulses at X-rays; a comparison in absolute time with the 610 MHz radio-profile shows alignment of the $`\gamma `$-ray pulses with two of the three radio pulses.
EGRET detected six pulsars with overwhelming statistical significance (Crab, Vela, Geminga, PSR B1706-44, PSR B1951+32 and PSR B1055-52; see e.g. the review by Thompson et al. 1997). Compared to these six, the modulation significance of PSR J0218$`+`$4232 falls only in the 3–4$`\sigma `$ range, similar to the significance of the weak timing signals found with EGRET from PSR B0656+14 (Ramanamurthy et al. (1996)) and PSR B1046-58 (Kaspi et al. (2000)). The additional circumstantial evidence for the detection of PSR J0218$`+`$4232 , particularly the similarity of the double-peaked X-ray and $`\gamma `$-ray pulse profile shapes, and the fact that the X-ray spectrum measured for PSR J0218$`+`$4232 below 10 keV is the hardest measured for any pulsar (Mineo et al. 2000) increases the likelihood of the detection. Nevertheless, confirmation of the detections of PSR B0656+14, PSR B1046-58 and PSR J0218$`+`$4232 by future high-energy $`\gamma `$-ray missions like the Italian AGILE and NASA’s GLAST is important.
The nearby 3C 66A obviously complicated the analyses, but its contribution to the $`\gamma `$-ray excess in the skymaps has consistently been taken into account. The events detected from this BL Lac have no systematic effect on the double-peaked timing signature assigned to PSR J0218$`+`$4232 in the timing analysis. However, our results show that earlier publications on the spectrum of 3C 66A (e.g. Dingus et al. (1996); Mukherjee et al. (1997); Lin et al. (1999)) should be revised, the time averaged spectrum is significantly harder than published earlier.
In Kuiper et al. (1998) and Mineo et al. (2000) the similarity of the double-peaked X-ray pulse profile of PSR J0218$`+`$4232 with that of the Crab pulsar was noted and discussed. It is now striking that the observed 100-1000 MeV pulse profile of PSR J0218$`+`$4232 shows one narrow ($`250\mu s`$) pulse preceded $`0.45`$ in phase by a broader pulse, again a morphology very similar to that of the Crab pulsar $`\gamma `$-ray profile. The latter exhibits two distinct pulses at $`0.4`$ phase separation at X-ray and $`\gamma `$-ray energies, with the X-ray and $`\gamma `$-ray pulses being aligned in absolute phase. Unfortunately, we cannot align the X-ray and $`\gamma `$-ray profiles of PSR J0218$`+`$4232 in absolute phase, but the similar phase separation suggests that the pulses are also aligned (see Fig. 7).
We noted in the Introduction that the surface magnetic field strengths of ms pulsars are 3 to 4 orders of magnitude weaker than that of normal radio pulsars. This makes the boundary condition for the production of $`\gamma `$-rays near the neutron star surface for ms pulsars much less favourable than for normal radio pulsars. It is, however, remarkable that the Crab pulsar and the members of the Class II ms pulsars have in common that the magnetic field strengths near the light cylinders $`B_{\text{lc}}`$ are comparable (in the range $`(310)\times 10^5`$ Gauß). In fact, ranking all known radio pulsars by $`B_{\text{lc}}`$, the three Class II ms pulsars rank number 1, 3 and 6, and Crab ranks number 2 (see also the discussion in Saito et al. (1997); Kuiper et al. (1998) and Takahashi et al. (1999)).
This is illustrated in Fig. 9, showing a scatter plot for all radio pulsars of $`B_{\text{lc}}`$ versus the spin-down flux, $`F_{\text{sd}}=\dot{E}/(4\pi d^2)`$, with $`\dot{E}`$ the total rotational energy loss rate and $`d`$ the distance. The three Class II ms pulsars are clearly located at the extreme of the $`B_{\text{lc}}`$ distribution. The two Class I ms pulsars possess significantly lower, more average values for ms pulsars. Also indicated are the 8 normal pulsars detected by EGRET in high-energy gamma-rays, as well as PSR B1509-58, detected by COMPTEL up to about 30 MeV (Kuiper et al. 1999b ). As has been noted in earlier papers, $`F_{\text{sd}}`$ is a good indicator for the probability to detect hard X-ray and high-energy gamma-ray emission from normal radio pulsars. The only normal pulsar near the top of the $`F_{\text{sd}}`$ distribution, not seen by EGRET is PSR B0540-69. This LMC pulsar is detected, however, at X-rays up to $`50`$ keV (Ulmer et al. (1999)). In order for ms pulsars to be seen with a hard X-ray spectrum (Class II), or even at high-energy gamma-rays (PSR J0218+4232) a high value for $`B_{\text{lc}}`$ seems to be required, in addition to a high $`F_{\text{sd}}`$. This suggests that $`B_{\text{lc}}`$ is a key parameter for models explaining the production of high-energy emission in the magnetospheres of ms pulsars.
Given in addition the similarities with the Crab of the high-energy pulse profiles (X-rays and also $`\gamma `$-rays for PSR J0218$`+`$4232 ) this suggests that the pulsed high-energy non-thermal emission from the Class II ms pulsars and the Crab pulsar have a similar origin in the pulsar magnetosphere, quite likely in a vacuum gap near the light cylinder. We know from radio observations, however, that the Crab has an orthogonal alignment, while PSR J0218$`+`$4232 is a nearly aligned rotator (Navarro et al. 1995, Stairs et al. 1999). Unfortunately, a parameter which is also important in this discussion on the geometry, the impact angle, has only been determined with large uncertainties, and therefore the line-of-sight information for PSR J0218$`+`$4232 is unconstrained (Stairs et al. 1999).
If indeed, X-ray emission and $`\gamma `$-ray emission from Class II ms pulsars has to be produced in a vacuum gap near the light cylinder, the vacuum gap has to be very short in order to have narrow and aligned pulses at X-rays and $`\gamma `$-rays, given the very strong curvature of the magnetic field lines in ms-pulsar magnetospheres. In addition, the potential drop has to be very high over this short length to accelerate the particles to the energies required for high-energy gamma-ray production. It is obvious that continuous acceleration of particles and production of X-rays and $`\gamma `$-rays from the surface of the neutron star along the curved magnetic field lines till the light cylinder radius (for PSR J0218$`+`$4232 only 111 km) will not render the narrow and aligned pulses at X-rays and $`\gamma `$-rays.
The Crab pulsar has also its two X-ray and $`\gamma `$-ray pulses aligned in absolute phase with two of the three radio pulses, leading to a consistent picture in which the high-energy pulses and the aligned radio pulses are produced in the same zones in the magnetosphere (see e.g. Romani & Yadigaroglu (1995)). The apparent alignment of the $`\gamma `$-ray pulses of PSR J0218$`+`$4232 with two of three pulses measured at 610 MHz suggests also that some of the radio pulses are produced in the same zones in the magnetosphere as the $`\gamma `$-ray pulses. However, we would first like to see a better radio estimate of the viewing angle for the PSR J0218$`+`$4232 system, and a confirmation of the absolute alignment using new and better observations at X-ray energies, before making further speculations on the geometry.
Theoretical models attempting to explain the high-energy electro-magnetic radiation from spin-down powered pulsars are divided in two main catagories distinguished by the production sites of the radiation in the pulsar magnetospheres. The first class of models, polar cap (PC) models, rely on the acceleration of charged particles along the open field lines near the magnetic pole(s) followed by cascade processes given rise to high-energy electro-magnetic radiation (see e.g Daugherty & Harding (1994), 1996). In the second class of models, outergap models (OG), the acceleration of charged particles and subsequent generation of high-energy radiation takes place in vacuum gaps near the pulsar light cylinder (see e.g. Cheng et al. 1986a,b and Ho (1989)). Unfortunately, for the case of ms pulsars no detailed self-consistent model calculations exist for either class of models, allowing predictions for different observational aspects, e.g. pulse phase resolved spectra, pulse shapes, efficiencies. In most cases only one aspect of the emergent high-energy radiation is addressed.
The PC model elaborated by a Polish group (Bulik, Dyks & Rudak), for example, only focusses on the emergent high-energy electro-magnetic spectrum from ms pulsars from X-rays up to high-energy $`\gamma `$-rays, while the pulse shape is ignored. This group predicts a dominating Synchrotron component over the entire X-ray/soft $`\gamma `$-ray band (0.1 keV - 1 MeV) with a spectral photon index of $`1.5`$ (Dyks & Rudak (1999)). This does not agree with the much harder photon indices of $`1`$ observed for PSR J0218+4232, PSR B1937+21 and PSR B1821-24. The predicted $`\gamma `$-ray spectrum, dominated by curvature radiation, peaks between 10 GeV and 100 GeV and even an inverse Compton scattering component is predicted at TeV energies (Bulik & Rudak (1999); Bulik et al. (2000)). The maximum in the observed spectrum of PSR J0218$`+`$4232 ($`\nu F_\nu `$ or $`E^2F`$ representation) is located in the 1–100 MeV range, also in contradiction with their model prediction (see Fig. 8). Their $`\gamma `$-ray flux prediction for PSR J0218$`+`$4232 is even more than a factor of 10 below the expected GLAST sensitivity level, thus not at all detectable by the less sensitive EGRET telescope for which we present the results.
The polar cap cascade model of Zhang & Harding (2000) including now also, compared to earlier versions, inverse Compton scattering of higher generation cascade pairs provides predictions for both the X-ray and $`\gamma `$-ray luminosities of spin-down powered pulsars, including ms pulsars. In the soft/medium energy X-ray band the model predicts a thermal origin of the spectral features of the pulsed emission from ms pulsars. This is inconsistent with the observed non-thermal (very) hard pulsed spectra of the 3 Class II ms pulsars. However, for the Class I ms pulsars this could be in agreement with the observed spectral properties.
Zhang and Harding also predict that ms pulsars usually have a considerable high-energy $`\gamma `$-ray luminosity, but due to their weak magnetic field strengths, resulting in quite high photon escape energies, the emergent $`\gamma `$-spectrum is very hard. The latter is not in agreement with the observed soft high-energy (photon Power-law index of $`2.6`$ for energies between 100 MeV and 1 GeV) $`\gamma `$-ray spectrum of PSR J0218$`+`$4232 .
Thus, so far the PC scenario based models appear to be unsuccessful in explaining the observed X-ray and $`\gamma `$-ray properties of the Class II ms pulsars.
An outergap model aiming at predicting pulsed and unpulsed $`\gamma `$-ray emission from ms pulsars was presented by Wei et al. (1996). This model predicts a spectral photon index of $`2`$ for the pulsed emission from a ms pulsar for the energy range of $`10`$ keV to $`500`$ MeV, in contradiction with the spectrum we show in Fig. 8 for PSR J0218$`+`$4232 . The model predicts also a harder unpulsed component with a spectral photon index of $`1.5`$, dominating the pulsed component above $`500`$ MeV. We have not detected this component for PSR J0218$`+`$4232 at energies above 100 MeV.
Concerning the energetics of the $`\gamma `$-ray emission of PSR J0218$`+`$4232 it is interesting to compare the observed $`\gamma `$-ray efficiency $`\eta _{\text{obs}}`$ (fraction of the total spin-down luminosity) of $`0.07`$ with theoretically derived efficiencies. For the PC model of Zhang & Harding (2000) the efficiency scales as $`\eta _{\text{PC}}p\tau ^{0.5}`$ with $`p`$ the pulse period and $`\tau `$ the characteristic age of the pulsar. Expressed in the Crab pulsar efficiency $`\eta _{\text{Crab}}`$ we find for PSR J0218$`+`$4232 that $`\eta _{\text{0218}}45\times \eta _{\text{Crab}}`$, which translates to an efficiency of $`\eta _{\text{0218}}0.05`$ substituting the measured Crab $`\gamma `$-ray efficiency of about $`0.001`$. The thick OG model of Zhang & Cheng (1998) yields the following expression for the $`\gamma `$-ray efficiency: $`\eta _{\text{OG}}p^2\tau ^{6/7}`$. This translates to $`\eta _{\text{0218}}300\times \eta _{\text{Crab}}`$, which means that $`\eta _{\text{0218}}0.33`$. Thus, within the framework of both PC and OG models the expected $`\gamma `$-ray conversion efficiency is very high, approximately in accordance with the measured efficiency of about $`0.1`$. However, it should be noted that both models predict an even higher efficiency for e.g. PSR J0437-4715, a Class I ms pulsar. This pulsar is very nearby but has not been detected as a $`\gamma `$-ray source/pulsar (Fierro et al. (1995)).
The circumstantial evidence presented in this paper for the detection of pulsed high-energy $`\gamma `$-rays from ms pulsar PSR J0218$`+`$4232 opens a new window in the study of the magnetospheric properties of spin-down powered pulsars. It is unfortunate that we cannot repeat this observation with EGRET anymore. Therefore, deep searches for high-energy $`\gamma `$-ray emission from the Class II ms pulsars with future more sensitive gamma-ray missions like GLAST and AGILE are very important. But also earlier sensitive observations at the harder X-rays above 10 keV are very important to bridge the observational gap. Particularly the ESA mission INTEGRAL might be able to extend the hard spectra measured below 10 keV to as high as a few MeV.
###### Acknowledgements.
This work is supported by the Netherlands Organisation for Scientific Research (NWO). We thank Michael Kramer for providing the 1400 MHz radio profile of PSR J0218$`+`$4232 obtained with the Effelsberg radio telescope.
|
warning/0005/math0005037.html
|
ar5iv
|
text
|
# References
Finite Groups Embeddable in Division Rings
T. Y. Lam
§1. Introduction
Finite groups that are embeddable in the multiplicative groups of division rings $`K`$ were completely determined by S. A. Amitsur \[Am\] in 1955. In case $`K`$ has characteristic $`p>0`$, the only possible finite subgroups of $`K^{}`$ are cyclic groups, according to a theorem of I. N. Herstein \[He\]. Thus, the only interesting case is when $`K`$ has characteristic $`\mathrm{\hspace{0.17em}0}`$; that is, when $`K`$.
In \[He\], Herstein conjectured that odd-order subgroups of division rings $`K`$ were cyclic, and he proved this to be the case when $`K`$ is the division ring of the real quaternions. Herstein’s conjecture was settled negatively in \[Am\]. As part of his complete classification of finite groups in division rings, Amitsur showed that the smallest noncyclic odd-order group that can be embedded in a division ring is one of order $`63`$ (and this group is unique).
Amitsur’s paper is daunting to read as it is long and technically complicated. In lecturing to a graduate class on division rings, I tried to find a simple reason for the “first exceptional odd order” $`63`$ (to Herstein’s conjecture). After some work, I did come up with a reason that was simple enough to be explained to my class, without having to go through any part of Amitsur’s paper. Furthermore, the method I used led easily to the second exceptional odd order, $`117`$ (which was not mentioned in Amitsur’s paper). Since this line of reasoning did not seem to have appeared in the literature before, I record it in this short note. To better motivate the results discussed here, I have also included a quick exposition on the beginning part of the theory of finite subgroups of division rings. As a result, this paper can be read independently of \[Am\].
§2. Background on Amitsur Groups
Let us say that a finite group $`G`$ is Amitsur if $`G`$ can be embedded in the multiplicative group $`K^{}`$ of some division ring $`K`$. We start with some background information on such Amitsur groups. First the following easy observation.
(2.1) Theorem. Any abelian Amitsur group $`G`$ is cyclic.
Proof. Say $`GK^{}`$, where $`K`$ is a division ring. If $`F`$ is the center of $`K`$, then the $`F`$-span $`L`$ of $`G`$ is a finite-dimensional $`F`$-algebra. Since $`L`$ is a commutative domain, it must be a field. From field theory (see, e.g., \[Ja : p. 132\]), it then follows that $`GL^{}`$ is cyclic. QED
Next, we classify Amitsur $`p`$-groups. (Here, $`p`$ denotes any prime number, unrelated to the characteristic of our division rings.) This is possible thanks to some standard results in $`p`$-group theory.
(2.2) Theorem. Let $`G`$ be an Amitsur $`p`$-group. Then $`G`$ is cyclic if $`p`$ is odd, and $`G`$ is cyclic or dicyclic if $`p=2`$.
Recall that a group is dicyclic (of order $`\mathrm{\hspace{0.17em}4}n`$) if it is generated by two elements $`x,y`$ subject to the relations $`x^{2n}=1,y^2=x^n`$, and $`yxy^1=x^1`$, where $`n2`$. Dicyclic groups are also known as generalized quaternion groups; they are $`\mathrm{\hspace{0.17em}2}`$-groups iff $`n`$ is a power of $`\mathrm{\hspace{0.17em}2}`$. It is easy to see that any dicyclic group $`x,y`$ as above is embeddable in the division ring of real quaternions, so the conclusion in (2.2) is indeed the best possible.
Proof of (2.2). In the theory of $`p`$-groups, there is a well-known result which guarantees that, if a $`p`$-group has a unique subgroup of order $`p`$, then it satisfies the conclusion of the present theorem; see Th. 12.5.2 in \[Ha\]. Thus, it suffices for us to verify that, if $`G\{1\}`$, $`G`$ has a unique subgroup of order $`p`$. Say $`GK^{}`$, where $`K`$ is a division ring, with center $`F`$.
By elementary group theory, we know that $`G`$ has a nontrivial center. Fix a central element $`xG`$ of order $`p`$, so $`|x|=p`$. If $`yG`$ is any subgroup of order $`p`$, then $`L:=F(x,y)`$ is a field. In this field, the equation $`t^p1=0`$ has at most $`p`$ solutions, so we must have $`y=x`$, as desired. QED
In order to get some general results for non $`p`$-groups, we introduce the following class of finite groups, where $`m,n,r`$ are natural numbers:
$$G_{m,n,r}=a,b:a^m=b^n=1,bab^1=a^r.$$
$`(2.3)`$
Whenever we use this notation, it will be assumed that $`r^n1(\text{mod}m)`$. This assumption guarantees that $`G_{m,n,r}`$ is a semidirect direct product of $`a`$ and $`b`$, with $`|a|=m`$ and $`|b|=n`$. In particular, $`|G_{m,n,r}|=mn`$. If necessary, we may assume $`\mathrm{\hspace{0.17em}0}r<m`$; but this is not always essential.
Let us say that a finite group $`G`$ is Sylow-cyclic if all of its Sylow subgroups are cyclic.<sup>1</sup><sup>1</sup>1Such groups are also called $`Z`$-groups in the literature, after Hans Zassenhaus. We use here the term “Sylow-cyclic” since it is completely self-explanatory. For instance, any $`G`$ with a square-free order is Sylow-cyclic, since each of its Sylow groups has a prime order. The following classical result of Hölder, Burnside, and Zassenhaus (see Th. 9.4.3 in \[Ha\]) gives a complete determination of all Sylow-cyclic finite groups.
(2.4) Hölder-Burnside-Zassenhaus Theorem. A finite group $`G`$ is Sylow-cyclic iff $`GG_{m,n,r}`$ for some $`m,n,r`$ with $`(m,n(r1))=1`$. (In particular, Sylow-cyclic groups are solvable.)
Coupling this powerful classification result with (2.2), we deduce the following important information on a class of Amitsur groups, including all the ones of odd order.
(2.5) Theorem. If $`G`$ is an Amitsur group with order not divisible by $`8`$, then $`G`$ is Sylow-cyclic, and hence $`GG_{m,n,r}`$ for some $`m,n,r`$ with $`(m,n(r1))=1.`$
According to this result, Amitsur groups with order not divisible by $`8`$ are “fairly close” to being cyclic: they are at least metacyclic, i.e., extensions of cyclic groups by cyclic groups. To find all odd-order Amitsur groups, for instance, we must determine, for $`m,n`$ odd, which of the metacyclic groups $`G_{m,n,r}`$ (with $`(m,n(r1))=1`$) are Amitsur. In the next section, we shall present some partial results in this direction, not only for odd-order groups, but for the groups $`G_{m,n,r}`$ in (2.3) in general. Note that the groups $`G_{m,n,r}`$ in this paper are “simpler versions” of the groups $`G_{m,r}`$ studied by Amitsur, in that the $`G_{m,n,r}`$’s are given as semidirect products, while Amitsur’s $`G_{m,r}`$’s are not. As we shall see in the next section, the semidirect product representation of $`G_{m,n,r}`$ makes it possible for us to obtain quickly some necessary conditions for such a group to be Amitsur.
§3. Necessary Conditions for the Embeddability of $`G_{m,n,r}`$
In this section, we shall study the groups $`G_{m.n,r}`$ that are embeddable in division rings. (We do not need to assume the condition $`(m,n(r1))=1`$.) The main result here is first stated in the following negative fashion.
(3.1) Theorem. Let $`G=G_{m,n,r}`$, where $`n>1`$ and $`o(\overline{r})`$ (the order of $`\overline{r}`$ in the unit group $`\text{U}(_m)`$) is exactly $`n`$. Then $`G`$ is not Amitsur.
Proof. Assume, instead, that $`GK^{}`$, where $`K`$ is a division ring. We use the presentation (2.3) for $`G`$. From $`bab^1=a^r`$, we get $`b^iab^i=a^{r^i}`$. Thus, $`b^ia=a^{r^i}b`$, and $`a^1b^ia=a^{r^i1}b^i`$. From $`b^n=1`$, we get $`b^{n1}+\mathrm{}+b+1=0`$ in $`K`$. Conjugating this equation by $`a`$, we get
$$a^{r^{n1}1}b^{n1}+a^{r^{n2}1}b^{n2}+\mathrm{}+a^{r1}b+1=0.$$
$`(3.2)`$
Subtracting the last two equations and cancelling $`b`$ (from the right) gives
$$\left(a^{r^{n1}1}1\right)b^{n2}+\left(a^{r^{n2}1}1\right)b^{n3}+\mathrm{}+(a^{r1}1)=0.$$
$`(3.3)`$
Conjugating this by $`a`$ again and subtracting (3.3) from the resulting equation, we get after another cancellation of $`b`$:
$$\left(a^{r^{n1}1}1\right)\left(a^{r^{n2}1}1\right)b^{n3}+\mathrm{}+(a^{r^21}1)(a^{r1}1)=0.$$
Carrying this process to the bitter end produces the equation
$$\left(a^{r^{n1}1}1\right)\left(a^{r^{n2}1}1\right)\mathrm{}(a^{r^21}1)(a^{r1}1)=0K.$$
$`(3.4)`$
Therefore, $`a^{r^i1}=1`$ for some $`i[1,n1]`$. But then $`r^i1(\text{mod}m)`$, which contradicts $`o(\overline{r})=n`$. QED
In my lectures, I referred to the above as the “triple $`(1)`$-proof”, because of the three remarkable layers of $`(1)`$’s in the first factor of the key equation (3.4)! Note that this proof actually showed that, under the assumption that $`o(\overline{r})=n>1`$, $`G_{m,n,r}`$ cannot be multiplicatively embedded in any domain (that is, a nonzero ring without 0-divisors). This, however, is not a real improvement, since it is easily seen that, once a finite (multiplicative) group $`G`$ is embeddable in a domain, then $`G`$ is in fact embeddable in a division ring.
We now record some consequences of (3.1).
(3.5) Corollary. If $`G=G_{m,n,r}`$ (generated by $`a,b`$ as in $`(2.3)`$) is Amitsur, then the largest square-free subgroup of $`b`$ must be central in $`G`$. In particular, if $`G=G_{m,n,r}`$ is Amitsur and $`n`$ is square-free, then $`G`$ is cyclic.
Proof. For any prime $`\mathrm{}`$ dividing $`n=o(b)`$, let $`c`$ be the $`\mathrm{}`$-Sylow group of $`b`$, say of order $`\mathrm{}^t`$. Then $`cac^1=a^s`$ for some $`s`$, and $`acG_{m,\mathrm{}^t,s}`$ is still Amitsur. By (3.1), we must have $`o(\overline{s})<\mathrm{}^t`$ in $`\text{U}(_m)`$. This means that $`c^{\mathrm{}^{t1}}`$ acts trivially on $`a`$ (by conjugation), and hence $`c^{\mathrm{}^{t1}}`$ is central in $`G`$. From this, the first conclusion in the Corollary follows. If $`n`$ happens to be square-free, this conclusion means that $`b`$ is central in $`G`$. Thus, $`G`$ is abelian, and hence cyclic by (2.1). QED
(3.6) Corollary. Let $`G`$ be an Amitsur group of order $`p^tq_1\mathrm{}q_k`$, where $`k1`$, and $`p>q_1>\mathrm{}>q_k`$ are primes. Then $`G`$ is cyclic.
Proof. Since $`k1`$, $`p`$ is an odd prime. Thus, $`G`$ is Sylow-cyclic, and hence solvable by (2.4). We proceed by induction on $`k`$. Note that the solvability of $`G`$ implies, by Philip Hall’s Theorem, that it has a subgroup $`H`$ of order $`m:=p^tq_1\mathrm{}q_{k1}`$ (see \[Ha: Th. 9.3.1\]). This Hall subgroup is still Amitsur, and hence cyclic by the inductive hypothesis. Also, since $`[G:H]=q_k`$ is the smallest prime divisor of $`|G|`$, $`H`$ must be normal in $`G`$. Thus, by taking a generator of $`H`$ and an element of order $`q_k`$ in $`G`$, we can represent $`G`$ as $`G_{m,q_k,r}`$ for some $`r`$. Since $`q_k`$ is a prime, (3.5) implies that $`G`$ is cyclic. Note that this argument is also sufficient to get the induction started, so the proof is complete. QED
A special case of (3.6) is that any square-free Amitsur group is cyclic. This recaptures Corollary 5 on p. 384 of \[Am\].
Note that, in (3.1) and (3.5), the integers $`m,n`$ and $`r`$ were arbitrary. To get some explicit numerical results, let us specialize these results to odd-order groups.
(3.7) Theorem. If $`G`$ is an Amitsur group of odd order $`<171`$, then $`G`$ is cyclic except possibly when $`|G|=63or\mathrm{\hspace{0.33em}\hspace{0.17em}117}.`$
Proof. If $`|G|`$ is either a prime power, or square-free, or of the form $`p^tq`$ where $`p>q`$ are primes, the foregoing results imply that $`G`$ is cyclic. Among odd integers from $`1`$ to $`169`$, the only ones not of any of the above forms are
$$53^2=45,\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.17em}7}3^2=63,\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.17em}11}3^2=99,\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.17em}13}3^2=117,\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.17em}5}3^3=135\text{and}\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.17em}17}3^2=153.$$
If $`|G|=45`$ or $`99`$, $`G`$ is easily seen to be abelian (by Sylow theory), and hence cyclic (by (2.1)). Next, consider $`|G|=135`$. Since $`G`$ is Amitsur, we can represent it (thanks to (2.5)) in the form $`G_{m,n,r}`$ with $`(m,n)=1`$. We may assume $`m,n>1`$, so we have $`m=5`$ or $`\mathrm{\hspace{0.17em}3}^3`$. But $`\text{U}(_5)`$ has no element of order $`\mathrm{\hspace{0.17em}3}`$, and $`\text{U}(_{3^3})`$ has no element of order $`\mathrm{\hspace{0.17em}5}`$. Therefore, $`G`$ must be abelian and hence cyclic. The case $`|G|=153`$ can be treated similarly (with exactly the same conclusion). Thus, $`G`$ is cyclic in all cases, except possibly when $`|G|=63`$ or $`\mathrm{\hspace{0.17em}117}`$. QED
What about $`|G|=63`$? There are, up to isomorphism, four groups of order 63; they are:
$$_{63},_3\times _{21},_3\times G_{7,3,2},\text{and}G_{7,9,2}.$$
The first group is cyclic, and hence Amitsur. The second and third ones are not Sylow-cyclic, and hence not Amitsur. The remaining question is whether the non-cyclic (but Sylow-cyclic) $`G_{7,9,2}`$ is Amitsur. Note that, since $`o(\overline{2})=3<9`$ in $`\text{U}(_7)`$, (3.1) does not apply to this group. Thus, there is a chance that $`G_{7,9,2}`$ is Amitsur. Indeed, the embeddability of this group was proved by Amitsur, and subsequent expositions on this have been given by C. Ford \[Fo\] and J. Dauns \[Da\]. For the sake of completeness, we give a quick sketch for the embeddability of $`G_{7,9,2}`$ below.
Let $`L=(\zeta )`$ where $`\zeta `$ is a primitive 21st root of unity, and let $`\sigma \text{Gal}(L/)`$ be defined by $`\sigma (\zeta )=\zeta ^{16}`$. Then $`o(\sigma )=3`$, so $`L`$ is a cubic extension of the fixed field $`F:=L^\sigma `$. Note that $`F`$ contains the primitive cubic root of unity $`\omega :=\zeta ^7`$, since
$$\sigma (\omega )=\zeta ^{716}=\zeta ^7=\omega .$$
On the other hand, for the primitive 7th root of unity $`a:=\zeta ^3`$, we have
$$\sigma (a)=\zeta ^{316}=\zeta ^6=a^2.$$
Now introduce a new symbol $`b`$, and form the cyclic $`F`$-algebra
$$K:=LLbLb^2,\text{with}b^3=\omega ,\text{and}b\mathrm{}=\sigma (\mathrm{})b(\mathrm{}L),$$
which has center $`F`$. We have $`\text{dim}_{}F=\phi (21)/3=4`$, so $`\text{dim}_{}K=3^24=36.`$ In the algebra $`K`$, the elements $`a,b`$ satisfy the equations
$$a^7=1,b^3=\omega ,b^9=\omega ^3=1,\text{and}ba=\sigma (a)b=a^2b.$$
Therefore, $`a,`$ and $`b`$ generate a group $`GG_{7,9,2}`$ in the group of units $`\text{U}(K)`$. It can be checked that $`K`$ is a division $`F`$-algebra (see, for instance, \[Fo\]), so this establishes the noncyclic $`G_{7,9,2}`$ as an Amitsur group (of order $`63`$). The center $`F`$ of the division ring $`K`$ is easily seen to be the biquadratic extension $`(\sqrt{3},\sqrt{7})`$ of the rationals.
As for the case $`|G|=117`$, we have a noncyclic candidate $`G_{13,9,9}`$ (noting that $`o(\overline{9})=3`$ in $`\text{U}(_{13})`$). This group can be shown to be Amitsur in the same way as $`G_{7,9,2}`$ was.
A substantial part of the argument in showing that a certain group is Amitsur is to establish that a certain $``$-algebra is a division algebra. This is by no means routine. In Amitsur’s paper, this part of the work is handled by number-theoretic tools, such as the Hasse Norm Theorem and the Hasse-Brauer-Noether Theorem. Thus, to fully understand Amitsur’s work would require considerable preparation in algebraic number theory.
In closing, we should mention that it is also of interest to fix a field $`F`$ (of characteristic zero), and to determine the finite groups $`G`$ that are embeddable in $`K^{}`$ for some finite-dimensional central $`F`$-division algebra $`K`$. This problem has been studied by B. Fein and M. Schacher \[FS\], who called such $`G`$$`F`$-adequate groups”. The embedding constructed for $`G=G_{7,9,2}`$ above showed that $`G`$ is $`(\sqrt{3},\sqrt{7})`$-adequate, but Fein and Schacher showed that $`G`$ is already $`\left(\sqrt{3}\right)`$-adequate. This is remarkable, since Fein and Schacher have also proved the following interesting positive result on Herstein’s conjecture: If $`[F:]2`$, then any $`F`$-adequate odd-order group is cyclic, except when $`F=\left(\sqrt{3}\right)`$!
University of California
Berkeley, Ca 94720
lam@math.berkeley.edu
|
warning/0005/cond-mat0005288.html
|
ar5iv
|
text
|
# Gaussian Tunneling Model of 𝑐-Axis Twist Josephson Junctions
\[
## Abstract
We calculate the critical current density $`J_c^J`$ for $`c`$-axis Josephson tunneling between identical high temperature superconductors twisted an angle $`\varphi _0`$ about the $`c`$-axis. We model the tunneling matrix element squared as a Gaussian in the change of wavevector $`𝐪`$ parallel to the junction, $`|t(𝐪)|^2\mathrm{exp}(𝐪^2a^2/2\pi ^2\sigma ^2)`$. The $`J_c^J(\varphi _0)/J_c^J(0)`$ obtained for the $`s`$\- and extended-$`s`$-wave order parameters (OP’s) are consistent with the Bi<sub>2</sub>Sr<sub>2</sub>CaCu<sub>2</sub>O<sub>8+δ</sub> data of Li et al., but only for strongly incoherent tunneling, $`\sigma ^20.25`$. A $`d_{x^2y^2}`$-wave OP is always inconsistent with the data. In addition, we show that the apparent conventional sum rule violation observed by Basov et al. might be understandable in terms of incoherent $`c`$-axis tunneling, provided that the OP is not $`d_{x^2y^2}`$-wave.
\]
Recently, there has been a resurgence of interest in the symmetry of the order parameter (OP) in the high temperature superconductors (HTSC). Although many phase-sensitive experiments were interpreted as giving evidence for an OP in YBa<sub>2</sub>Cu<sub>3</sub>O<sub>7-δ</sub> (YBCO) that was consistent with the $`d_{x^2y^2}`$-wave form, the same type of phase-sensitive experiments on the electron-doped HTSC Nd<sub>1.85</sub>Ce<sub>0.15</sub>CuO<sub>4-y</sub> and Pr<sub>1.85</sub>Ce<sub>0.15</sub>CuO<sub>4-y</sub> (PCCO) also were interpreted in terms of a $`d_{x^2y^2}`$-wave OP, in apparent contradiction with other results, suggesting a possible problem with the former experiments. In addition, new phase-sensitive experiments on Bi<sub>2</sub>Sr<sub>2</sub>CaCu<sub>2</sub>O<sub>8+δ</sub> (Bi2212) gave strong evidence that the OP contains an isotropic component, since it has a non-vanishing average over its Fermi surface. First, $`c`$-axis Josephson tunneling between Bi2212 and either Pb or Nb demonstrated that an isotropic $`s`$-wave component of the OP exists in Bi2212. Second, $`c`$-axis tunneling across the junctions of Bi2212 intercalated with HgBr<sub>2</sub> has been studied using mesas. In these experiments, increasing the $`c`$-axis spacing by 6.3Å increased the normal state resistivity by a factor of 200. In the superconducting state, $`R_n`$ and $`I_c`$ changed by comparable factors, but their product $`I_cR_n`$ 10 mV, about half the optimal value expected in the Ambegaokar-Baratoff (AB) model of purely incoherent $`c`$-axis tunneling between identical $`s`$-wave superconductors. Such behavior is very difficult to understand in terms of a $`d_{x^2y^2}`$-wave OP.
More directly, a new phase-sensitive experiment which can test the symmetry of the OP over the entire range of temperatures $`T`$ below the superconducting transition temperature $`T_c`$ was performed. In this experiment, a single crystal of Bi2212 was cleaved in the $`ab`$-plane, the two cleaves were twisted a chosen angle $`\varphi _0`$ about the $`c`$-axis with respect to each other, and fused back together. After lead attachment, the critical currents $`I_c^J(T)`$ and $`I_c^S(T)`$ across the twist junction and single crystals were measured, and their ratio was found to be directly proportional to the ratio $`A^J/A^S`$ of their areas at $`T/T_c=0.9`$. Those authors then claimed that (a) the intrinsic junctions and the twist junction behaved identically, (b) the $`c`$-axis tunneling is strongly incoherent, and (c) the OP contains an isotropic component, but not any purported $`d_{x^2y^2}`$-wave component for $`T<T_c`$, except possibly below a second, unobserved phase transition.
Since then, the group theoretic arguments upon which conclusion (c) were based have been published. In addition, an exact calculation of the possible roles of coherent $`c`$-axis tunneling has been presented. For the tight-binding Fermi surface generally thought to be applicable to Bi2212, , it was shown that such coherent tunneling was inconsistent with the data, even for an isotropic $`s`$-wave OP. Since the claim (a) of Li et al. is just a statement of their experimental observations, and the claim (b) is clearly correct in the limit of purely incoherent tunneling, it remains to quantify precisely just how incoherent the tunneling must be in order to fit the data.
Since Bi2212 behaves as a stack of weakly-coupled Josephson junctions, the critical current density $`J_c`$ across each junction may be evaluated by neglecting the couplings between the other junctions. Specifically, $`J_c^J`$ across the twist junction between layers 1 and 2 is given by
$$J_c^J=|4eT\underset{\omega }{}f^J(𝐤𝐤^{})F_1(𝐤_J)F_2^{}(𝐤_{𝐉}^{}{}_{}{}^{})|,$$
(1)
where $`\omega `$ represents the Matsubara frequencies, $`f^J(𝐤𝐤^{})`$ is the spatial average of the tunneling matrix element squared, $`\mathrm{}`$ represents two-dimensional integrals over each of the two first Brillouin zones (BZ’s), and the wavevectors $`𝐤_J`$ and $`𝐤_J^{}`$ are obtained from $`𝐤=(k_x,k_y)`$ and $`𝐤^{}=(k_x^{},k_y^{})`$ by rotations of $`\pm \varphi _0/2`$ about the $`c`$-axis, respectively. The anomalous Green’s functions $`F_n=\mathrm{\Delta }_n/[\omega ^2+\xi _n^2+|\mathrm{\Delta }_n|^2]`$ and $`F_n^{}=F_n^{}`$, where $`\mathrm{\Delta }_n(𝐤)`$ and $`\xi _n(𝐤)`$ are the OP and quasiparticle dispersion on the $`n`$th layer, respectively. For Bi2212, we assume the quasiparticle dispersions $`\xi _n`$ have identical tight-binding forms, differing only by the wavevector rotations,
$`\xi _n(𝐤)`$ $`=`$ $`t[\mathrm{cos}(k_xa)+\mathrm{cos}(k_ya)]`$ (3)
$`+t^{}\mathrm{cos}(k_xa)\mathrm{cos}(k_ya)\mu ,`$
where we take $`t=`$306 meV, $`t^{}/t=0.90`$ and $`\mu /t=0.675`$ to give a good fit to the Fermi surface of Bi2212, for which $`\xi _n(𝐤_F)=0`$. These values are slightly different from those used previously. A plot of this Fermi surface is shown in Fig. 1. We study three OP’s. These are the isotropic $`s`$-wave, a constant, the $`d_{x^2y^2}`$-wave, topologically equivalent to $`\mathrm{cos}(k_xa)\mathrm{cos}(k_ya)`$, or an extended-$`s`$-wave, which we take to be the absolute magnitude of the particular $`d_{x^2y^2}`$-wave form. To obtain the particular $`d_{x^2y^2}`$-wave form we used the repulsive interaction of the form
$$V(𝐪)=V_0\underset{𝐐=(\pm 1,\pm 1)\pi /a}{}\mathrm{\Gamma }/[(𝐪𝐐)^2+\mathrm{\Gamma }^2],$$
(4)
where $`V_0=556`$ meV and $`\mathrm{\Gamma }=0.1`$. The BCS equation is solved for $`\mathrm{\Delta }(𝐤)`$ in terms of $`V(𝐤𝐤^{})`$. The extended-$`s`$-wave OP was taken to have the absolute magnitude of the $`d_{x^2y^2}`$-wave form obtained from Eq. (4). The isotropic $`s`$-wave OP was taken to have the maximum magnitude of it. We included the umklapp terms which occur with rotated Fermi surfaces, but the effects of doing so were very small in all cases we studied.
Previously, we investigated the case in which $`f^J(𝐤𝐤^{})\delta (𝐤𝐤^{})`$, appropriate for coherent tunneling. We found that it made very little difference whether or not one included umklapp terms in the integrals in Eq. (1). However, we found that for Fermi surfaces similar to that pictured in Fig. 1, it was not possible to obtain a quantitative fit to the data. We then argued that for extremely incoherent tunneling, with $`f^J(𝐤𝐤^{})=f_0^J`$, a constant, the $`s`$-wave and extended-$`s`$-wave OP’s could both fit the data equally well. However, we did not discuss the intermediate case in any detail, and that is our purpose here. We therefore take the tunneling matrix element squared to be a Gaussian in the momentum change parallel to the junction,
$$f^J(𝐤𝐤^{})=f_0^J\mathrm{exp}[(𝐤𝐤^{})^2a^2/2\pi ^2\sigma ^2].$$
(5)
Since $`\sigma =0`$ and $`\sigma \mathrm{}`$ result in purely coherent and incoherent tunneling, $`\sigma `$ is a dimensionless parameter quantifying the incoherence of the tunneling. In Fig. 1, the shaded concentric circles for $`\sigma ^2=0.005`$ and 0.050 are pictured for the case in which the tunneling on one side of the junction occurs from the intersection of the Fermi surface and the line connecting the $`(0,0)`$ and $`(\pi ,\pi )`$ points in the BZ. In the shaded regions, $`|𝐤𝐤^{}|\sigma \pi \sqrt{2}/a`$, so that $`1/ef^J(𝐤𝐤^{})/f_0^J1`$.
In the case of a circular Fermi surface cross-section, as might be expected for electron doped HTSC, we could write $`𝐤=k_F(\mathrm{cos}\varphi _𝐤,\mathrm{sin}\varphi _𝐤)`$, etc. and Eq. (5) can be rewritten as
$$f^J(\varphi _𝐤,\varphi _𝐤^{})=f^J(\gamma )\mathrm{exp}[\gamma \mathrm{cos}(\varphi _𝐤\varphi _𝐤^{})],$$
(6)
where $`\gamma =(ak_F/\pi \sigma )^2`$ and $`f^J(\gamma )=f_0^J\mathrm{exp}(\gamma )`$. This form is identical to that of Graf et al., provided that $`f_0^J=\mathrm{exp}(\gamma )/[\tau _{}I_0(\gamma )]`$, where $`I_n(z)`$ is a Bessel function and $`1/\tau _{}`$ is an effective interlayer tunneling rate. The denominator contains the normalization factor $`I_0(\gamma )`$. Then just below $`T_c`$, $`\mathrm{\Delta }(T)0`$, and
$$J_{c\zeta }^J=C_0|f^J(\varphi _𝐤,\varphi _𝐤^{})\mathrm{\Delta }_{1\zeta }(\varphi _{𝐤_+})\mathrm{\Delta }_{2\zeta }(\varphi _𝐤_{}^{})_{\varphi _𝐤,\varphi _𝐤^{}}|,$$
(7)
where $`C_0=em^2/(4T_c)`$, $`m`$ is the in-plane effective mass, $`\zeta =s,d,e`$ indexes the OP’s, and $`\varphi _{𝐤_+}=\varphi _𝐤+\varphi _0/2`$, $`\varphi _𝐤_{}^{}=\varphi _𝐤^{}\varphi _0/2`$. These OP’s are $`\mathrm{\Delta }_{ns}=\mathrm{\Delta }_0`$, $`\mathrm{\Delta }_{nd}(\varphi )=\mathrm{\Delta }_0\mathrm{cos}(2\varphi )`$, and $`\mathrm{\Delta }_{ne}(\varphi )=|\mathrm{\Delta }_0\mathrm{cos}(2\varphi )|`$, respectively. We find
$`J_{cs}^J`$ $`=`$ $`C_0\mathrm{\Delta }_0^2/\tau _{},`$ (8)
$`J_{cd}^J`$ $`=`$ $`C_0\mathrm{\Delta }_0^2|\mathrm{cos}(2\varphi _0)|f_d(\gamma )/\tau _{},`$ (9)
and
$`J_{ce}^J`$ $`=`$ $`C_0\mathrm{\Delta }_0^2f_e(\gamma ,\varphi _0)/\tau _{},`$ (10)
where
$`f_d(\gamma )`$ $`=`$ $`{\displaystyle \frac{I_2(\gamma )}{2I_0(\gamma )}}`$ (11)
and
$`f_e(\gamma ,\varphi _0)`$ $`=`$ $`{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{4(2\delta _{n,0})\mathrm{cos}(4n\varphi _0)I_{4n}(\gamma )}{\pi ^2(4n^21)^2I_0(\gamma )}}.`$ (12)
In this simple model, $`J_{ce}^J/J_{cs}^J=f_e(\gamma ,\varphi _0)`$ at $`T_c`$, which is periodic in $`\varphi _0`$ with period $`\pi /2`$, satisfying $`f(\gamma ,\pm \pi /2\varphi _0)=f(\gamma ,\varphi _0)`$. In the incoherent limit, $`f_e(0,\varphi _0)=(2/\pi )^2`$, a constant. However, in the coherent limit $`f_e(\mathrm{},\varphi _0)=[|\mathrm{sin}(2\varphi _0)|+(\pi /22|\varphi _0|)\mathrm{cos}(2\varphi _0)]/\pi `$ in the domain $`|\varphi _0|\pi /2`$. $`f_e(\mathrm{},\varphi _0)`$ has maxima of $`1/2`$ at $`\varphi _0=n\pi /2`$ and minima of $`1/\pi `$ at $`\varphi _0=(2n+1)\pi /4`$ for integer $`n`$. The $`\varphi _0`$ and $`\gamma `$ dependencies of $`f_e(\gamma ,\varphi _0)`$ are shown in Figs. 2 and 3.
However, for all $`\gamma `$ values, $`J_{cd}^J|\mathrm{cos}(2\varphi _0)|`$, which vanishes at $`\varphi _0=45^{}`$, as shown in Fig. 4. In addition, $`J_{cd}^J(\varphi _0=0)`$ differs from the $`s`$-wave case by the additional factor $`f_d(\gamma )=I_2(\gamma )/[2I_0(\gamma )]`$. In the coherent limit, $`\gamma \mathrm{}`$, $`f_d(\gamma )1/2`$, but in the incoherent limit $`\gamma 1`$, $`f_d(\gamma )\gamma ^2/16`$. $`f_e(\gamma ,0)`$ also decreases from $`1/2`$ at $`\gamma \mathrm{}`$ with decreasing $`\gamma `$, but not nearly as dramatically, approaching the constant value $`(2/\pi )^2`$ as $`\gamma 0`$. The functions $`f_d(\gamma )`$ and $`f_e(\gamma ,0)`$ are compared in Fig. 3, where they are plotted as functions of $`1/\gamma `$ for clarity.
We now consider the full calculations with the Fermi surface shown in Fig. 1, and Gaussian tunneling, Eq. (5). We performed explicit calculations at $`T/T_c=0.5,0.9`$. In Figs. 4-6, we plotted our $`J_c^J(\varphi _0)/J_c^J(0)`$ results at $`T/T_c=0.5`$ for the $`d_{x^2y^2}`$-wave, extended-$`s`$-wave, and isotropic $`s`$-wave OP’s, respectively, for $`0^{}\varphi _045^{}`$. Results for $`T/T_c=0.9`$ are very similar. All results are periodic in $`\varphi _0`$ with period $`\pi /2`$ and symmetric about $`\varphi _0=\pm \pi /4`$. Note that this presentation is equivalent to $`J_c^J(\varphi _0)/J_c^S`$, since $`J_c^S`$ is the critical current density across the untwisted single crystal, which is itself a stack of essentially equivalent Josephson junctions. Hence, $`J_c^J(0)=J_c^S`$. In each case, we show the results for $`\sigma ^2=0,0.005,0.05,0.25,`$ and 0.50.
For each OP, coherent tunneling causes $`J_c^J(\varphi _0)/J_c^J(0)`$ to decrease sharply as $`\varphi _0`$ increases from $`0^{}`$. This is because the tight-binding Fermi surfaces are only identical for $`\varphi _0=0^{}`$, and the overlap of the Fermi surfaces changes dramatically from a continuous curve to a set of four points for $`\varphi _0>0^{}`$. In addition, there are local maxima in $`J_c^J(\varphi _0)/J_c^J(0)`$ at about $`\varphi _020^{}`$, which are absent for a circular Fermi surface cross-section. For this twist angle, the rotated Fermi surfaces begin to overlap at eight points. This increased overlap is responsible for the more gradual decrease of $`J_c^J(\varphi _0)`$ at larger $`\varphi _0`$ values. We also note that the coherent tunneling curves for the extended-$`s`$\- and $`d`$-wave OP’s are nearly equal for $`\varphi _030^{}`$. This is also true for the $`\sigma ^2=0.005`$ curves. However, the $`d`$-wave curves are distinctly different from the $`s`$-wave and extended-$`s`$-wave curves in the vicinity of $`\varphi _0=45^{}`$, where the $`d`$-wave curves all vanish by symmetry. In addition, the $`d`$-wave curves for $`\sigma ^2=0.25`$ and $`\sigma ^2=0.50`$ are essentially identical and quantitatively in agreement with the function $`|\mathrm{cos}(2\varphi _0)|`$. This is the same $`\varphi _0`$ dependence that we obtained analytically for all $`\gamma `$ in the case of a circular Fermi surface cross-section.
In comparing the cases of the isotropic and extended-$`s`$-wave OP’s, we see that the main differences appear for rather coherent tunneling, for which the extended-$`s`$-wave curves have a more dramatic $`\varphi _0`$-dependence. However, both OP’s are consistent with the data of Li et al. for $`\sigma ^20.25`$, and neither one is consistent with the data for $`\sigma ^20.05`$. By examining Fig. 1, we see that $`\sigma ^2=0.25`$ corresponds to a tunneling strength satisfying $`1/ef^J(𝐤𝐤^{})/f_0^J1`$ over 39% of the first BZ, which is strongly incoherent. Thus, these calculations imply that it is impossible to distinguish the isotropic-$`s`$-wave from the extended-$`s`$-wave OP’s in the $`c`$-axis twist junction experiments. We can only say that there must be an isotropic $`s`$-wave OP component for $`T<T_c`$, which is projected out from the OP anisotropy, if any, by averaging over a large portion of the BZ.
Finally, we have studied the reductions in the product $`I_cR_n`$ of the critical current times the resistance across a junction from the AB limit ($`I_cR_n(T)|_{AB}`$) for the case of an extended-$`s`$-wave OP as $`\sigma \mathrm{}`$. We first define
$$I_cR_n(T)=C(T/T_c)I_cR_n(T)|_{AB},$$
(13)
where the AB limit curve corresponds to the isotropic $`s`$-wave case with $`\sigma \mathrm{}`$ for any Fermi surface. A few limits can be investigated analytically. For a circular Fermi surface cross-section, we can analytically evaluate $`I_cR_n`$ for this model both at $`T_c`$ and as $`T0`$. We find $`C(1)=(2/\pi )^20.405`$, as above, and $`C(0)=(2/\pi )^2\mathrm{ln}40.562`$.
Similar results can be obtained for different Fermi surfaces. For the Fermi surface shown in Fig. 1, we find numerically that $`C(0.9)=0.416`$, $`C(0.5)=0.465`$, and $`C(0)=0.572`$. For the slightly different Fermi surface studied elsewhere, with $`t^{}/t=1.3`$ and $`\mu /t=0.6`$, we found $`C(0.9)=0.400`$, $`C(0.5)=0.450`$, and $`C(0)=0.578`$. Thus, the result of Yurgens et al. that $`I_cR_n(T=0)`$ 10 mV for the HgBr<sub>2</sub>-intercalated Bi2212 is in rather good agreement with that expected for an extended-$`s`$ OP.
We remark that there have been a number theories of the HTSC that relied upon coherent tunneling in the $`c`$-axis, with a $`𝐤`$-dependent matrix element, between each of the layers. This model was shown previously to give the worst agreement with the experiment of Li et al. of all of the coherent tunneling models studied, even for an $`s`$-wave OP. Here we showed that the experiment is easy to understand if the interlayer tunneling is strongly incoherent, provided that the OP integrates to a non-vanishing value on the two-dimensional Fermi “surface”.
Although there have not yet been many infrared reflectance measurements on Bi2212, the available experiments strongly suggest that the $`c`$-axis tunneling is not metallic. In that work, not only was no Drude edge above $`30`$cm<sup>-1</sup> seen in the $`c`$-axis conduction in the normal state, none was seen in the superconducting state as well. In addition, there is now strong evidence at all underdoped HTSC have incoherent $`c`$-axis normal state tunneling, with the sole exception of YBCO with $`\delta 0.15`$. This is also the case in the recent measurements on the electron doped material NCCO. Although above $`T_c`$, the non-metallic behavior is clearly seen, below $`T_c`$, metallic-like behavior reminiscent of a Drude edge for wavevectors in the range 10 to 200 cm<sup>-1</sup> has been observed, although none has yet been seen in Bi2212. This metallic-like behavior is most likely associated with the $`c`$-axis supercurrent. Thus, if Bi2212 were to behave the same as the other materials with an incoherent $`c`$-axis normal state conduction, one would expect it to show this same metallic-like behavior below $`T_c`$.
There have been two schools of thought on this issue. One is that the interlayer tunneling processes change dramatically from incoherent to coherent below $`T_c`$. Our analysis of the $`c`$-axis twist experiments of Li et al. strongly contradicts this idea, since the quasiparticle tunneling below $`T_c`$ must be strongly incoherent. The second school holds that there is no quasiparticle tunneling below $`T_c`$ due to an “orthogonality catastrophe”, but that the only tunneling process occurs by the simultaneous tunneling of pairs. However, the fact that HgB<sub>2</sub>-intercalated Bi2212 has nearly the same $`T_c`$ as does unintercalated Bi2212 argues strongly that this interlayer pair tunneling model for the occurrence of superconductivity is not correct.
Thus, if the superconductivity arises from intralayer pairing, then the tunneling process must be tunneling by individual quasiparticles. The $`c`$-axis twist experiments provide strong evidence that this tunneling is incoherent, meaning that the momenta of the quasiparticles parallel to the layers before and after the tunneling process are not correlated, or random. This is in addition to any possible changes in the kinetic energy that might also occur.
With incoherent normal state $`c`$-axis tunneling, it is extremely difficult to obtain a significant amount of $`c`$-axis critical current with an OP that averages to zero on the Fermi “surface”. Since there is a very strong consensus that the superconducting critical current in all HTSC is three-dimensional, incoherent tunneling in the $`c`$-axis direction necessarily implies that the OP must have an $`s`$-wave component, so that its Fermi “surface” average is non-vanishing at all $`TT_c`$. This is also true in the organic layered superconductor $`\kappa `$-(ET)<sub>2</sub>Cu\[N(CN)<sub>2</sub>\]Br, which has recently been shown to have extremely incoherent $`c`$-axis propagation, both above and below $`T_c`$. Combined with this supporting evidence that the $`c`$-axis tunneling in most HTSC is predominantly incoherent, the twist experiments of Li et al. provide very strong evidence of an $`s`$-wave superconducting OP in the HTSC.
Very recently, a simple model for the apparent violation of the conventional sum rule observed by Basov et al. was investigated theoretically by Kim and Carbotte (KC). Those workers assumed a $`d_{x^2y^2}`$-wave OP, and an incoherent interlayer tunneling matrix element squared of the form $`f^J(\varphi _𝐤,\varphi _𝐤^{})=|V_0|+|V_1|^2\mathrm{cos}(2\varphi _𝐤)\mathrm{cos}(2\varphi _𝐤^{})`$. By equating $`1/\lambda _c^2`$ derived from the conductivity and the superfluid density $`\rho _s`$, each evaluated to lowest order in $`f^J`$, they wrote,
$$\delta N_\zeta (T/T_c)=\frac{1}{2}+\frac{\underset{\omega }{}f^J(\varphi _𝐤,\varphi _𝐤^{})[1\omega _{\zeta 𝐤}\omega _{\zeta 𝐤^{}}]}{2_\omega f^J(\varphi _𝐤,\varphi _𝐤^{})\delta _{\zeta 𝐤}\delta _{\zeta 𝐤^{}}},$$
(14)
where $`\delta N_\zeta =(N_nN_{s\zeta })/\rho _{s\zeta }`$, $`\omega _{\zeta 𝐤}=\omega /\mathrm{\Omega }_{\zeta 𝐤}`$, $`\omega _{\zeta 𝐤^{}}=\omega /\mathrm{\Omega }_{\zeta 𝐤^{}}`$, $`\delta _{\zeta 𝐤}=\mathrm{\Delta }_\zeta (\varphi _𝐤)/\mathrm{\Omega }_{\zeta 𝐤}`$, $`\delta _{\zeta 𝐤^{}}=\mathrm{\Delta }_\zeta (\varphi _𝐤^{})/\mathrm{\Omega }_{\zeta 𝐤^{}}`$, $`\mathrm{\Omega }_{\zeta 𝐤}=[\omega ^2+\mathrm{\Delta }_\zeta ^2(\varphi _𝐤)]^{1/2}`$, and $`\mathrm{\Omega }_{\zeta 𝐤^{}}=[\omega ^2+\mathrm{\Delta }_\zeta ^2(\varphi _𝐤^{})]^{1/2}`$. It is easy to show that for coherent tunneling $`\delta N_\zeta (T/T_c)=1`$ for any OP, for all $`T/T_c1`$. It is also easy to show that $`\delta N_s(T/T_c)=1`$, regardless of the form of $`f^J(\varphi _𝐤,\varphi _𝐤^{})`$. For the $`d`$-wave case, KC showed that incoherent tunneling gives a conductivity sum rule violation that is strong and of the wrong sign. They gave a lower limit on the violation, based upon restrictions of the parameters $`V_0`$ and $`V_1`$. However, KC did not calculate the extended- $`s`$-wave case, which is actually very interesting to do.
We studied the one-parameter model of Graf et al., , which interpolates smoothly between coherent and incoherent tunneling. In this model, we can analytically perform the calculations at $`T_c`$ for the $`\zeta =d,e`$ cases. We find that $`\delta N_d(1)=[2+1/f_d(\gamma )]/4`$ and $`\delta N_e(1)=[2+1/f_e(\gamma ,0)]/4`$. In the coherent limit $`\gamma \mathrm{}`$, all three OP’s give no sum rule violation, as expected. However, in the incoherent limit $`\gamma 0`$, $`\delta N_d(1)4/\gamma ^2`$, which diverges strongly, and $`\delta N_e(1)(1+\pi ^2/8)/21.117`$. As $`T0`$ in the coherent limit $`\gamma \mathrm{}`$, we again have $`\delta N_\zeta (0)=1`$ for $`\zeta =s,d,e`$, as expected. As $`T0`$ in the incoherent limit $`\gamma 0`$, for both $`\zeta =d,e`$, we can evaluate the denominator in Eq. (14) exactly, and the numerator numerically. We find $`\delta N_e(0)1.087`$ and $`\delta N_d(0)\mathrm{}`$. We thus conclude that none of the three OP’s gives a $`\delta N<1`$, but the $`d`$-wave case is by far the worst, especially in the limit of strongly incoherent tunneling. Thus, it is likely to be much easier to construct a theory that can incorporate both incoherent interlayer tunneling and a $`\delta N<1`$ if the OP is an anisotropic $`s`$-wave one, and not $`d`$-wave. In particular, theories based upon two gaps, a non-superconducting pseudogap and a superconducting gap, appear likely to satisfy the experimental sum rule violation, provided that the OP describing the superconducting state is either $`s`$-wave or extended-$`s`$-wave.
In conclusion, we found that the $`c`$-axis twist experiments of Li et al. provide compelling evidence that the $`c`$-axis tunneling in Bi2212 is strongly incoherent. As a consequence, the experiment cannot distinguish between an isotropic $`s`$-wave and an extended-$`s`$-wave OP. However, the purported $`d_{x^2y^2}`$-wave OP can be ruled out in Bi2212. The other HTSC and organic layered superconductors which also have such incoherent $`c`$-axis tunneling also cannot have a $`d_{x^2y^2}`$-wave OP. Such incoherent $`c`$-axis tunneling is likely to be consistent with the available $`c`$-axis infrared reflectivity measurements, provided that the OP is not of the $`d_{x^2y^2}`$-wave form.
The authors thank G. Arnold, R. Kleiner, Qiang Li, and M. Mößle for useful discussions. This work was supported by USDOE-BES Contract No. W-31-109-ENG-38, by NATO Collaborative Research Grant No. 960102, and by the DFG through the Graduiertenkolleg “Physik nanostrukturierter Festkörper.”
|
warning/0005/cond-mat0005347.html
|
ar5iv
|
text
|
# Parity Effect in a mesoscopic superconducting ring
## Abstract
We study a mesoscopic superconducting ring threaded by a magnetic flux when the single particle level spacing is not negligible. It is shown that, for a superconducting ring with even parity, the behavior of persistent current is equivalent to what is expected in a bulk superconducting ring. On the other hand, we find that a ring with odd parity shows anomalous behavior such as fluxoid quantization at half-integral multiples of the flux quantum and paramagnetic response at low temperature. We also discuss how the parity effect in the persistent current disappears as the temperature is raised or as the size of the ring increases.
What happens to superconductivity when the sample is made very small? Anderson already addressed this fundamental question in 1959 and argued that as the size of a superconductor decreases and, accordingly, the average level spacing $`\delta `$ becomes larger than the BCS gap $`\mathrm{\Delta }`$, superconductivity is no longer possible. Recent experiments on ultrasmall “superconducting” nanoparticles have led to reconsider this old, but fundamental question. In a series of experiments the authors of studied transport through nanometer-scale Al grains and succeeded to get the discrete eigenspectrum of a single superconducting grain. The results were found to depend on the parity, i.e. on the electron number in the grain being even or odd. These experiments initiated several theoretical investigations. von Delft et al. used a model of uniform level spacing in a parity-projected mean field theory and found that the breakdown of superconductivity occurs at a value of $`\delta /\mathrm{\Delta }O(1)`$ which is parity-dependent. This parity effect has been shown to increase when including the effects of level statistics . Effects of quantum fluctuations , canonical description of BCS superconductivity , as well as transport theory for a nanoparticle coupled to superconducting leads have also been subjects of the study along this line.
Another interesting example for studying the size effect on superconductivity is a mesoscopic superconducting ring threaded by a magnetic flux $`\mathrm{\Phi }`$. It is well known that a conventional BCS superconducting ring exhibits fluxoid quantization at integer multiples of the flux quantum $`\mathrm{\Phi }_0=hc/2e`$ and a diamagnetic response at $`\mathrm{\Phi }=n\mathrm{\Phi }_0`$ with $`n`$ being an integer . In this Letter, we address the following question which is essentially the same as in a simply connected grain: What happens in a superconducting ring when the size of the ring becomes very small? For this purpose, we adopt the parity-projected mean field theory for an ideal mesoscopic ring. It is shown that the order parameter strongly depends on the parity when the size of the ring is small enough, as in the case of a grain. The most dramatic feature we show in our study is the behavior of the supercurrent (or persistent current) which strongly depends on the parity. For a ring with even parity, the behavior of the supercurrent is identical to that of a bulk superconducting ring. It exhibits fluxoid quantization at integer multiples of the flux quantum, and a diamagnetic response for small deviations of the flux from the integer multiples of the flux quantum. On the other hand, the characteristics are found to be very different for a small, odd parity ring. In a superconductor with odd parity, there is one unpaired quasiparticle. We find that at low temperature the existence of this quasiparticle drives the superconducting ring to a half-integral fluxoid quantization and a paramagnetic response at small values of the flux. We further show that the anomalous behavior in an odd-parity superconductor disappears at temperatures higher than the level spacing of single electron spectra, where one recovers the behavior of a conventional superconducting ring. Finally, this parity-dependent behavior of supercurrent is shown to disappear in the thermodynamic limit.
The ideal superconducting ring can be described by the Hamiltonian:
$$H=\underset{j\sigma }{}\epsilon _j^0c_{j\sigma }^{}c_{j\sigma }\lambda \delta \underset{i,j}{}c_i^{}c_{\overline{i}}^{}c_{\overline{j}}c_j.$$
(1)
The single particle energy is given by $`\epsilon _j^0=\frac{\mathrm{}^2}{2mR^2}(jf/2)^2`$, where $`R`$ is the radius of the ring, $`f=\mathrm{\Phi }/\mathrm{\Phi }_0`$ is the external flux divided by the flux quantum $`\mathrm{\Phi }_0=hc/2e`$, and $`j`$ is an integer which corresponds to an angular momentum quantum number. This is obtained by solving the Schrödinger equation in a 1D noninteracting ring. Note that $`\overline{j}=j`$. $`\lambda `$ is the dimensionless BCS coupling constant. $`\delta =\mathrm{}^2N/8mR^2`$ is the level spacing at the Fermi energy, where $`N`$ is the number of electrons in the ring. We don’t take into account the Zeeman splitting, namely $`h`$, because it is negligible unless the radius of the ring is very small. For $`\mathrm{\Phi }\mathrm{\Phi }_0`$ with a uniform magnetic field, the ratio of the level spacing $`\delta `$ to the Zeeman splitting is proportional to $`R`$ which is estimated as $`\delta /h10^{20}r_sR/m^2`$, where $`r_s`$ is the average distance between electrons. For example, in a typical superconductor such as Al with $`R1\mu m`$, $`\delta /h10^4`$.
A simple way of describing a mesoscopic superconductor with fixed number parity $`P`$ (denoted by $`e`$ for even, and $`o`$ for odd parity) is to adopt the parity-projected grand canonical partition function
$$Z_P(\mu )=\text{Tr}\frac{1}{2}[1\pm (1)^N]e^{\beta (H\mu N)}.$$
(2)
We evaluate $`Z_P`$ using the BCS-type mean field approximations, which consists in neglecting quadratic terms of the fluctuations :
$`c_i^{}c_{\overline{i}}^{}c_{\overline{j}}c_j`$
$``$ $`c_i^{}c_{\overline{i}}^{}c_{\overline{j}}c_j+c_i^{}c_{\overline{i}}^{}c_{\overline{j}}c_jc_i^{}c_{\overline{i}}^{}c_{\overline{j}}c_j`$
$`+`$ $`\delta _{ij}\left(c_i^{}c_ic_{\overline{i}}^{}c_{\overline{i}}+c_i^{}c_ic_{\overline{i}}^{}c_{\overline{i}}c_i^{}c_ic_{\overline{i}}^{}c_{\overline{i}}\right).`$
The ensemble average $`\mathrm{}`$ should be evaluated in a given parity $`P=e`$ or $`P=o`$. The first three terms on the r.h.s of Eq.(Parity Effect in a mesoscopic superconducting ring) correspond to the mean-field approximation for the superconducting pairing. The last three terms are usually not considered since they give no contribution in the thermodynamic limit. However, those terms cannot be ignored in mesoscopic systems, though they were neglected in the previous mean-field description for ultrasmall superconducting grains . Note that the validity of our mean-field treatment is limited to the $`\delta <\mathrm{\Delta }`$ limit where the usual BCS approximation can be applied. As a result we get the following expression for the mean-field Hamiltonian
$$H=C_P+\underset{j\sigma }{}\stackrel{~}{E}_j\gamma _{j\sigma }^{}\gamma _{j\sigma }+\mu N,$$
(4)
where $`\gamma _{j\sigma }`$ ($`\gamma _{j\sigma }^{}`$) destroys (creates) a quasiparticle; $`\gamma _{j\sigma }=u_jc_{j\sigma }\sigma v_jc_{\overline{j}\sigma }^{}`$, and the constant $`C_P`$ and the quasiparticle energy $`\stackrel{~}{E}_j`$ are given by
$`C_P`$ $`=`$ $`{\displaystyle \underset{j}{}}\left[{\displaystyle \frac{1}{2}}(\epsilon _j+\epsilon _{\overline{j}})E_j\right]+\mathrm{\Delta }_P^2/\lambda \delta `$
$`+`$ $`\lambda \delta {\displaystyle \underset{j}{}}\left[u_j^2f_{\overline{j}\overline{\sigma }}+v_j^2(1f_{j\sigma })\right]\left[u_j^2f_{j\sigma }+v_j^2(1f_{\overline{j}\overline{\sigma }})\right],`$
and
$$\stackrel{~}{E}_j=\frac{1}{2}(\epsilon _j\epsilon _{\overline{j}})+E_j,$$
(6)
respectively. Here $`\mathrm{\Delta }_P=\lambda \delta _jc_{\overline{j}}c_j`$ is the parity-dependent order parameter which has to be calculated self-consistently. $`\epsilon _j`$ and $`E_j`$ are defined as $`\epsilon _j=\epsilon _j^0\mu \lambda \delta [u_j^2f_{\overline{j}\overline{\sigma }}+v_j^2(1f_{j\sigma })]`$ and $`E_j=\sqrt{\mathrm{\Delta }_P^2+(\epsilon _j+\epsilon _{\overline{j}})^2/4}`$, respectively. For simplicity we neglect the last term of $`\epsilon _j`$. It does not change the qualitative feature of our results since its role is only to increases somewhat the effective level spacing near the Fermi level for large $`\delta `$. $`f_{j\sigma }=\gamma _{j\sigma }^{}\gamma _{j\sigma }`$ is the average occupation number of a state $`j`$ with spin $`\sigma `$. It is parity-dependent and given by
$$f_{j\sigma }=\frac{f_+(\stackrel{~}{E}_j)Z_+f_{}(\stackrel{~}{E}_j)Z_{}}{Z_+\pm Z_{}},$$
(7)
for $`P=e`$ (upper sign) and $`P=o`$ (lower sign), where $`f_\pm (E)=1/(e^{\beta E}\pm 1)`$ and $`Z_\pm =_{j\sigma }(1\pm e^{\beta \stackrel{~}{E}_j})`$. $`u_j,v_j`$ are BCS coherence factors:
$$u_j^2=1v_j^2=\frac{1}{2}\left(1+\frac{\epsilon _j+\epsilon _{\overline{j}}}{2E_j}\right).$$
(8)
The parity-dependent chemical potential $`\mu _P`$ is determined by the relation
$$N=\frac{1}{\beta }\frac{}{\mu }\mathrm{log}Z_P(\mu )|_{\mu =\mu _P},$$
(9)
that holds provided that $`\mu _P`$ lies halfway between the last filled and first empty level if $`P=e`$, and on the singly occupied level if $`P=o`$ .
The mean-field self-consistency condition $`\mathrm{\Delta }_P=\lambda \delta _jc_{\overline{j}}c_j`$ leads to the relation
$$1=\lambda \delta \underset{j=j_c}{\overset{j_c}{}}\frac{1}{2E_j}(1f_{\overline{j}}f_j),$$
(10)
where $`j_c`$ is the cutoff value of $`j`$ in the summation. At $`T=0`$ with $`\mathrm{\Phi }=0`$ the occupation of quasiparticles reduces to $`f_{j\sigma }=\frac{1}{4}`$ if the level lies on the chemical potential, namely $`j=\pm j_F`$, for $`P=o`$, and zero otherwise. The factor $`1/4`$ is due to the orbital degeneracy of the ring, $`\epsilon _j^0=\epsilon _{\overline{j}}^0`$, as well as the spin degeneracy. For nonzero external flux, the orbital degeneracy is lifted and $`f_{j_F\sigma }=1/2`$ (but $`f_{\overline{j}_F\sigma }=0`$) for odd parity. This nonzero value of quasiparticle occupation gives rise to the so called “blocking effect”. That is, the odd parity superconductor has one unpaired electron, which prevents pair scattering of other pairs into/out of the singly occupied state and reduces the order parameter as compared to that of an even parity superconductor.
In the framework of the parity-projected grand canonical description, three different cases appear in solving the “gap” equation: (i) Even parity with fully occupied highest level ($`N=4j_F+2`$), (ii) even parity with partially occupied highest level ($`N=4j_F`$), and (iii) odd parity ($`N=4j_F\pm 1`$). It is important to note that the self-consistent equation (10) does not depend on the flux if $`N`$ is kept unchanged under variation of the flux. The condition of constant $`N`$ leads to the flux-dependent chemical potential
$$\mu _P(f)=\mu _P(0)+\frac{\mathrm{}^2}{8mR^2}f^2.$$
(11)
$`E_j`$ is independent of $`f`$ with this condition, and accordingly Eq.(10) is also flux-independent at low temperature, $`k_BT<\delta ,\mathrm{\Delta }_P`$.
Fig.1 shows the pairing parameter as a function of $`\delta `$ keeping the electron density constant. In solving the equation we chose $`\lambda =0.2`$, close to that of Al , and $`j_c=2j_F`$. The pairing parameter which is obtained by solving Eq.(10) depends strongly on the parity as in the grain superconductor. In the odd-parity superconductor the order parameter is suppressed compared to the one for even parity. However, the behavior of the pairing parameter as a function of the level spacing is somewhat different from the grain superconductor with equal level spacing or with randomly distributed levels . It is due to the existence of orbital degeneracy in the ideal ring geometry. If the highest occupied level is fully filled (case (i)), the chemical potential lies halfway between the last filled and the first empty level. In this case there exists a critical level spacing $`\delta _c/\mathrm{\Delta }_0O(1)`$ where $`\mathrm{\Delta }_e`$ goes to zero as in the mean-field solution for the grain superconductivity. If the highest occupied level is partially filled (both for even and odd parity) the chemical potential lies on that level. Pair scattering of electrons between these orbitally degenerate levels makes it impossible to have a solution with $`\mathrm{\Delta }_P=0`$. On the other hand, the presence of any weak disorder will lift the orbital degeneracy and give a solution with $`\mathrm{\Delta }_P=0`$. Note that, in a very small grain with $`\delta \mathrm{\Delta }_0`$, quantum fluctuations are important and the mean-field treatment becomes invalid .
Next we discuss the parity-dependent behavior of the persistent current in the “superconducting” state. Because we deal with an isolated ring, the current should be calculated in the canonical description, while we use the parity-projected grand canonical partition function. However, the number fluctuation in the grand canonical treatment is very small at low temperature and the free energy with a fixed number $`N`$, namely $`F_N`$, can be written as
$$F_N\frac{1}{\beta }\mathrm{log}Z_P+\mu _PN.$$
(12)
Note that this relation is exact at $`T=0`$. The persistent current is given by
$$I=c\frac{F_N}{\mathrm{\Phi }},$$
(13)
which gives the following expression by using the mean field Hamiltonian (4)
$$I=I_{dia}+I_{para},$$
(14)
where $`I_{dia}`$ and $`I_{para}`$ are the diamagnetic and the paramagnetic contribution to the current, respectively,
$`I_{dia}`$ $`=`$ $`c{\displaystyle \frac{\mu _P}{\mathrm{\Phi }}}N,`$ (15)
$`I_{para}`$ $`=`$ $`c{\displaystyle \underset{j\sigma }{}}{\displaystyle \frac{\stackrel{~}{E}_j}{\mathrm{\Phi }}}f_{j\sigma }.`$ (16)
Note that at low temperature ($`k_BT<\delta `$, $`\mathrm{\Delta }_P`$) the constant term $`C_P`$ in the Hamiltonian (4) does not contribute to the current since it is independent of the flux.
For even parity, the paramagnetic contribution to the persistent current is absent at $`k_BT\mathrm{\Delta }_e`$ since $`f_{j\sigma }=0`$ for $`P=e`$. Thus, the persistent current for $`P=e`$, $`I^e`$, is equal to $`I_{dia}`$ and obtained from Eq.(11) and (15):
$$I^e=2I_0f,$$
(17)
where $`I_0=ev_F/(2\pi R)`$ with $`v_F=\frac{\mathrm{}}{m}\frac{j_F}{R}`$ being the Fermi velocity and $`f=\mathrm{\Phi }/\mathrm{\Phi }_0`$. (See Fig.2(a).) The behavior of $`I^e`$ is essentially equivalent to what is expected in the bulk superconducting ring.
On the other hand, the paramagnetic contribution is important for $`P=o`$ because of an unpaired quasiparticle. $`I_{dia}`$ is identical for $`P=o`$ and $`e`$. At low temperature ($`k_BT<\delta `$) $`I_{para}`$ can be written as follows
$$I_{para}2I_0\left(f_{j_F\sigma }f_{\overline{j}_F\sigma }\right),$$
(18)
where
$$f_{\pm j_F\sigma }\frac{1}{2}\frac{e^{\pm f/t}}{\left(e^{f/t}+e^{f/t}\right)}.$$
(19)
$`t`$ is dimensionless temperature $`t=k_BT/\delta `$. At zero temperature $`I_{para}`$ reduces to $`\text{sgn}[f]I_0`$ and the total current for $`P=o`$, $`I^o`$, is given by
$$I^o=I_0(\text{sgn}[f]2f).$$
(20)
As shown in the solid line of Fig.2(b), the current vanishes at half-flux quantum, $`f=\pm 1/2`$, while it has maximum value of $`|I^o|`$ at $`f=0`$. This implies that the possible values of fluxoid in a ring with odd parity are half-integers, $`\pm \mathrm{\Phi }_0/2`$, $`\pm 3\mathrm{\Phi }_0/2`$, etc, in contrast to the conventional fluxoid quantization at integer multiples of $`\mathrm{\Phi }_0`$. Note that, the fluxoid $`\mathrm{\Phi }^{}`$ is defined in a conventional way
$$\mathrm{\Phi }^{}=\mathrm{\Phi }+\frac{(2m)c}{2e}v_s𝑑l,$$
(21)
where $`v_s`$ is the supercurrent velocity.
The parity-dependent behavior of the persistent current found above should be distinguished from that of a normal ring of noninteracting electrons with spin $`1/2`$ considered by Loss and Goldbart. In a normal ring with spin $`1/2`$, the persistent current depends on the number of particles with modulo 4. Periodicity and oscillation amplitude depend on the number, and the current response is diamagnetic only for $`N=4j_F+2`$ (the case of fully occupied top level). On the other hand, a mesoscopic superconducting ring shows only the parity dependence that originates from the electron paring. The periodicity and amplitude of oscillation remain unchanged as $`\mathrm{\Phi }_0(=hc/2e)`$ and $`I_0`$, respectively.
The effect of temperature on the persistent current is shown in Fig.2 (b). $`I_{para}`$ is reduced as the temperature is raised. It is clear that the effect of the unpaired quasiparticle disappears with increasing $`T`$; the persistent current will show a behavior like in Fig.2(a), at temperature higher than $`t1`$. Actually there is a crossover temperature $`t^{}=0.5`$ where the current response at small flux changes from paramagnetic to diamagnetic. (That is $`\frac{dI^o}{d\mathrm{\Phi }}|_{\mathrm{\Phi }=0}(t=t^{})=0`$.)
Let us, finally, discuss the condition under which the parity effect can be observed in the persistent current. As noted above, the condition for the existence of the paramagnetic contribution to the persistent current is $`T<T^{}`$ with $`k_BT^{}/\delta =0.5`$. Thus, the crossover temperature is given by
$$k_BT^{}=2.5\text{(eV)}\frac{r_s}{R},$$
(22)
where $`r_s`$ is the average distance between electrons. For Al $`r_s=1.10`$Å$`=2.07a_0`$, with $`a_0`$ being the Bohr radius. For a ring with $`R5\times 10^4r_s5\mu m`$, $`T^{}1\text{K}`$ is comparable to the transition temperature of bulk Al. If the ring is not strictly 1D, the level spacing is reduced and, accordingly, the crossover temperature $`T^{}`$ is lowered as compared to the ideal 1D ring with the same radius. Furthermore, $`T^{}0`$ if $`R\mathrm{}`$. In other words, the parity effect disappears in the thermodynamic limit, and our treatment coincides with the conventional BCS description.
In conclusion, we have investigated the parity-dependent properties of a mesoscopic superconducting ring threaded by a magnetic flux. The properties in the persistent current of an even-parity ring are similar to those of a bulk superconductor. On the other hand, a ring with odd parity shows unconventional behavior: the fluxoid quantization at half-integer multiples of the flux quantum. Moreover, there exists a paramagnetic response at temperatures below a characteristic temperature of the order of the level spacing. We have also shown that this parity effect of the persistent current disappears as the temperature is raised or as the size of the ring increases.
The author wishes to acknowledge valuable discussions with A. Bill. This work has been supported by the Visitors Program of the MPI-PKS, and also by the BK21 project of the Korean Ministry of Education.
|
warning/0005/cond-mat0005404.html
|
ar5iv
|
text
|
# Side-gate modulation of critical current in mesoscopic Josephson junction
## 1 Introduction
Devices with semiconducting materials as barriers in Josephson weak links offer the possibility to modulate the junction properties in order to investigate both fundamental transport mechanisms as well as for use in applications of superconducting devices. Several methods for tunable Josephson junctions have been studied both theoretically and experimentally. Tunable effects have been observed in e.g. Josephson-field-effect-transistors , non-equilibrium junctions , optically modulated weak links , and recently in side-gate modulated junctions .
Following Ref. we consider superconductor-2DEG-superconductor structures where the indirect ballistic transport between the non-opposite superconducting contacts can be controlled by a voltage applied to a side-gate (see Fig. 2). The geometry has many similarities with the T-stub wave guide geometry but here the two interfaces to the superconductors are non-opposite. The aim is to demonstrate the principle behind side-gate modulation of the critical current and also to investigate the local density-of-states in the 2DEG in relation to the discussion of “non-local” modes .
In this work we focus on the normal state conductance and the Josephson current in the case where electron and hole propagation in the normal region is phase-coherent. For convenience we consider the zero temperature limit. The normal state conductance is then given by the Landauer formula which at $`T=0\mathrm{K}`$ has the particularly simple form
$$G_\mathrm{N}=\frac{2e^2}{h}\mathrm{Tr}tt^{}=\frac{2e^2}{h}\underset{n=1}{\overset{N}{}}T_n,$$
(1)
where $`2e^2/h`$ is the quantum unit of conductance and $`T_n`$ ($`n=1,2,\mathrm{}N`$) are the transmission eigenvalues of $`tt^{}`$ which is an $`N\times N`$ matrix, $`N`$ being the number of propagating modes at the Fermi level. Beenakker and van Houten considered the Josephson current through a superconducting quantum point contact and found that the critical current was quantized in units of $`e\mathrm{\Delta }/\mathrm{}`$, $`\mathrm{\Delta }`$ being the energy gap of the superconductor. Subsequently it was shown by Beenakker that just like the normal state conductance, also the Josephson current between two superconducting leads can also be expressed in terms of $`T_n`$ of the normal region which couples the two superconductors. At $`T=0\mathrm{K}`$ the Josephson current can be written as
$`I_\mathrm{J}(\varphi )`$ $`=`$ $`{\displaystyle \frac{e\mathrm{\Delta }}{2\mathrm{}}}\mathrm{sin}\varphi \mathrm{Tr}tt^{}\left[\widehat{1}tt^{}\mathrm{sin}^2(\varphi /2)\right]^{1/2}`$ (2)
$`=`$ $`{\displaystyle \frac{e\mathrm{\Delta }}{2\mathrm{}}}\mathrm{sin}\varphi {\displaystyle \underset{n=1}{\overset{N}{}}}T_n\left[1T_n\mathrm{sin}^2(\varphi /2)\right]^{1/2},`$
where $`\varphi `$ is the phase difference between the two superconductors. Eq. (2) is valid in the short-junction limit where the dimension of the normal region is short compared to the superconducting coherence length $`\xi `$ which in the ballistic regime is $`\xi =\mathrm{}v_\mathrm{F}/\pi \mathrm{\Delta }`$, $`v_\mathrm{F}`$ being the Fermi velocity.
We note that in Eqs. (1) and (2) all transverse degrees-of-freedom are incorporated in the transmission amplitude matrix $`t`$ and that the angle-dependence of the Andreev scattering discussed in Ref. follows directly from the angle (mode) dependence of $`t`$. Scattering due to non-matching Fermi velocities or Fermi momenta can be included in several ways: i) by calculating the transmission from the composite scattering matrix $`S=S_{\mathrm{I}_1}S_{2\mathrm{D}\mathrm{E}\mathrm{G}}S_{\mathrm{I}_2}`$ where the two interfaces ($`\mathrm{I}_1`$ and $`\mathrm{I}_2`$) are described by the scattering matrices $`S_{\mathrm{I}_1}`$ and $`S_{\mathrm{I}_2}`$ as in Ref. , ii) by adding an effective potential to the 2DEG at the interfaces, or iii) by explicitly taking the different Fermi velocities and Fermi momenta into account. In this work we will however for simplicity assume matching Fermi properties of the superconductor and the 2DEG.
To illustrate how the transport properties depend on the transmission let us consider the case of single-mode leads. Fig. 1 shows how the normal state conductance $`G_\mathrm{N}`$, the critical current $`I_\mathrm{c}=\mathrm{max}I_\mathrm{J}(\varphi )`$, and the product $`I_\mathrm{c}R_\mathrm{N}=I_\mathrm{c}/G_\mathrm{N}`$ depend on the transmission probability $`𝒯=|t|^2`$. Compared to the off-resonance regime with $`𝒯0`$ (the Ambegaokar–Baratoff regime ) the $`I_\mathrm{c}R_\mathrm{N}`$ product is enhanced by a factor-of-two when the 2DEG region is tuned to resonance ($`𝒯1`$). This factor-of-two enhancement can be considered as a signature of Andreev mediated transport at resonance whereas the transport is tunneling-like at off-resonance.
Equations (1) and (2) form the basis for our calculations of the transport properties and since they only depend on the normal state transmission properties of the 2DEG region we can employ standard methods for quantum transport in semiconductor structures. The transmission can conveniently be calculated from the retarded Green function $`𝒢_\epsilon ^r(𝐫_1,𝐫_2)`$ and also the local density-of-states follows from $`𝒢_\epsilon ^r(𝐫_1,𝐫_2)`$. We calculate $`𝒢_\epsilon ^r(𝐫_1,𝐫_2)`$ by applying a finite differences method to the Hamiltonian of the 2DEG region.
The paper is organized as follows: Section II introduces the finite differences method and formulates Eqs. (1) and (2) in terms of the retarded Green function, section III presents our geometry and model, section IV contains our results, and in section V discussions and conclusions are given.
## 2 Finite differences method
We discretize the continuous Hamiltonian by writing the Laplacian as finite differences . This gives rise to a tight-binding-like representation
$`\left\{H_c\right\}_{ij}`$ $`=`$ $`z\times \gamma +U(𝐫_i)i=j,`$
$`=`$ $`\gamma i,j\mathrm{NN},`$
$`=`$ $`0\mathrm{otherwise},`$
where $`z=4`$ is the number of nearest neighbors (NN) and $`\gamma =\mathrm{}^2/2ma^2`$ corresponds to a hopping matrix element, $`a`$ being the lattice spacing. Comparing the energy dispersion to the parabolic dispersion of the continuous problem shows that the finite differences method is a very accurate description for energies $`\epsilon <\gamma `$ . Making the grid finer ($`a`$ smaller) improves the accuracy and/or allows for a treatment of higher energies.
With the above definition of the Hamiltonian we label the lattice points by a number from $`1`$ to $`M`$, $`M`$ being the number of lattice points. The retarded Green function can then be written as an $`M\times M`$ matrix
$$𝒢^\mathrm{r}(\epsilon )=\left[\epsilon \widehat{1}H_c\mathrm{\Sigma }^\mathrm{r}(\epsilon )\right]^1,$$
(4)
where the element $`\left\{𝒢^\mathrm{r}(\epsilon )\right\}_{ij}`$ corresponds to $`𝒢_\epsilon ^r(𝐫_i,𝐫_j)`$. Here, $`H_c`$ is the tight-binding Hamiltonian of the conductor (describing the conductor as a closed system) and $`\mathrm{\Sigma }^\mathrm{r}=\mathrm{\Sigma }_1^\mathrm{r}+\mathrm{\Sigma }_2^\mathrm{r}`$ is the self-energy describing the coupling to leads $`1`$ and $`2`$. For the self-energies the only non-zero elements are those where the leads couple to the 2DEG (scattering region). For large $`M`$ it can be useful to employ a recursive method for calculating $`𝒢^\mathrm{r}(\epsilon )`$ but even for $`M1000`$ the direct matrix inversion in Eq. (4) can be done without too much numerical effort.
The transmission amplitude matrix $`t`$ can now be calculated from the Fisher–Lee relation which gives
$$t(\epsilon )=\left[\mathrm{\Gamma }_1(\epsilon )\right]^{1/2}𝒢^\mathrm{r}(\epsilon )\left[\mathrm{\Gamma }_2(\epsilon )\right]^{1/2},$$
(5)
where $`\mathrm{\Gamma }_j(\epsilon )=i\left(\mathrm{\Sigma }_j^\mathrm{r}(\epsilon )[\mathrm{\Sigma }_j^\mathrm{r}(\epsilon )]^{}\right)`$. Substituting into Eqs. (1) and (2) gives
$$G_\mathrm{N}=\frac{2e^2}{h}\mathrm{Tr}\mathrm{\Gamma }_1(\epsilon )𝒢^\mathrm{r}(\epsilon )\mathrm{\Gamma }_2(\epsilon )[𝒢^\mathrm{r}(\epsilon )]^{},$$
(6)
and
$`I_\mathrm{J}(\varphi )`$ $`=`$ $`{\displaystyle \frac{e\mathrm{\Delta }}{2\mathrm{}}}\mathrm{sin}\varphi \mathrm{Tr}\mathrm{\Gamma }_1(\epsilon )𝒢^\mathrm{r}(\epsilon )\mathrm{\Gamma }_2(\epsilon )[𝒢^\mathrm{r}(\epsilon )]^{}`$
$`\times \left[\widehat{1}\mathrm{\Gamma }_1(\epsilon )𝒢^\mathrm{r}(\epsilon )\mathrm{\Gamma }_2(\epsilon )[𝒢^\mathrm{r}(\epsilon )]^{}\mathrm{sin}^2(\varphi /2)\right]^{1/2},`$
where we have used the cyclic invariance of the trace. Once the retarded Green has been obtained (by a single matrix inversion) these relations directly provide the essential transport properties. The local density-of-states (in the normal state) can also be calculated directly from the retarded Green function (see e.g. )
$$\rho (𝐫_j,\epsilon )=\frac{1}{\pi }\mathrm{Im}\left\{𝒢^\mathrm{r}(\epsilon )\right\}_{jj},$$
(8)
and this function is useful in obtaining insight into the spatial variations of the states and revealing the nature of the conduction through the sample.
## 3 Geometry and model
We consider the geometry shown in Fig. 2. The 2DEG is confined also in the direction parallel to the side-gate so that the system acts as a cavity or quantum dot coupled to two superconductors and an additional side-gate. We note that our geometry is slightly different from that of Ref. where there is no confinement of the 2DEG in the direction parallel to the side-gate. This means that here the side-gate is used in tuning the cavity to resonance whereas in Ref. it rather acts as a ’classical’ mirror which can be used in focusing the incident wave from one lead onto the other lead.
We treat the transmission problem fully quantum mechanically by means of the presented finite differences method and the lattice version of the sample shown in Fig. 2. In this case $`a=W/(M_\mathrm{L}+1)`$ with $`M_\mathrm{L}=4`$ sites in the transverse direction of the leads of width $`W`$. In the continuous description of the leads $`N=\mathrm{Int}(k_\mathrm{F}W/\pi )`$ gives the number of propagating modes at the Fermi level, $`\mathrm{Int}(x)`$ being the integer part of $`x`$. The dimensions are $`W^{}=13\sqrt{2}a3.7\times W`$ and $`W^{\prime \prime }=(13+1/2)\sqrt{2}a3.8\times W`$. We choose $`W`$ and the Fermi level such that $`N=1`$. The lattice model then gives a reasonable description of the continuous problem – the threshold energy $`E_1`$ of the first mode deviates only by $`3\%`$ from the continuous result.
The side-probe is assumed to act as a gate and only affect the potential $`U`$ of the 2DEG through an electro-static coupling. Here, we note that fluctuations in the gate potential may lead to dephasing of the electron and hole propagation in the 2DEG (see e.g. Ref. ) and the same can also be the case if the probe acts as a voltage probe (see e.g. Ref. ). Here, we use the model in Ref. \[Eq. (19)\] to account for the potential modification in the 2DEG due to the side-gate. The distance $`\zeta `$ from the side-gate to the 2DEG edge position is then given by
$$\zeta =\frac{ϵV_\mathrm{g}}{4\pi ^2n_0|e|}=\frac{ϵV_\mathrm{g}}{2\pi k_\mathrm{F}^2|e|},$$
(9)
where $`V_\mathrm{g}`$ is the side-gate voltage, $`ϵ=ϵ_rϵ_0`$ is the dielectric constant of the semiconductor ($`ϵ_r14`$ for GaAs), and $`n_0=k_\mathrm{F}^2/2\pi `$ the 2DEG density. With the lattice shown in Fig. 2 the distance $`\zeta `$ can be changed in steps of $`\delta \zeta =a/\sqrt{2}`$ corresponding to $`\delta V_\mathrm{g}=\sqrt{2}\pi k_\mathrm{F}^2a|e|/ϵ`$.
## 4 Results
In Fig. 3 we show the transmission as a function of the energy for different values of the gate voltage. Since the interfaces between the leads and the 2DEG are fully transparent the transmission shows a broad resonance behavior. By changing the gate voltage the cavity is ’squeezed’ and as expected the resonance shifts towards higher energies. For a given Fermi level the side-gate can thus also be used in tuning the transmission to resonance.
In Fig. 4 we show the normal state conductance $`G_\mathrm{N}`$, the critical current $`I_\mathrm{c}`$, and the $`I_\mathrm{c}R_\mathrm{N}`$ product as a function of gate voltage for an energy $`\epsilon =2.5\times E_1`$. By increasing the gate voltage the cavity is tuned to resonance at $`V_\mathrm{g}6\times \delta V_\mathrm{g}`$ where the normal state conductance equals the quantum unit of conductance $`2e^2/h`$ and the critical current equals the quantum unit of critical current $`e\mathrm{\Delta }/\mathrm{}`$. The $`I_\mathrm{c}R_\mathrm{N}`$ product changes by a factor-of-two from the Ambegaokar–Baratoff value $`\mathrm{\Delta }/8\pi e`$ to its quantum unit $`\mathrm{\Delta }/4\pi e`$ at resonance.
In the inset of Fig. 4 we show the local density of states $`\rho (𝐫,\epsilon )`$ (in the normal state) at the energy $`\epsilon =2.5\times E_1`$ for a gate voltage corresponding to the resonance condition. The transparent interfaces give rise to a local density of states forming a relatively smooth and connected ’path’ between the two leads. The plot also clearly shows the wave nature of the electron and hole propagation. Thus a more simple semi-classical trajectory model with point-like particles and a side-gate acting as a classical mirror would be inadequate for the present situation. This is often the case and as studied in e.g. Ref. complicated multiple scattering processes can in a full quantum mechanical treatment give rise to effects not found in a semi-classical study.
## 5 Discussion and conclusion
Hybrid semiconductor-superconductor structures offer interesting possibilities for investigating fundamental transport phenomena as well as for potential applications. We have studied the possibility of side-gate modulating the critical current in a mesoscopic superconductor-2DEG-superconductor Josephson junction. Side-gate modulation offers a new alternative to devices based on field-effects, non-equilibrium effects, and optical effects. In the case of side-gate modulation the side-gate is used in tuning the transmission of the 2DEG to resonance.
Our calculations are based on a numerical treatment of the transmission properties of the 2DEG region. Using the Landauer formula and a similar formula for the Josephson current we have then calculated the essential transport properties: the normal state conductance $`G_\mathrm{N}`$, the critical current $`I_\mathrm{c}`$, and the $`I_\mathrm{c}R_\mathrm{N}`$ product. The conductance in the superconducting state can unfortunately not be expressed directly in terms of the transmission eigenvalues and is complicated due to the presence of multiple Andreev reflections (see however Ref. for a possible solution). The considered geometry is comparable to the situation in recent experiments even though some simplifications have been made. A quantitative comparison would require i) that the different Fermi properties of the superconductor and the 2DEG are taken into account , ii) more lattice points ($`M_\mathrm{L}N`$) in order to account correctly for the case of multi-mode leads ($`N16`$ in Ref. ), and iii) that the width of the 2DEG should be much larger than the separation $`W^{}`$ of the two leads. However, our model studies demonstrate how the side-gate modulation can be used in controlling both the normal state conductance and the critical current.
Our results are in qualitative agreement with the recent experimental findings and confirm the possibility of modulating the transport properties by a side-gate. Studies of the local density of states confirm the existence of “non-local” modes and indicate that more simple trajectory models can not fully account for the detailed electron and hole propagation – rather a full quantum mechanical treatment is necessary.
Acknowledgements — We would like to thank M. Brandbyge, A.-P. Jauho, K. Flensberg, and H. Takayanagi for useful discussions.
|
warning/0005/quant-ph0005116.html
|
ar5iv
|
text
|
# Universal Quantum Computation with the Exchange Interaction
Experimental implementations of quantum computer architectures are now being investigated in many different physical settings. The full set of requirements that must be met to make quantum computing a reality in the laboratory is daunting, involving capabilities well beyond the present state of the art. In this report we develop a significant simplification of these requirements that can be applied in many recent solid-state approaches, using quantum dots , and using donor-atom nuclear spins or electron spins . In these approaches, the basic two-qubit quantum gate is generated by a tunable Heisenberg interaction (the Hamiltonian is $`H_{ij}=J(t)\stackrel{}{S}_i\stackrel{}{S}_j`$ between spins $`i`$ and $`j`$), while the one-qubit gates require the control of a local Zeeman field. Compared to the Heisenberg operation, the one-qubit operations are significantly slower and require substantially greater materials and device complexity, which may also contribute to increasing the decoherence rate. Here we introduce an explicit scheme in which the Heisenberg interaction alone suffices to exactly implement any quantum computer circuit, at a price of a factor of three in additional qubits and about a factor of ten in additional two-qubit operations. Even at this cost, the ability to eliminate the complexity of one-qubit operations should accelerate progress towards these solid-state implementations of quantum computation.
The Heisenberg interaction has many attractive features that have led to its being chosen as the fundamental two-qubit interaction in a large number of recent proposals: Its functional form is very accurate — deviations from the isotropic form of the interaction, arising only from relativistic corrections, can be very small in suitably chosen systems. It is a strong interaction, so that it should permit very fast gate operation, well into the GHz range for several of the proposals. At the same time, it is very short ranged, arising from the spatial overlap of electronic wavefunctions, so that it should be possible to have an on-off ratio of many orders of magnitude. Unfortunately, the Heisenberg interaction by itself is not a universal gate, in the sense that it cannot generate any arbitrary unitary transformation on a collection of spin-1/2 qubits. So, every proposal has supplemented the Heisenberg interaction with some other means of applying independent one-qubit gates (which can be thought of as time-dependent local magnetic fields). But the need to add this capability to the device adds considerably to the complexity of the structures, by putting unprecedented demands on “g-factor” engineering of heterostructure materials, requiring that strong, inhomogeneous magnetic fields be applied , or involving microwave manipulations of the spins that may be slow and may cause heating of the device . These added complexities may well exact a high cost, perhaps degrading the quantum coherence and clock rate of these devices by orders of magnitude.
The reason that the Heisenberg interaction alone does not give a universal quantum gate is that it has too much symmetry: it commutes with the operators $`\widehat{S}^2`$ and $`\widehat{S}_z`$ (for the total spin angular momentum and its projection on the $`z`$ axis), and therefore it can only rotate among states with the same $`S,S_z`$ quantum numbers. But by defining coded qubit states, ones for which the spin quantum numbers always remain the same, the Heisenberg interaction alone is universal, and single-spin operations and all their attendant difficulties can be avoided.
Recent work has identified the coding required to accomplish this. Starting with early work that identified techniques for suppressing phase-loss mechanisms due to coupling with the environment, more recent studies have identified encodings that are completely immune from general collective decoherence, in which a single environmental degree of freedom couples in the same way to all the spins in a block. These codes are referred to both as decoherence-free subspaces (and their generalization, the decoherence-free subsystems) , and also as noiseless subspaces and subsystems. The noiseless properties of these codes are not relevant to the present work; but they have the desired property that they consist of states with definite angular momentum quantum numbers.
So, in principle, the problem has been solved: the Heisenberg interaction alone is universal and can be used for quantum computation. However, a very practical question still remains: how great is the price that must be paid in return for eliminating single-spin operations? In particular, how many applications of the Heisenberg interaction are needed to complete some desired quantum gate operation? The only guidance provided by the existing theory comes from a theorem of Solovay and Kitaev, which states that “efficient” approximations exist: given a desired accuracy parameter $`ϵ`$, the number $`N`$ of exchange operations required goes like $`NK\mathrm{log}^c(1/ϵ)`$, where $`c4`$ and $`K`$ is an unknown positive constant. However, this theorem provides very little useful practical guidance for experiment; it does not show how to obtain the desired approximating sequence of exchange operations, and, since $`K`$ is unknown, it gives no clue of whether the number of operations needed for a practical accuracy parameter is 10 or 10000. In the following we remedy these inadequacies by showing that the desired quantum logic operations can be obtained exactly using sequences of exchange interactions short enough to be of practical significance for upcoming experiments.
In the scheme we analyze here, we use the smallest subspace with definite angular-momentum quantum numbers that can be used to encode a qubit; this subspace is made up of three spins. It should be noted that in principle the overhead in spatial resources could be made arbitrarily small: asymptotically the rate of encoding into such noiseless subsystems converges to unity. The space of three-spin states with spin quantum numbers $`S=1/2`$, $`S_z=+1/2`$ is two dimensional and will serve to represent our coded qubit. A good explicit choice for the basis states of this qubit are $`|0_L=|S|`$, $`|1_L=\sqrt{2/3}|T_+|\sqrt{1/3}|T_0|`$. Here $`|S=\sqrt{1/2}(||)`$ is the singlet state of spins 1 and 2 (see Fig. 1a) of the three-spin block, and $`|T_+=|`$ and $`|T_0=\sqrt{1/2}(|+|)`$ are triplet states of these two spins. For these states we have constructed an explicit exchange implementation of the basic circuit elements of quantum logic; in particular, we discuss how one obtains any coded one-qubit gate, and a specific two-qubit gate, the controlled NOT (cNOT).
It is easy to understand how one-qubit gates are performed on a single three-spin block. We note that Hamiltonian $`H_{12}`$ generates a rotation $`U_{12}=\mathrm{exp}(i/\mathrm{}J\stackrel{}{S}_1\stackrel{}{S}_2𝑑t)`$ which is just a $`z`$-axis rotation (in Bloch-sphere notation) on the coded qubit, while $`H_{23}`$ produces a rotation about an axis in the x-z plane, at an angle of 120<sup>o</sup> from the $`z`$-axis. Since simultaneous application of $`H_{12}`$ and $`H_{23}`$ can generate a rotation around the $`x`$-axis, three steps of 1D parallel operation (defined in Fig. 1a) permit any one-qubit rotation, using the classic Euler-angle construction. In serial operation, we find numerically that four steps are always adequate when only nearest-neighbor interactions are possible (eg, the sequence $`H_{12}`$-$`H_{23}`$-$`H_{12}`$-$`H_{23}`$ shown in Fig. 2a, with suitable interaction strengths), while three steps suffice if interactions can be turned on between any pair of spins (eg, $`H_{12}`$-$`H_{23}`$-$`H_{13}`$, see Fig. 2b).
We have performed numerical searches for the implementation of two-qubit gates using a simple minimization algorithm. Much of the difficulty of these searches arises from the fact that while the four basis states $`|0_L,1_L|0_L,1_L`$ have total spin quantum numbers $`S=1`$, $`S_z=+1`$, the complete space with these quantum numbers for six spins has nine states, and exchanges involving these spins perform rotations in this full nine-dimensional space. So, for a given sequence, eg the one depicted in Fig. 2c, one considers the resulting unitary evolution in this nine-dimensional Hilbert space as a function of the interaction times $`t_1`$, $`t_2`$, … $`t_N`$. This unitary evolution can be expressed as a product $`U(t_1,\mathrm{},t_N)=U_N(t_N)\mathrm{}U_2(t_2)U_1(t_1)`$, where $`U_n(t_n)=\mathrm{exp}(it_nH_{i(n),j(n)}/\mathrm{})`$. The objective of the algorithm is to find a set of interaction times such that the total time evolution describes a cNOT gate in the four-dimensional logic subspace $`U(t_1,\mathrm{},t_N)=U_{\mathrm{cNOT}}A_5`$. The matrix $`A_5`$ can be any unitary $`5\times 5`$ matrix (consistent with $`U`$ having a block diagonal form). The efficiency of our search is considerably improved by the use of two invariant functions $`m_{1,2}(U)`$ identified by Makhlin , which are the same for any pair of two-qubit gates that are identical up to one-qubit rotations. It is then adequate to use an algorithm that searches for local minima of the function $`f(t_1,\mathrm{},t_N)=_i(m_i(U_{\mathrm{cNOT}})m_i(U(t_1,\mathrm{},t_N)))^2`$ with respect to $`t_1`$, …$`t_N`$ ($`m_i`$ is understood only to act on the $`4\times 4`$ logic subspace of $`U`$). Finding a minimum for which $`f=0`$ identifies an implementation of cNOT (up to additional one-qubit gates, which are easy to identify) with the given sequence $`(i(n),j(n))_n`$, $`i(n)j(n)`$ of exchange gates. If no minimum with $`f=0`$ is found after many tries with different starting values (ideally mapping out all local minima), we have strong evidence (although not a mathematical proof) that the given sequence of exchange gates cannot generate cNOT.
The optimal serial-operation solution is shown in Fig. 2c. Note that by good fortune this solution happens to involve only nearest neighbors in the 1D arrangement of Fig. 1a. The circuit layout shown obviously has a high degree of symmetry; however, it does not appear possible to give the obtained solution in a closed form. (Of course, any gate sequence involving non-nearest neighbors can be converted to a local gate sequence by swapping the involved qubits, using the SWAP gate, until they are close; here however the minimal solution found does not require such manipulations.) We have also found (apparently) optimal numerical solutions for parallel operation mode. For the 1D layout of Fig. 1a, the simplest solution found involves 8 clock cycles with just 8\*4 different interaction-time parameters ($`H_{12}`$ can always be zero in this implementation). For the 2D parallel mode of Fig. 1b, a solution was found using just 7 clock cycles (7\*7 interaction times).
It is worthwhile to give a complete overview of how quantum computation would proceed in the present scheme. It should begin by setting all the computational qubits to the $`|0_L`$ state. This state is easily obtained using the exchange interaction: if a strong $`H_{12}`$ is turned on in each coded block and the temperature made lower than the strength $`J`$ of the interaction, these two spins will equilibrate to their ground state, which is the singlet state. The third spin in the block should be in the $`|`$ state, which can be achieved by also placing the entire system in a moderately strong magnetic field $`B`$, such that $`k_BT<<g\mu _BB<J`$. Then, computation can begin, with the one- and two-qubit gates implemented according to the schemes mentioned above. For the final qubit measurement, we note that determining whether the spins 1 and 2 of the block are in a singlet or a triplet suffices to perfectly distinguish $`|0_L`$ from $`|1_L`$ (again, the state of the third spin does not enter). Thus, for example, the AC capacitance scheme for spin measurement proposed by Kane is directly applicable to the coded-qubit measurement.
There are several issues raised by this work that deserve further exploration. The $`S=1/2`$, $`S_z=+1/2`$ three-spin states that we use are a subspace of a decoherence-free subsystem that has been suggested for use in quantum computing by exchange interactions . Use of this full subsystem, in which the coded qubit can be in any mixture of the $`S_z=+1/2`$ and the corresponding $`S_z=1/2`$ states, would offer immunity from certain kinds of interactions with the environment, and would not require any magnetic field to be present, even for initialization of the qubits. In this modified approach, the implementation of one-qubit gates is unchanged, but the cNOT implementation must satisfy additional constraints – the action of the exchanges on both the $`S=1`$ and the $`S=0`$ six-spin subspaces must be considered. As a consequence, implementation of cNOT in serial operation is considerably more complex; our numerical studies have failed to identify an implementation (even a good approximate one) for sequences of up to 36 exchanges (cf. 19 in Fig. 2c). On the other hand, we have found implementations using 8 clock cycles for 1D and 2D parallel operation (again for the 1D case $`H_{12}`$ can be zero), so use of this larger Hilbert space may well be advantageous in some circumstances.
Finally, we note that further work is needed on the performance of quantum error correction within this scheme. Our logical qubits can be used directly within the error correction codes that have been shown to produce fault tolerant quantum computation. Spin decoherence will primarily result in “leakage” errors, which would take our logical qubits into states of different angular momentum (eg, $`S=3/2`$). Our preliminary work indicates that, with small modifications, the conventional error correction circuits will not cause uncontrolled propagation of leakage error. In addition, the general theory shows that there exist sequences of exchange interactions which directly correct for leakage by swapping a fresh $`|0_L`$ into the coded qubit if leakage has occurred, and doing nothing otherwise; we have not yet identified numerically such a sequence. If fast measurements are possible, teleportation schemes can also be used in leakage correction.
To summarize, the present results offer a new alternative route to the implementation of quantum computation. The tradeoffs are clear: for the price of a factor of three more devices, and about a factor of ten more clock cycles, the need for stringent control of magnetic fields applied to individual spins is dispensed with. We are hopeful that the new flexibility offered by these results will make easier the hard path to the implementation of quantum computation in the lab.
Acknowledgments: DPD, DB, JK, and KBW acknowledge support from the National Security Agency (NSA) and the Advanced Research and Development Activity (ARDA). DPD also thanks the UCLA DARPA program on spin-resonance transistors for support, and is also grateful for the hospitality of D. Loss at the University of Basel, where much of this work was completed. JK also acknowledges support from the US National Science Foundation. The work of GB is supported in part by the Swiss National Science Foundation. Discussions with P. O. Boykin and B. M. Terhal are gratefully acknowledged.
|
warning/0005/nucl-ex0005001.html
|
ar5iv
|
text
|
# Antideuteron yield at the AGS and coalescence implications
## I Introduction
The production of antinucleons and antinuclei in heavy ion collisions is of significant interest for several reasons . Because the initial colliding system contains no antibaryons, their yields and spectra are determined solely by collision dynamics. At AGS energies ($`\sqrt{s}`$ = 4.8 A GeV), nucleon-antinucleon pair production is above threshold in individual $`Nucleon+Nucleon`$ collisions, but direct deuteron-antideuteron pair production is not. However, a small fraction of the colliding nucleon pairs which have high relative Fermi momenta may exceed the threshold for $`d`$-$`\overline{d}`$ pair production. We have estimated this contributes at a level at least two orders of magnitude below the measured yield from our experiment. Antideuterons are therefore predominantly formed in secondary interactions between directly produced antinucleons from the collision. Their production is then highly dependent on the total abundances and spatial distribution of the antinucleons.
Simple coalescence and thermal models indicate that differences in the measurement of the coalescence parameter between nuclei ($`B_A`$) and their antinuclei counterparts ($`\overline{B_A}`$) are due to differences in the source volumes, where
$`B_A`$ $``$ $`{\displaystyle \frac{\left(\frac{1}{2\pi p_t}\frac{d^2N_A}{dydp_t}\right)}{\left(\frac{1}{2\pi p_t}\frac{d^2N_p}{dydp_t}\right)^Z\left(\frac{1}{2\pi p_t}\frac{d^2N_n}{dydp_t}\right)^{AZ}}}`$ (1)
$``$ $`{\displaystyle \frac{\left(\frac{1}{2\pi p_t}\frac{d^2N_A}{dydp_t}\right)}{\left(\frac{1}{2\pi p_t}\frac{d^2N_p}{dydp_t}\right)^A}},`$ (2)
$`\overline{B_A}`$ $``$ $`{\displaystyle \frac{\left(\frac{1}{2\pi p_t}\frac{d^2N_{\overline{A}}}{dydp_t}\right)}{\left(\frac{1}{2\pi p_t}\frac{d^2N_{\overline{p}}}{dydp_t}\right)^A}}.`$ (3)
We have used an assumption in Equation 2 that $`n`$ and $`p`$ abundances in the region of the measurement are similar, which is only approximately true . Antineutron yields remain unmeasured.
The small binding energy of (anti)deuterons requires that they be formed near the hypersurface of the fireball where their mean free path is sufficiently large that they suffer no further collisions. The additional annihilation cross section for antinucleons implies that densities must be even lower for $`\overline{d}`$s than $`d`$s to have sufficient mean free paths for survival , possibly resulting in a more shell-like spatial formation zone for $`\overline{d}`$. This has fueled predictions that $`\overline{B_2}`$ will be notably smaller than $`B_2`$ .
However, other thermal model calculations suggest $`B_2`$ and $`\overline{B_2}`$ may be very similar. Microscopic models demonstrate that nucleons may undergo 20-30 collisions before their final interactions in heavy ion collisions at AGS and SPS energies . The more collisions the constituents undergo, the further towards thermal and chemical equilibrium the system is driven. If equilibrium is complete, and all particles freeze-out along the same hypersurface, no difference would be expected between $`B_2`$ and $`\overline{B_2}`$ . An analysis of equilibrium conditions in the collisions studied by E864 is presented elsewhere .
A sample of two $`\overline{d}`$s was previously measured in the AGS Experiment 858 in minimum bias Si + Au collisions at 14.6 GeV/c per nucleon, indicating that in such a system $`\overline{B_2}`$ is below or at the level of $`B_2`$ . However, this measurement involves a smaller system size where antimatter annihilation may not be as prominent as in central Au + Pt collisions. Antiproton yields are significantly suppressed from first collision scaling in the larger colliding systems compared to peripheral collisions or Si + Au collisions . Some transport models indicate that over 90% of the originally produced $`\overline{p}`$s are annihilated, while other calculations include a screening of the annihilation in this dense environment thus reducing the losses . These large collisions, consequently, provide a good place to see the effects of annihilation on $`\overline{d}`$ production. Interestingly, there may be additional processes in these collisions whose contributions to antimatter coalescence rates are not completely understood, such as findings that a significant portion of the antibaryon number is carried away in the form of strange antibaryons .
## II EXPERIMENT
In order to search for rare products from heavy ion collisions at the AGS, Experiment 864 was designed as an open geometry, high rate spectrometer . It features a multiplicity detector for triggering on central collisions , and a level 2 trigger capable of selecting events in which high mass objects traverse the spectrometer, using a hadronic calorimeter to provide fast energy and time measurements . The calorimeter energy measurement encompasses the kinetic energy for normal hadronic matter, with an additional annihilation energy of approximately twice the particle mass for antimatter. Tracking of charged particles is performed primarily with three scintillating hodoscope TOF walls interspersed with two straw tube tracking stations for improved track spatial resolution. The tracking system is used to determine velocities and rigidities of charged particles accurately, providing mass resolutions typically better than 5% .
The spectrometer magnets can be set to different field strengths and polarities which optimize acceptance for particles of interest. During the 1996-1997 run of the experiment, over 250 million triggers were taken with both magnets set to -0.75T. This setting allows reasonable acceptance for $`\overline{d}`$s while most positively charged particles and lighter negatively charged particles are swept out of the acceptance. Figure 1 shows the $`\overline{d}`$ geometric acceptance for this field setting.
## III ANALYSIS
Figure 2a shows the mass spectrum of charge $`1`$ particles from the data set in the region near the $`\overline{d}`$ mass ($`1.8<y<2.2`$) for tracks which have passed basic quality cuts. Simulations have been performed to study the background evident in the data, which is understood to have two predominant sources. In the rapidity range shown in Figure 2a, the background is dominated by neutrons passing through the spectrometer magnets before undergoing charge-exchange reactions ($`n+Xp+X^{}`$), resulting in stiff tracks which are calculated with incorrect masses and charge signs. However, the energy measured by E864’s hadronic calorimeter should be consistent with the kinetic energy of a $`p`$ traveling at the speed calculated from the hodoscope TOF measurements . This energy measurement can then be used to cut away background whose calorimeter response is reasonably consistent with that of a $`p`$. Such a cut is used in the mass spectrum shown in Figure 2b, revealing a clear peak at the $`\overline{d}`$ mass. The chosen cut requires that the calorimeter response is greater than $`4\sigma `$ above the expected response for a $`p`$ in order to remove as much background as possible without destroying the $`\overline{d}`$ signal (the cut is 95% efficient for $`\overline{d}`$s, aided by the significant energy contributions from annihilation in the calorimeter). While a $`4\sigma `$ cut would normally be expected to remove nearly 100% of the scattered $`p`$s, the efficiency is reduced in this case because of the bias from the level 2 trigger in selecting candidates whose calorimeter responses are large. Because this cut is independent of the background’s calculated mass, the shape of the background is unaffected. A fit to the shape is made before the cut and used in fitting for signal plus background after the cut. The resulting $`\overline{d}`$ yield is $`17.6\pm 7.5`$ counts, where the statistical error includes contributions from a background of $`34.1\pm 5.8(Poissonstat.)\pm 2.3(normalization)`$.
Similarly, in the rapidity range between 1.4 and 1.8 the background is understood to be dominated by $`\overline{p}`$s which have scattered by small angles near the spectrometer magnets. Again, the calorimeter can be used to reject $`\overline{p}`$s in the mass spectrum, although the additional energy deposited from annihilation in the calorimeter leads to a smaller separation from $`\overline{d}`$s for $`\overline{p}`$s than $`p`$s. A $`4\sigma `$ cut would severely impinge upon the $`\overline{d}`$ signal, so a less stringent $`2\sigma `$ cut is chosen with the resulting efficiency for $`\overline{d}`$s at 86%. A $`\overline{d}`$ signal is found with $`4.6\pm 3.3`$ counts, where the statistical error includes contributions from a background of $`5.4\pm 2.3(Poissonstat.)\pm 0.7(normalization)`$.
In order to calculate invariant yields, the transverse momentum distribution of the measured $`\overline{d}`$s must also be understood as well as possible. Because it is not known which candidates under the mass peak are the true $`\overline{d}`$s, the $`p_t`$ distribution is calculated for random selections of the candidates. For each selection, the acceptance-corrected count is determined by acceptance correcting every candidate at its measured $`y`$ and $`p_t`$. Millions of random selections are made and a distribution of acceptance-corrected counts is found with a most probable value and widths which indicate the systematic errors of the method. Between the rapidities of 1.8 and 2.2, the candidates have an acceptance-weighted $`<`$$`p_t`$$`>`$ of 0.35 GeV/c, and have an upper limit of $`p_t=1`$ GeV/c. Correcting for all efficiencies results in a $`\overline{d}`$ invariant yield of $`3.5\pm 1.5(stat.)_{0.5}^{+0.9}(sys.)\times 10^8`$ GeV$`{}_{}{}^{2}c_{}^{2}`$.
In the rapidity range between 1.4 and 1.8, the candidates have an acceptance-weighted $`<`$$`p_t`$$`>`$ of 0.26 GeV/c, and have an upper limit of $`p_t=0.5`$ GeV/c. All possible selections of the candidates are tried, determining a yield of $`3.7\pm 2.7(stat.)_{1.5}^{+1.4}(sys.)\times 10^8`$ GeV$`{}_{}{}^{2}c_{}^{2}`$.
E864 has also measured $`\overline{p}`$ yields in 10% central Au + Pb collisions at 11.5 GeV/c per nucleon . These yields are shown along with the new $`\overline{d}`$ measurements in Figure 3 reflected about mid-rapidity. The new $`\overline{d}`$ yields are in agreement with upper limits previously published by E864 taken from a smaller data sample .
The measured $`\overline{p}`$ and $`\overline{d}`$ yields allow us to calculate the coalescence factor $`\overline{B_2}`$ using Equation 3. The E864 $`\overline{p}`$ measurement must be corrected for contributions from antihyperon decays as these do not participate in the coalescence process. Here, we will use the most probable value for $`\overline{Y}/\overline{p}=3.5`$, and use the 98% confidence limit of $`\overline{Y}/\overline{p}>2.3`$ to define the systematic error of the correction . This is a significant correction made from an indirect measurement of $`\overline{Y}/\overline{p}`$ which attributes the entire difference in $`\overline{p}`$ yields between two experiments to their acceptance for antihyperon decay contributions, introducing sizable uncertainties into our calculation of $`\overline{B_2}`$. Additionally, as coalescence is a process affecting co-moving nucleons, yields must be taken at the same $`<`$$`p_t`$$`>`$$`/`$$`A`$. In the mid-rapidity bin ($`1.4<y<1.8`$), this means using the $`\overline{p}`$ yield at $`p_t=0.13`$ GeV/c, for which E864’s $`\overline{p}`$ measurements are valid. The final value of $`\overline{B_2}`$ comes out as $`4.1\pm 2.9(stat.)_{2.4}^{+2.3}(sys.)\times 10^3`$ GeV$`{}_{}{}^{2}c_{}^{3}`$, where the systematic error includes contributions from the $`\overline{d}`$ acceptance correction and the antihyperon-feeddown correction to the $`\overline{p}`$s.
E864 has also measured $`B_2`$ in the range $`0.1<p_t/A<0.2`$ GeV/c as $`1.06\pm 0.15\times 10^3`$ GeV$`{}_{}{}^{2}c_{}^{3}`$ . While E864’s $`\overline{B_2}`$ measurement is above its $`B_2`$ measurement, the two are within statistical and systematic errors of each other. These values are shown in Figure 4 along with those from other experiments studying collisions with very large numbers of participant nucleons ($`400`$). Along with the SPS results, the evidence at this point suggests that there may be no difference between coalescence of matter and antimatter in central heavy ion collisions. This may be an indication that the freeze-out hypersurface for antideuterons is not substantially modified by antimatter annihilation from that of deuterons, and is consistent with predictions based on a thermalized source . Predictions of $`\overline{B_2}`$ supression from $`B_2`$ are dependent on such modifications and diminish in this scenario.
## IV SUMMARY AND ACKNOWLEDGEMENTS
E864 has measured an antideuteron signal in central heavy ion collisions at the AGS, where coalescence is likely to be the dominant method for production. The measured invariant yields are $`3.5\pm 1.5(stat.)_{0.5}^{+0.9}(sys.)\times 10^8`$ GeV$`{}_{}{}^{2}c_{}^{2}`$ ($`1.8<y<2.2`$, $`<`$$`p_t`$$`>=0.35`$ GeV/c) and $`3.7\pm 2.7(stat.)_{1.5}^{+1.4}(sys.)\times 10^8`$ GeV$`{}_{}{}^{2}c_{}^{2}`$ ($`1.4<y<1.8`$, $`<`$$`p_t`$$`>=0.26`$ GeV/c). The measured coalescence parameters from E864 for matter ($`B_2=1.06\pm 0.15\times 10^3`$ GeV$`{}_{}{}^{2}c_{}^{3}`$) and antimatter ($`\overline{B_2}=4.1\pm 2.9(stat.)_{2.4}^{+2.3}(sys.)\times 10^3`$ GeV$`{}_{}{}^{2}c_{}^{3}`$) at mid-rapidity are within statistical and systematic errors of each other.
We gratefully acknowledge the excellent support of the AGS staff. This work was supported in part by grants from the U.S. Department of Energy’s High Energy and Nuclear Physics Divisions, the U.S. National Science Foundation.
Present address: Vanderbilt University, Nashville, Tennessee 37235
Present address: Istituto di Cosmo-Geofisica del CNR, Torino, Italy / INFN Torino, Italy
Present address: Anderson Consulting, Hartford, CT
Present address: Univ. of Denver, Denver CO 80208
Deceased.
Present address: Cambridge Systematics, Cambridge, MA 02139
Present address: McKinsey & Co., New York, NY 10022
Present address: Department of Radiation Oncology, Medical College of Virginia, Richmond, VA 23298
Present address: University of Tennessee, Knoxville, TN 37996
Present address: Institut de Physique Nucléaire, 91406 Orsay Cedex, France
Present address: Institute for Defense Analysis, Alexandria, VA 22311
Present address: MIT Lincoln Laboratory, Lexington, MA 02420-9185
|
warning/0005/cond-mat0005337.html
|
ar5iv
|
text
|
# Crystal Field -𝐴𝑆_𝑧² Does Not Produce One-Phonon Transitions With Δ𝑆_𝑧=±2 [Comment on EPL 46, 692 (1999) by Leuenberger and Loss]
Recently Leuenbeger and Loss suggested a theory of phonon-assisted relaxation in a molecular nanomagnet Mn-12 that “contrary to previous results is in reasonably good agreement not only with the relaxation data but also with all experimental parameter values known so far” . The purpose of this Comment is to show that the model of Leuenberger and Loss and its comparison with experiment are premised upon a principal error.
The theory of Ref. follows the footsteps of our earlier work that describes phonon-assisted magnetic relaxation in Mn-12 in terms of the master equation for the density matrix. The dominant term in the crystal field of Mn-12 is $`AS_z^2`$. Performing rotation of the anisotropy axis due to the elastic deformation $`𝐮`$, one obtains the following magnetoelastic coupling
$$A(\omega _{xz}\{S_x,S_z\}+\omega _{yz}\{S_y,S_z\}=(A/2)[(\omega _{xz}i\omega _{yz})\{S_+,S_z\}+(\omega _{xz}+i\omega _{yz})\{S_{},S_z\}],$$
(1)
where $`\omega _{\alpha \beta }=\frac{1}{2}(_\beta u_\alpha _\alpha u_\beta )`$. Operators $`S_\pm `$ change the $`S_z`$ projection of spin by $`\pm 1`$, while $`𝐮`$ is linear on the operators of creation and annihilation of phonons. Consequently, Eq. (1), when quantized, describes emission and absorption of one phonon accompanied by $`\mathrm{\Delta }S_z=\pm 1`$.
According to Leuenberger and Loss , the crystal field $`AS_z^2`$ also produces the magnetoelastic coupling of the form
$$A(ϵ_{xx}ϵ_{yy})(S_x^2S_y^2)=(A/2)(ϵ_{xx}ϵ_{yy})(S_+^2+S_{}^2).$$
(2)
where $`ϵ_{\alpha \beta }`$ is the strain tensor. This coupling generates “second-order” spin-phonon transitions with $`\mathrm{\Delta }S_z=\pm 2`$, which “lead to a much faster relaxation of the spin system” than “first-order” transitions with $`\mathrm{\Delta }S_z=\pm 1`$.
The error stems from the use by Leuenbeger and Loss of the linear formula for the strain rensor,
$$ϵ_{\alpha \beta }=\frac{1}{2}\left(\frac{u_\alpha }{x_\beta }+\frac{u_\beta }{x_\alpha }\right),$$
(3)
instead of the exact expression ,
$$ϵ_{\alpha \beta }=\frac{1}{2}\left(\frac{u_\alpha }{x_\beta }+\frac{u_\beta }{x_\alpha }+\frac{u_\gamma }{x_\alpha }\frac{u_\gamma }{x_\beta }\right).$$
(4)
To obtain Eq. (2) Leuenberger and Loss computed the rotation matrix $`\widehat{R}`$ and the corresponding deformation, $`𝐮=(\widehat{R}_z\widehat{R}_y\widehat{R}_x1)𝐱`$, up to the second order in infinitesimal rotation $`\delta \varphi _\alpha `$. They obtained
$`u_x`$ $`=`$ $`\delta \varphi _yz\delta \varphi _zy(1/2)(\delta \varphi _y^2+\delta \varphi _z^2)x`$
$`u_y`$ $`=`$ $`\delta \varphi _zx\delta \varphi _xz(1/2)(\delta \varphi _x^2+\delta \varphi _z^2)y`$
$`u_z`$ $`=`$ $`\delta \varphi _xy\delta \varphi _yx(1/2)(\delta \varphi _x^2+\delta \varphi _y^2)z`$ (5)
The incorrect Eq. (3) then gives (incorrectly)
$$ϵ_{xx}=(\delta \varphi _y^2+\delta \varphi _z^2)/2,ϵ_{yy}=(\delta \varphi _x^2+\delta \varphi _z^2)/2,ϵ_{zz}=(\delta \varphi _x^2+\delta \varphi _y^2)/2,$$
(6)
and one gets $`\delta \varphi _x^2=ϵ_{xx}ϵ_{yy}ϵ_{zz}`$ and cyclic permutations for $`\delta \varphi _y^2`$ and $`\delta \varphi _z^2`$. Substituting this into $`\widehat{R}`$ and inserting the rotated spin vector $`R_xR_y𝐒`$ into $`AS_z^2`$, Leuenberger and Loss obtained Eq. (2). One should notice, however, that the substitution of Eq. (5) into the correct expression for the strain tensor, Eq. (4), yields
$$ϵ_{xx}=ϵ_{yy}=ϵ_{zz}=0,$$
(7)
in accordance with the fact that rotation does not change the volume, $`\delta V=Vϵ_{\alpha \alpha }=0`$.
In fact, to see that Eq. (2) cannot be right no calculation is needed. Indeed, the operator $`AS_z^2`$ conserves $`S_z`$. In order to change the magnetic quantum number of Mn-12 molecule by 2, one would have to assign to phonons spin 2. Phonons, however, being described by a vector field $`𝐮`$, cannot possess spin other than 1 . One-phonon processes accompanied by $`\mathrm{\Delta }S_z=\pm 2`$ can only be generated by terms in the crystal field which do not conserve $`S_z`$. For Mn-12 these are tunneling terms which are orders of magnitude smaller than the uniaxial crystal field. Leuenberger and Loss (see Appendix D of Ref. ) find support of their incorrect statement on p. 563 of Abragam and Bleaney . However, Abragam and Bleaney, when talking about “first-” and “second-order” transitions, mean two-phonon Raman processes, which, of course, are not prohibited by the above argument. These two-phonon processes die out at low temperature. Their rate is inversely proportional to the tenth power of the sound velocity, as compared to the fifth power for the one-phonon processes. Leuenberger and Loss compare their theory with experiment based upon the extraction of the sound velocity from the measured relaxation rate, using a wrong type of the phonon process. Their claim of agreement with experiment is, therefore, completely invalidated by their incorrect model for the spin-phonon coupling.
This work has been supported by the NSF Grant No. DMR-9978882.
|
warning/0005/hep-th0005083.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
In recent years, a lot of activity has been devoted to the study of Quantum Field Theory in the framework of the Exact Renormalization Group Equation (ERGE) of Wilson . The principal advantage of this approach is the fact that the ERGE can be solved numerically at the non-perturbative level: this feature in principle pave the way to a number of important applications. However, in practice, the method works well for scalar theories, where there is an extensive literature and a collection of reliable numerical results (see for a recent review and a complete bibliography) whereas its application to gauge theories should be regarded as problematic. The root of the difficulties becomes clear if we remind the logic of the ERGE approach in the formalism of the one-particle irriducible (1PI) effective action , which is the following: i) take a given Quantum Field Theory and introduce a fictitious infrared cutoff $`\mathrm{\Lambda }`$ by modifing the bare propagator in some way; ii) write the functional identity which describes how the 1PI effective action of the theory changes when the infrared cutoff is lowered; iii) fix the boundary conditions at the ultraviolet scale $`\mathrm{\Lambda }_0`$ by identifying the effective action with the local bare action and integrate down numerically or analytically in perturbation theory the equation up to the the infrared value $`\mathrm{\Lambda }\mathrm{\Lambda }_0`$.
This logic is perfectly legitimate in scalar theories, but quite delicate in the case of gauge theories since there is no way of inserting the infrared cutoff in a way consistent with gauge-invariance (we mean with BRST-invariance): as a consequence the $`\mathrm{\Lambda }0`$ theory is gauge-dependent and unphysical and the limit $`\mathrm{\Lambda }0`$ must be taken strictly. This generates two distinct problems.
The first problem, which has been extensively studied in the literature , is the following. Since the infrared cutoff breaks BRST-invariance, the usual Slavnov-Taylor identities have to be modified and assume a very complicate form such that it is practically impossible to find truncations of the effective action consistent with the modified identities. As a consequence at the non-perturbative level there is no rigorous control on the fact that the gauge-invariant theory is recovered when the infrared cutoff is removed. On the contrary at the perturbative level this is guaranteed by a theorem: no matter how the infrared cutoff is inserted, the $`\mathrm{\Lambda }0`$ theory is gauge invariant provided that we choose the ultraviolet bare action in a very precise way, which can be established in perturbation theory to all orders by solving the fine-tuning equations discussed in . Therefore, at least in principle we could consider solved this problem, even if in practice the fine-tuning equations are quite involved and cannot be solved out of perturbation theory.
The second problem, which in our knowledge has been treated only partially in the literature, is the question of the regularity of the $`\mathrm{\Lambda }0`$ limit. Actually the smoothness of this limit is guaranteed only for the proper vertices at non-exceptional configuration of momenta : but these are not physical quantities. On the contrary, the limit is extremely delicate in the computation of physical quantities as for instance the interquark potential: we will see in this paper that the limit must be taken with great care in order to have sensible results. In such a situation, a naive numerical analysis of the problem is certainly questionable. Apparently, this subtle point has never been recognized in the previous literature and is explicitly analyzed here for the first time.
The guideline of our research program consists in setting up a formalism where these problems can be faced analytically with perturbative computations, in such a way to have definite answers on the recovering of the gauge invariance and on the regularity or the singularity of the $`\mathrm{\Lambda }0`$ limit for physical quantities. In this logic, in a couple of recent papers we presented a Wilsonian formulation of gauge theories in which the Wilsonian infrared cutoff $`\mathrm{\Lambda }`$ is introduced as a mass-like term for the propagating fields. In this way the Wilson’s Renormalization Group Equation for the 1PI effective action becomes consistent with the Ward-Takahashi identities to all scales. This important property, which allows to solve the problem of finding gauge-consistent truncations but not the question on the regularity of the $`\mathrm{\Lambda }0`$ limit, was proved in the Abelian case in covariant gauges and in the non-Abelian case in algebraic non-covariant gauges . In this latter case, the crucial point was the fact that in algebraic non-covariant gauges the ghost fields decouple and the gauge-symmetry can be implemented via simple Ward-Takahashi identities instead that in terms of complicated Slavnov-Taylor identities. However, as we said in , even if the implementation of the gauge symmetry is quite efficient in these formulations there are various disadvantages to be taken in account. The first disadvantage, common also to the covariant version of the method, is the need for an explicit regularization of the evolution equation, since when the mass cutoff is employed the ultraviolet momenta are not sufficiently suppressed. However, in our view this is a little price to pay since, at least in perturbation theory, there exist gauge-invariant techniques to manage the ultraviolet problem (dimensional regularization, higher derivative regularization, etc.). The second disadvantage is the gauge-dependence of the would-be physical quantities at $`\mathrm{\Lambda }0`$. However, even this problem is not so dramatic since the gauge-dependence can be controled throught the generalized functional identities described in . It is only the third disadvantage, i.e. the fact that non-covariant gauges have a very delicate $`\mathrm{\Lambda }0`$ limit, which is a very serious one. This feature gives rise to difficulties since if the $`\mathrm{\Lambda }0`$ limit does not exist for quantities which are instead well defined in other gauges the whole approach should be considered as suspicious. In practice this fact does not prevent phenomenological applications since there are infrared safe quantities which can be computed in some approximation (i.e. by a truncation of the evolution equation) without encountering problems. Nevertheless in questions of principle one would like to have a formalism with a well definite perturbative expansion and fully consistent with other gauges for any physical quantity. This is why in this paper we carefully study the infrared limit of the non-Abelian theory in different algebraic non-covariant gauges. We will see in particular that the planar gauge and the light-cone gauge are regular in the $`\mathrm{\Lambda }0`$ limit whereas the axial gauge is singular. The origin of the problems of the axial gauge will be connected with the spurious infrared divergences introduced by the double pole term in the propagator, i.e. the term proportional to $`n^2q_\mu q_\nu /[nq]^2`$ . This is hardly unexpected, since it is well known that in the standard approach using the Cauchy Principal Value (CPV) prescription the singularities connected with the double pole are responsible for many serious problems, as the loss of unitarity and the failure of the Wilson loop consistency check . In particular this latter quantity is the most interesting to study since it is connected with the interquark potential, which is the typical quantity one would like to compute non-perturbatively in the Wilson renormalization group approach . We recall that problems are expected to be present in the perturbative computation of this quantity since it was shown long time ago, working in the temporal axial gauge that the Wilson loop of sides $`2L\times 2T`$ which directly corresponds to the interquark potential computed at order $`g^4`$ is different from the corresponding computation in covariant gauges. More dramatically, in the standard perturbative expansion, the temporal gauge Wilson loop does not exponentiate in the limit $`T\mathrm{}`$ i.e. it is not of the form
$$W_{\mathrm{\Gamma }_{LT}}\stackrel{T\mathrm{}}{=}\mathrm{exp}[i\mathrm{\hspace{0.33em}2}TV(2L)],$$
(1)
(see for a critical analysis). The same problem was found in the spatial axial gauge for loops in the Euclidean space with the side $`T`$ in the direction of the axial vector $`n^\mu `$ <sup>1</sup><sup>1</sup>1Obviously there exist loops which give the same result in axial gauge and in Feynman gauge and in general smooth loops are expected to be safe. . The origin of the problem was found in the Cauchy Principal Value Prescription which in fact breaks gauge-invariance. There has been a lot of work in the literature to try to solve this problem (see the references cited in the monograph ). In the Wilsonian approach one naturally obtains a regularization of spurious divergences different from the CPV prescription and which is consistent with unitarity. Nevertheless, as we anticipated in , still one expects to have problems in the infrared limit $`\mathrm{\Lambda }0`$. These problems are discussed in detail in this paper: in addition we prove that they are absent in more stable algebraic non-covariant gauges, as the planar gauge and the light-cone gauge. This is not a surprise since it is well known in the literature that these gauge are safe with traditional prescriptions (the CPV for the planar gauge and the Mandelstam-Leibbrandt (ML) for the light-cone gauge) and in particular the Wilson loop consistency check is successfully passed at least at order $`O(g^4)`$ in perturbation theory. The light-cone gauge in particular should be considered as a promising starting point for our analysis. However we stress that to prove that the Wilsonian i.e. massive formulation tends to the standard massless formulation (which is safe) in the $`\mathrm{\Lambda }0`$ limit is a subtle point since there are infrared sensitive quantities where the massive version and the massless version of the same gauge choice are different, i.e. the $`\mathrm{\Lambda }0`$ limit is singular. The non-trivial point is to show that this does not happen for physical quantities. In principle this could also be false in a finite order perturbative computation. As we said, one of the motivation of this paper it to set up a workable formalism where such a question can be answered with a reasonable analytic effort.
In order to compare and contrast with the previous Wilsonian literature, we remind that a Wilsonian approach to algebraic non-covariant gauges has already been presented in ; nevertheless our analysis differs from that approach, where a generic cutoff function is employed thus totally obscuring a number of important properties and difficulties of the formalism: in particular the gauge-invariance issue and the infrared problems can be clearly studied only within our framework. In spite of the peculiarities of our approach, we stress that our results should be regarded as quite general, since there is a large class of cutoff functions which in the infrared behave as a mass term: for instance the most used cutoff in the numerical ERGE literature the exponential cutoff
$$R_\mathrm{\Lambda }(q)=\frac{q^2\mathrm{exp}(q^2/\mathrm{\Lambda }^2)}{1\mathrm{exp}(q^2/\mathrm{\Lambda }^2)}$$
(2)
is in this class since $`R_\mathrm{\Lambda }(q)\mathrm{\Lambda }^2`$ for $`q^2\mathrm{\Lambda }^2`$ and thus soft momenta are effectively screened as if there were a mass $`\mathrm{\Lambda }^2`$. This is not, however, the general case, since for instance the sharp cutoff is not in this class<sup>2</sup><sup>2</sup>2 This can be seen for instance by using the approximation to the sharp cutoff
$$R_\mathrm{\Lambda }(q)=\frac{q^2\mathrm{exp}((q^2/\mathrm{\Lambda }^2)^b)}{1\mathrm{exp}((q^2/\mathrm{\Lambda }^2)^b)}$$
with $`b\mathrm{}`$. In this case the infrared modes are very strongly suppressed since $`R_\mathrm{\Lambda }(q)`$ is divergent as $`q^20`$. The same is obtained with other regularizations of the sharp cutoff.. In general we expect that the situation of the infrared limit for other cutoff functions is worse than in the mass-cutoff case, since the gauge-symmetry is completely broken, Ward-Takahashi identities have a cumbersome form, and possible gauge-cancellations between different divergent contributions are lost. Actually one could speculate that at the non-perturbative level, due to infinite resummations of classes of Feynman diagram the infrared problem could be eventually solved. However, as a matter of fact, the perturbative expansion is the only controled analytical tool we have at our disposition. This is why we are interested in giving a Wilsonian formulation in which i) a well defined perturbative expansion can be defined and ii) it is possible to perform practical, analytical computations with an effort not too much bigger than in the standard massless approach. In particular in this work we discuss a general method and a series of tricks to analytically compute finite parts of one-loop Feynman diagrams in the $`\mathrm{\Lambda }0`$ limit. This is relevant in preparation to a computation of the Wilson loop up to order $`O(g^4)`$.
The plan of the paper is the following: in section 2 we review the basics of the standard gauge-fixing procedure, pointing out some subtle aspects which are usually not noticed since an infrared regularization consistent with BRST symmetry is assumed, which is not our case. In section 3 we study the properties of the general linear gauge $`L^\mu (p)A_\mu ^a(p)=0`$ in presence of a mass-like cutoff. In particular the massive axial gauge, the massive planar gauge and the massive light-cone gauge are analyzed. In section 4 we discuss the problems of the axial gauge as connected to the presence of the double pole in the propagator, whereas in sections 5 and 6 we show that these problems are absent in planar and light-cone gauge. In section 7 the Wilson loop for a finite rectangular path is explicitly computed up to order $`O(g^2)`$ and it is shown that the physical limit $`\mathrm{\Lambda }0`$ exists and is independent of the gauge-fixing choice. Section 8 resume our work and contains our conclusions. Two appendices close the paper: appendix A is a simple but instructive exercice that shows the gauge-dependence problem in the computation of a physical quantity, the pressure of a free photon gas; appendix B collects the technical tools and tricks which are used in the text to compute one-loop Feynman diagrams.
## 2 Remarks on the gauge-fixing procedure
In a strictly gauge-invariant theory one cannot define a gauge-field propagator since strict gauge-invariance implies that the free two-point function is transverse and therefore cannot be inverted. As a consequence the starting point of the perturbative analysis in terms of Feynman diagrams is missing. Thus, in order to define a perturbative quantum field theory from a classical gauge invariant field theory one is forced to break gauge invariance through the addition of a gauge-fixing term to the classical action<sup>3</sup><sup>3</sup>3Outside perturbation theory, for instance in the lattice approach, one can define the theory without an explicit gauge-fixing. However in order to show that the continuum limit exists, which has been rigorously proved only in perturbation theory, a gauge-fixing term is needed even in the lattice formulation. In general the gauge-fixing term is unavoidable to make contact with the perturbative formulation, which is the only one we analytically control.. The way gauge-invariance is broken depends on the chosen gauge-fixing term, which is at large extent arbitrary: the essential requirement is that the gauge-fixing term, depending on the gauge field and other fields (ghosts, antighosts and auxiliary fields), must be a BRST-cocycle: in this hypothesis it is possible to prove that the physical quantities are gauge-fixing independent and therefore that there are not physical arbitrarities.
In the path integral approach the gauge-fixing is usually introduced by means of the Faddeev-Popov argument: since there is an infinite number of gauge-equivalent configurations giving the same contribution to the functional integral the partition function is ill defined. Then the problem is solved by integrating only on representative of any gauge orbit. The representative are fixed by means of a gauge fixing condition of kind $`^a(A)=0`$; typical choices are $`^a(A)=^\mu A_\mu ^a`$ (covariant gauges) or $`^a(A)=n^\mu A_\mu ^a`$ (algebraic non-covariant gauges). These latter are interesting even because they avoid the problem of Gribov copies, i.e. the choice of the representative of the gauge orbit is unique .
Now it is convenient to introduce a bit of notations (the full list of our notations is reported in ). A gauge transformations of parameters $`\delta \omega =\delta \omega T=\delta \omega ^aT_a`$ is denoted by
$$\delta _GA_\mu =D_\mu \delta \omega ,\delta _G\psi =ig\delta \omega \psi ,\delta _G\overline{\psi }=ig\overline{\psi }\delta \omega ,$$
(3)
where $`T_a`$ are the generators of $`SU(N_c)`$ in the fundamental representation and $`N_c=3`$ is the number of colors. The set of physical fields of theory is denoted by $`\varphi =(A_\mu ^a(p),\psi _i(p),\overline{\psi }^i(p))`$ and the bare action with $`S_B(\varphi )`$; the expression $`\frac{\delta }{\delta \omega }`$ denotes the differential operator $`\frac{\delta }{\delta A_\mu }D_\mu `$ . A gauge-invariant ultraviolet regularization, which exists in non-chiral theories, is understood. With these notations the gauge-fixed partition function can be written as
$$Z_{}=[d\varphi ]det\frac{\delta }{\delta \omega }\delta ((A))e^{iS_B(\varphi )}.$$
(4)
The usefulness of the gauge-fixed approach is based in the following fundamental property: if the observable $`O(\varphi )`$ is gauge-invariant, then its mean value in the gauge $`^a(A)=0`$, defined as
$$<O(\varphi )>_{}=\frac{1}{Z_{}}[d\varphi ]det\frac{\delta }{\delta \omega }\delta ((A))e^{iS_B(\varphi )}O(\varphi ),$$
(5)
is independent of the gauge-fixing function $`^a`$:
$$\delta _GO(\varphi )=0<O>_{}=<O>_{^{}},^{}.$$
(6)
This means that in the computation of physical quantities we have the freedom of choosing any possible gauge-fixing.
We think that it is useful to review here the formal argument that support this statement with the scope of showing where and why this argument fails in the Wilsonian formalism.
The first step consists in introducing anticommuting fields $`C`$ and $`\overline{C}`$ and commuting auxiliary fields $`\lambda `$ in order to rewrite the determinant $`det\frac{\delta }{\delta \omega }`$ as a functional integral over Grassmannian variables<sup>4</sup><sup>4</sup>4Here we use the hermiticity conditions $`C=C^{}`$ and $`\overline{C}=\overline{C}^{}`$ such as the relation $`(\overline{C}\frac{\delta }{\delta \omega }C)^{}=\overline{C}\frac{\delta }{\delta \omega }C`$ holds.
$$det\frac{\delta }{\delta \omega }[dCd\overline{C}]\mathrm{exp}\left[i\overline{C}\frac{\delta }{\delta \omega }C\right]$$
(7)
and the functional delta function as
$$\delta ((A))=[d\lambda ]\mathrm{exp}\left[i\lambda (A)\right];$$
(8)
then the partition function (defined up to a multiplicative constant) can be rewritten in the form
$$Z_{}=[dCd\overline{C}d\lambda d\varphi ]\mathrm{exp}\left[iS_{}(\varphi ,C,\overline{C},\lambda )\right]$$
(9)
where the total bare action
$$S_{}=S_B+\lambda (A)\overline{C}\frac{}{A_\mu }D_\mu C$$
(10)
is invariant under the BRST-tranformation
$$sA_\mu =D_\mu C,sC=\frac{1}{2}gC\times C,s\overline{C}=\lambda ,s\lambda =0$$
(11)
for any choice of the gauge-fixing function. The cohomological property of the BRST tranformation $`s^2=0`$ allows us to prove the independence of BRST-invariant quantities from the gauge-fixing choice. Consider for instance a class of gauge-fixing functions $`_\epsilon ^a(A)`$ determined by one or more continuous parameters $`\epsilon `$: then we have to prove that physical quantities are independent of $`\epsilon `$. This can be easily shown by observing that the gauge-fixing term can be rewritten as a trivial cocicle
$$\lambda _\epsilon (A)\overline{C}\frac{\delta _\epsilon }{\delta A_\mu }D_\mu C=s\left(\overline{C}_\epsilon (A)\right).$$
(12)
Using the fact that trivial cocycles do not contribute to the partition function, i.e.
$$[d\mathrm{\Phi }]e^{iS_{}(\mathrm{\Phi })}sf(\mathrm{\Phi })0,$$
(13)
(we have denoted by $`\mathrm{\Phi }=(\varphi ,C,\overline{C},\lambda )`$ the full set of fields of the theory) one sees that the partition function is $`\epsilon `$independent
$$\frac{\text{d}}{\text{d}\epsilon }Z_\epsilon =[d\mathrm{\Phi }]e^{iS_{}(\mathrm{\Phi })}s\left(\overline{C}\frac{\text{d}}{\text{d}\epsilon }_\epsilon (A)\right)=0$$
(14)
and the same holds for BRST-invariant observables $`O(\mathrm{\Phi })`$ such as $`sO=0`$:
$$\frac{\text{d}}{\text{d}\epsilon }<O>_\epsilon =\frac{1}{Z_\epsilon }[d\mathrm{\Phi }]e^{iS_{}(\mathrm{\Phi })}s\left(O(\mathrm{\Phi })\overline{C}\frac{\text{d}}{\text{d}\epsilon }_\epsilon (A)\right)=0.$$
(15)
Unfortunately this proof is purely formal: a rigorous analysis must face the question of infrared divergences which can invalidate the argument. In fact, the perturbative expansion has no meaning at all, if we do not specify carefully as the infrared divergences are managed. The need for an infrared regularization can conflict with BRST-invariance and in general to show that gauge-invariance is recovered when the infrared regulator is removed is non-trivial since the existence of the infrared limit for physical quantities is a very delicate point. To our knowledge this issue has been not enough emphasized in the literature on the Wilson Renormalization Group, in spite of the fact that it is a very crucial one. Actually, even if the recovering of the gauge symmetry at $`\mathrm{\Lambda }=0`$ has been proved for proper vertices at non-exceptional configurations of momenta , there are no theorems guaranteeing the infrared safety of important physical quantities as the interquark potential, which is the first thing one would like to compute in the Wilson renormalization group approach . In particular, as we said in the introduction, there are problems in its perturbative computation. The reason is that the fundamental property of the exponentiation of the Wilson loop at $`T\mathrm{}`$ is lost at $`\mathrm{\Lambda }0`$ since it is related to highly non-trivial gauge-cancellations between different contributions which are lost when BRST-invariance is broken by the infrared cutoff. In other terms we expect a non-commutativity of limits $`T\mathrm{}`$ and $`\mathrm{\Lambda }0`$. This can be seen even at $`O(g^2)`$ in perturbation theory using the non-covariant formalism and it will be discussed at length in section 7. This peculiarity suggests that numerical computations of the interquark potential should be performed with great care.
## 3 The general linear gauge
In general we are interested in linear gauges where $`(A)`$ is a linear function of the gauge field,
$$(A)=L^\mu (p)A_\mu (p),L_\mu (p)=an_\mu +bp_\mu .$$
(16)
Notice that this is a class of interpolating gauges involving both the covariant gauges (if $`a=0`$) and the algebraic non-covariant gauges (if $`b=0`$). Other non-covariant gauges, as for instance the Coulomb one, can be obtained for particular values of $`a`$ and $`b`$. In order to integrate out the auxiliary fields it is convenient to introduce a small convergence factor in the functional integral measure by adding to the BRST action a term
$$_x\frac{1}{2}\xi _2\lambda \lambda ,$$
where $`\xi _2`$ is a positive parameter of mass dimension $`2`$. This terms is a trivial cocycle and therefore does not change the essential properties of the system; in particular by using standard techniques it is possibile to prove that physical quantities are formally independent of $`\xi _2`$. By using the linear equations of motion
$$\lambda ^a=\frac{1}{\xi _2}L^\mu A_\mu ^a$$
(17)
we can eliminate the auxiliary fields $`\lambda ^a`$ and consider the reduced action
$$S_{lin.g.}^{red}=_x\frac{1}{4}F_{\mu \nu }F^{\mu \nu }\overline{C}L^\mu D_\mu C+_x\frac{1}{2\xi _2}L^\mu A_\mu L^\nu A_\nu .$$
(18)
To this reduced three-level action, we add an infrared cutoff as a mass-like term
$$S_{lin.g.}^{red}(A,C,\overline{C},\mathrm{\Lambda })=S_{lin.g.}^{red}(A,C,\overline{C})+_x\frac{1}{2}\mathrm{\Lambda }^2A^2.$$
(19)
Now we can define an invertible massive propagator for the gauge fields. The explicit form of the propagator can be obtained in Euclidean space (Euclidean notations are recalled in appendix B) by inverting the matrix
$$D_{\mathrm{\Lambda },\mu \nu }^1(p_E)=\left(t_{\mu \nu }(p_E)p_E^2+\frac{1}{\xi _2}L_\mu (p_E)L_\nu (p_E)+\mathrm{\Lambda }^2\delta _{\mu \nu }\right),$$
(20)
where $`t_{\mu \nu }(p_E)`$ denotes the transverse projector in Euclidean space
$$t_{\mu \nu }(p_E)=\delta _{\mu \nu }\frac{p_{E,\mu }p_{E,\nu }}{p_E^2}.$$
(21)
In particular we are interested in the $`\xi _20`$ limit, when one eigenvalue of the propagator vanishes, due to the transversality property
$$L^\mu D_{\mathrm{\Lambda },\mu \nu }(p_E)\stackrel{\xi _20}{=}0,\mathrm{\Lambda }$$
(22)
and therefore $`D_{\mathrm{\Lambda },\mu \nu }`$ is not invertible (for any $`\mathrm{\Lambda }`$). The final result is
$$\begin{array}{c}\hfill D_{\mathrm{\Lambda },\mu \nu }(p_E)\stackrel{\xi _20}{=}\frac{1}{p_E^2+\mathrm{\Lambda }^2}\delta _{\mu \nu }\frac{pL(L_\mu p_{E,\nu }+L_\nu p_{E,\mu })}{(p_E^2+\mathrm{\Lambda }^2)((p_EL)^2+L^2\mathrm{\Lambda }^2)}+\\ \hfill \frac{p_{E,\mu }p_{E,\nu }L^2}{(p_E^2+\mathrm{\Lambda }^2)((p_EL)^2+L^2\mathrm{\Lambda }^2)}\frac{\mathrm{\Lambda }^2L_\mu L_\nu }{(p_E^2+\mathrm{\Lambda }^2)((p_EL)^2+L^2\mathrm{\Lambda }^2)}.\end{array}$$
(23)
Notice that in limit $`\xi _20`$ the propagator is invariant up to rescaling of the gauge-fixing $`L(p_E)C(p_E^2)L(p_E)`$. In particular for $`L_\mu n_\mu `$ we recover the massive axial gauge introduced in which satisfies simple Ward identities, whereas for $`L_\mu p_\mu `$ we obtain a massive version of the usual Landau gauge, which does not satisfies linearly broken Ward identities, but instead satisfies a little modification of the usual Slavnov-Taylor identities (this is the Curci-Ferrari model in Landau gauge ). We have also computed the explicit form of the propagator at $`\xi _20`$, but the resulting expression is not particularly illuminating and there is no scope in writing it here.
### 3.1 Planar gauge
The standard massless version of the planar gauge is formally obtained from the generalized axial gauge if we replace the parameter $`\xi _2`$ with the momentum dependent function $`\xi _2(p)=n^2/p^2`$; in this case the BRST action reads
$$S_{planar}^{BRST}=_x\frac{1}{4}F_{\mu \nu }F^{\mu \nu }\overline{C}n^\mu D_\mu C+\lambda n^\mu A_\mu +\frac{1}{2}\lambda \frac{n^2}{^2}\lambda $$
(24)
or, after elimination of the auxiliary fields,
$$S_{planar}^{red}=_x\frac{1}{4}F_{\mu \nu }F^{\mu \nu }\overline{C}n^\mu D_\mu C\frac{1}{2}n^\mu A_\mu ^2n^\nu A_\nu .$$
(25)
The planar gauge is less problematic than the axial gauge, but the price to pay is in a more complicate realization of symmetries: in opposition to the axial gauge case, the action (25) does not satisfies simple Ward-Takahashi identities (the reason being the presence of derivatives in the gauge-fixing term) but instead Slavnov-Taylor-like identities; moreover ghost fields play a non-trivial role even if less crucial than in covariant gauges<sup>5</sup><sup>5</sup>5For instance the planar gauge has the interesting feature that all diagrams involving ghost loops identically vanishes . This property is preserved even in the massive version. Nevertheless, there are non-vanishing vertex corrections including ghosts as external lines. In principle this could be avoided by considering a formulation without ghost fields but then the Ward identities becomes quite cumbersome.. The planar gauge has been studied quite intensively in the literature and has some interest in itself, and also in comparison with the axial gauge and the light-cone gauge in the problem of inconsistencies of perturbation theory. Moreover the computation of Feynman diagrams in planar gauge is similar but simpler than in light-cone gauge. For these reasons we will analyze in detail this gauge choice. We found that the more convenient way to insert the infrared cutoff in order to have a simple propagator is to modify the massless BRST action as
$$\begin{array}{cc}\hfill S_{planar}(\mathrm{\Lambda })=& _x\frac{1}{4}F_{\mu \nu }F^{\mu \nu }\overline{C}n^\mu D_\mu C+\lambda n^\mu A_\mu \hfill \\ & +_x\frac{1}{2}\mathrm{\Lambda }^2A_\mu A^\mu +\frac{1}{2}\lambda \frac{n^2}{^2+\mathrm{\Lambda }^2}\lambda .\hfill \end{array}$$
(26)
This gives as tree level reduced action
$$\begin{array}{cc}\hfill S_{planar}^{red}(\mathrm{\Lambda })=& _x\frac{1}{4}F_{\mu \nu }F^{\mu \nu }+\frac{1}{2}\mathrm{\Lambda }^2AA+\hfill \\ & \frac{1}{2n^2}n_\mu A^\mu (^2+\mathrm{\Lambda }^2)n_\mu A^\mu \overline{C}n^\mu D_\mu C.\hfill \end{array}$$
(27)
Notice that the “mass” $`\mathrm{\Lambda }^2`$ multiplies the term
$$\frac{1}{2}A^\mu \left(g_{\mu \nu }\frac{n_\mu n_\nu }{n^2}\right)A^\nu =\frac{1}{2}A^ig_{ij}A^j,i,j\{0,1,2\}$$
and thus only transverse (with respect to $`n_\mu `$) degrees of freedom are screened. There are other possible ways of introducing the infrared cutoff in the planar gauge, but this is the more interesting one in the sense that one obtains a propagator which is simple and very similar to the propagator of the light-cone gauge. Its explicit form in Minkowsky space is
$$D_{\mathrm{\Lambda },\mu \nu }(p)=\frac{1}{p^2\mathrm{\Lambda }^2+i\epsilon }\left\{g_{\mu \nu }\frac{n_\mu p_\nu +n_\nu p_\mu }{[pn]}+\frac{\mathrm{\Lambda }^2n_\mu n_\nu }{[pn]^2}\right\}.$$
(28)
The most important difference with respect to the massive axial gauge case is the fact that in the massive planar gauge there are spurious divergences at $`pn=0`$ even at $`\mathrm{\Lambda }0`$. Therefore we need an explicit prescription to manage them. The simplest choice which works, at least up to the order $`O(g^4)`$ in the stardard massless case<sup>6</sup><sup>6</sup>6 Actually in higher order computations the CPV prescription could be problematic; see the discussion in section 5.1., is the CPV prescription
$$\frac{1}{[pn]}\underset{\epsilon 0}{lim}\frac{pn}{(pn)^2+\epsilon ^2}.$$
(29)
We will use this prescription even in the massive case. We stress that if we split the propagator in two pieces as
$$D_{\mathrm{\Lambda },\mu \nu }=\overline{D}_{\mathrm{\Lambda },\mu \nu }+\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }$$
(30)
where $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ is the term proportional to $`n_\mu n_\nu /n^2`$,
$$\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }(p)=\frac{\mathrm{\Lambda }^2n_\mu n_\nu }{[pn]^2(p^2\mathrm{\Lambda }^2+i\epsilon )},$$
(31)
we can see that our formulation reduces to the usual planar gauge expression in the $`\mathrm{\Lambda }0`$ limit if the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ term can be neglected. This is obvious in the computation of infrared finite Feynman diagrams, far from being obvious in the computation of infrared divergent Feynman diagrams, and definitely non-trivial for physical quantities sensitive to the infrared, such as the interquark potential.
### 3.2 Light-cone gauge
The light-cone gauge is the most used and the most tested algebraic non-covariant gauge. At the present it passed many serious consistency checks and should be considered at the same level of safety of the covariant gauge, at least for what concerns the perturbative expansion in four dimensional gauge theories. In particular it has been explicitly proved at order $`O(g^4)`$ in perturbation theory that the Wilson loop computed in the light-cone gauge gives the same result of the Wilson loop in covariant gauge, provided that we use the Mandelstam-Leibbrandt (ML) prescription to regularize the spurious singularities .
The Euclidean massive light-cone propagator can be formally extracted from espression (23) for light-like vectors $`L_\mu =n_{E,\mu }`$ such as $`n_E^2=0`$:
$$D_{E,\mu \nu }(p_E)=\frac{1}{p_E^2+\mathrm{\Lambda }^2}\left\{\delta _{\mu \nu }\frac{n_{E,\mu }p_{E,\nu }+n_{E,\nu }p_{E,\mu }}{p_En_E}\frac{\mathrm{\Lambda }^2n_{E,\mu }n_{E,\nu }}{(p_En_E)^2}\right\}.$$
(32)
The light-cone condition can be realized in Euclidean space if the gauge vector $`n_E=(n_1,n_2,n_3,n_4)`$ has the form $`n_E=(0,0,1,i)`$ or $`n_E^{}=(0,0,1,i)`$, or an equivalent one after a $`SO(4)`$ rotation. Writing
$$p_3=p_{}\mathrm{cos}\theta _{},p_4=p_{}\mathrm{sin}\theta _{},p_{}^2=p_En_Ep_En_E^{}=p_3^2+p_4^2,$$
(33)
we obtain the explicit expressions
$$\frac{1}{p_En_E}=\frac{1}{ip_4+p_3}=\frac{ip_4+p_3}{p_4^2+p_3^2}=\frac{p_En_E^{}}{p_{}^2}=\frac{e^{i\theta _{}}}{p_{}}.$$
(34)
After Wick rotation in Minkowsky space $`p_0=ip_4`$ we see that this approach corresponds to regularize the spurious poles with the ML prescription. The Minkowskian propagator reads
$$D_{\mathrm{\Lambda },\mu \nu }(p)=\frac{1}{p^2\mathrm{\Lambda }^2+i\epsilon }\left\{g_{\mu \nu }\frac{n_\mu p_\nu +n_\nu p_\mu }{[[pn]]}+\frac{\mathrm{\Lambda }^2n_\mu n_\nu }{[[pn]]^2}\right\},$$
(35)
with
$$\frac{1}{[[pn]]}\frac{pn^{}}{(pn)(pn^{})+i\epsilon },n=(1,0,0,1),n^{}=(1,0,0,1)$$
(36)
and reduces to the standard one at $`\mathrm{\Lambda }0`$ if the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ term can be neglected. Differently from the planar gauge, the light-cone choice is extremely convenient since not only it avoids the double pole problem but still mantains the advantages of the axial gauge. i.e. the transversality property which garantees the full decoupling of ghost fields,
$$n^\mu D_{\mu \nu ,\mathrm{\Lambda }}(q)=0.$$
(37)
Furthermore the simple Ward identities
$$q^\mu D_{\mathrm{\Lambda },\mu \nu }(q)=\frac{n_\nu }{[[nq]]}$$
(38)
and
$$p^\mu D_\mathrm{\Lambda }^{\lambda \nu }(q)V_{\nu \mu \rho }(q,p,qp)D_\mathrm{\Lambda }^{\rho \tau }(q+p)=D_\mathrm{\Lambda }^{\lambda \tau }(q)D_\mathrm{\Lambda }^{\lambda \tau }(q+p)$$
(39)
hold, where
$$V_{\mu \nu \rho }(p,q,r)=(qr)_\mu g_{\nu \rho }+(pq)_\rho g_{\mu \nu }+(rp)_\nu g_{\rho \mu }$$
(40)
is the non-Abelian three-gluon vertex. By using these ingredients we explicitly checked the transversality of the gluon propagator and in general the Ward identities on the proper vertices with direct diagrammatic considerations, for any value of $`\mathrm{\Lambda }`$. We repeat that in planar gauge, on the contrary, the Ward identities are much more complicated and the gluon propagator is not transverse with respect to $`n_\mu `$ or $`p_\mu `$.
It is important to stress that: i) it is impossible to impose condition (39) in covariant gauges: this is the reason why the gauge symmetry is unavoidably more complicate in the covariant case and ghost fields have to be taken into account ii) differently from planar gauge, we checked that there are no other possibile forms of the light-cone propagator consistent with (37),(38) and (39): the transversality constraint plus the symmetry requirement fixes univocally the form (35) for the light-cone propagator. In other words, equation (35) is the unique way of introducing a Wilsonian infrared cutoff in a non-Abelian gauge theory consistently with Ward identities.
## 4 Problems of the pure axial gauge
As we recalled in the introduction, the standard approach to axial gauge with the CPV prescription has various problems and in particular the test of exponentiation for a properly chosen Wilson loop fails. There has been a lot of work in the literature to solve this problem (see for instance the citation list in ), but in spite of this effort in our opinion at present there are no completely satisfactory solutions. For instance there is an apparently simple solution consisting in changing the prescriptions on the gluon propagator and the ghost interaction in such a way that the Wilson loop becomes the same as computed in covariant gauges. This is the logic of Cheng and Tsai and also of the approches based on interpolating gauges, in which one try to define the axial gauge as a deformation of a more regular gauge. This kind of approaches has a long hystory (starting from the old work of until very recent papers ) nevertheless they are not completely satisfactory since i) the transversality property $`n^\mu D_{\mu \nu }=0`$ is lost; ii) the ghost fields are no longer decoupled; iii) the Ward identities have no more a simple form; iv) a careful study of the infrared singularities appearing in the limit in which the modified gauge tends to the pure axial gauge is needed. For such reasons these modifications are so drastic that in fact they should be interpreted as a switch to a truly different gauge . The situation in the Wilsonian approach is different but still there are problems which will be discussed in the following sections. Our final conclusion will be that the axial gauge is definitively sick, at least in perturbation theory, and alternative gauge choices should be considered.
### 4.1 The CPV regularization of the double pole
In the Cauchy Principal Value prescription, the double pole is defined as the derivative of the single pole,
$$\frac{1}{[p_3^2]}=\frac{}{p_3}\frac{1}{[p_3]}$$
(41)
where $`1/[p_3]`$ is defined as in (29). This means that the double pole is regularized as
$$\frac{1}{[p_3^2]}=\underset{\mathrm{\Lambda }0}{lim}\frac{p_3^2\mathrm{\Lambda }^2}{(p_3^2+\mathrm{\Lambda }^2)^2}.$$
(42)
We stress two things:
1. One could expect some problem with this regularization of the double pole, since we loose the property of the Euclidean propagator of being positive definite. This can be seen by looking at the eigenvalues of the Euclidean pure axial gauge propagator (at zero mass and with the CPV prescription), which are
$$0,\frac{1}{p_E^2},\frac{1}{p_E^2},\frac{1}{[p_3]^2}.$$
(43)
The first three eigenvalues are obviously non negative; the problem is with the last one, which is not positive definite. This means that the integral
$$I_3(f)=_{p_3}\frac{1}{[p_3^2]}f(p_3)<0$$
(44)
can be negative even if $`f(p_3)`$ is a regular positive function. For instance it is trivial to check that this happens for $`f(p_3)=1/(p_3^2+m^2)`$. Therefore $`D_{E,\mu \nu }(p_E)`$ is no more positive definite and clearly this is a potential source of problems for the perturbative expansion.
2. The Wilsonian prescription of the double pole
$$\frac{1}{[p_3^2]_W}=\underset{\mathrm{\Lambda }0}{lim}\frac{1}{p_3^2+\mathrm{\Lambda }^2}$$
(45)
is different from the CPV prescription (42). As a matter of fact the eigenvectors of the Euclidean propagator are explicitly non-negative
$$0,\frac{1}{p_E^2+\mathrm{\Lambda }^2},\frac{1}{p_E^2+\mathrm{\Lambda }^2},\frac{1}{p_3^2+\mathrm{\Lambda }^2}$$
(46)
and therefore one could argue that the Wilsonian prescription is better then the CPV. Unfortunately, the Wilsonian prescription has other problems that we will describe in the next section.
### 4.2 The double pole problem in the Wilsonian formulation
We found two major infrared problems in the Wilsonian formulation of the axial gauge: i) the Fourier transform of the propagator is divergent at $`\mathrm{\Lambda }0`$; ii) one loop Feynman diagrams which are infrared finite in covariant gauges becomes divergent in axial gauge for all configurations of momenta. In principle, this is not enough to prove the inconsistency of this gauge, since we should prove that at least one physical quantity is ill defined in the $`\mathrm{\Lambda }0`$ limit. Indeed one could think that due to miracolous cancellations related to the Ward-identities (which are respected) physical quantities are indeed finite in the $`\mathrm{\Lambda }0`$ limit as it happens for instance for the Wilson loop at order $`O(g^2)`$ (see section 7). However this is not garanteed in general. Moreover, even if this were true, the singularity of the $`\mathrm{\Lambda }0`$ limit would forbid in practice any numerical application, since even extremely small numerical breaking of the Ward-identities would result in enormous (at the limit infinite) differences in the final result at $`\mathrm{\Lambda }0`$. This is a serious drawback of the axial gauge Wilsonian formulation. We stress that, even if not noticed in those references, the problems we discuss here also apply to the formulations in where they are probably even worse due to the wild breaking of gauge-symmetry. They were not recognized before simply because they do not affect the computation of the one-loop beta function, which is an universal quantity not affected by the way the infrared cutoff in inserted and insensitive to the gauge fixing choice.
The origin of all problems comes from the part of the propagator proportional to $`p_\mu p_\nu `$,
$$D_{\mathrm{\Lambda },\mu \nu }^{pp}=\frac{p_\mu p_\nu n^2}{(p^2+\mathrm{\Lambda }^2)((pn)^2+n^2\mathrm{\Lambda }^2)}.$$
(47)
This quantity is a messy source of infrared divergencies due to the identity (in the sense of distributions)
$$\frac{1}{(pn)^2+\mathrm{\Lambda }^2}\stackrel{\mathrm{\Lambda }0}{=}\frac{\pi }{\mathrm{\Lambda }}\delta (pn)$$
(48)
This means that for any regular function $`f(p_3,\mathrm{\Lambda })`$ such as $`f(0,0)0`$ we have
$$\underset{\mathrm{\Lambda }0}{lim}_{p_3}\frac{f(p_3,\mathrm{\Lambda })}{p_3^2+\mathrm{\Lambda }^2}=\underset{\mathrm{\Lambda }0}{lim}\frac{f(0,0)}{2\mathrm{\Lambda }}=\mathrm{}.$$
(49)
We will see in next subsections the disastrous consequences of this fact on the computation of various quantities.
### 4.3 The $`x`$space propagator.
Here we show that even if the $`x`$space propagator
$$D_{\mathrm{\Lambda },\mu \nu }(x)=_pe^{ipx}D_{\mathrm{\Lambda },\mu \nu }(p)$$
(50)
is perfectly defined for any $`\mathrm{\Lambda }0`$, the infrared limit $`\mathrm{\Lambda }0`$ does not exist due to a strong infrared divergence coming from the double pole part of the propagator proportional to $`p_\mu p_\nu `$ (the double pole part proportional to $`\mathrm{\Lambda }^2n_\mu n_\nu `$ and the single pole part are regular as $`\mathrm{\Lambda }0`$). This can be proved with a direct computation by using the result (49). One obtains $`D_{\mathrm{\Lambda },i3}=D_{\mathrm{\Lambda },3i}=D_{\mathrm{\Lambda },33}=0`$ and
$$D_{\mathrm{\Lambda },ij}(x)\stackrel{\mathrm{\Lambda }0}{=}\frac{1}{2\mathrm{\Lambda }}_{\overline{p}}\frac{p_ip_j}{\overline{p}^2}e^{i\overline{p}\overline{x}}=\frac{1}{2\mathrm{\Lambda }}\frac{1}{3}\delta _{ij}\delta ^{(3)}(\overline{x})\stackrel{\mathrm{\Lambda }0}{=}\mathrm{}.$$
(51)
Therefore the Fourier transform of the propagator does not exist at $`\mathrm{\Lambda }0`$.
The level of danger of this problem is unclear at this level, since it could not affect physical quantities. We will see in section 7 that the simplest physical quantity we can compute, i.e. the Wilson loop at order $`O(g^2)`$, only depends on the Fourier transform of the transverse part of the propagator,
$$D_{\mathrm{\Lambda },\mu \nu }^T(x)_pe^{ipx}t_{\mu \lambda }(p)D_\mathrm{\Lambda }^{\lambda \rho }(p)t_{\rho \nu }(p),$$
(52)
which is safe in the limit $`\mathrm{\Lambda }0`$. To show this point we have to prove that the $`\stackrel{~}{D}_{\mu \nu ,\mathrm{\Lambda }}(x)`$ part of the propagator,
$$\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }(x)=D_\mathrm{\Lambda }^{nn}(x)\frac{n_\mu n_\nu }{n^2}$$
(53)
which is the only one which transverse part depends on the gauge vector $`n_\mu `$ is vanishing at $`\mathrm{\Lambda }0`$. A direct computation in Euclidean space gives
$$D_\mathrm{\Lambda }^{nn}(x_E)=_{p_E}\frac{\mathrm{exp}(ip_Ex_E)}{p_E^2+\mathrm{\Lambda }^2}\left[\frac{\mathrm{\Lambda }^2}{p_3^2+\mathrm{\Lambda }^2}\right].$$
(54)
By direct inspection using (48) one sees that this term is linearly vanishing with $`\mathrm{\Lambda }`$; therefore the infrared limit of the transverse part of the propagator exists finite and is independent of the gauge vector $`n_\mu `$:
$$D_{\mathrm{\Lambda },\mu \nu }^T(x_E)\stackrel{\mathrm{\Lambda }0}{=}\left(\delta _{\mu \nu }\frac{_{E,\mu }_{E,\nu }}{_E^2}\right)\frac{1}{4\pi ^2x_E^2}.$$
(55)
The crucial point here is the fact that the singular $`D_{\mathrm{\Lambda },\mu \nu }^{pp}`$ term (47) does not contributes to the transverse part of the propagator.
### 4.4 Computation of a one-loop integral
The simplest perturbative quantity sensitive to the double pole problem is the one-loop quark self-energy. In this section we will see that the $`p_\mu p_\nu `$ part of the propagator gives an infinite contribution to this quantity in the $`\mathrm{\Lambda }0`$ limit. Unfortunately this is a gauge-dependent quantity therefore in principle we cannot positively conclude about the inconsistency of the axial gauge choice with the Wilsonian prescription (45). However in practice this is a serious drawback since it forbids the direct computation of any perturbative quantity except the one-loop anomalous dimensions and the beta function .
In order to simplify the computation, by avoiding the inessential complications with the gamma matrices, we consider the self-energy of a scalar quark. Moreover, in order to avoid infrared divergences we take a quark mass $`m0`$ and in order to avoid ultraviolet divergences we derive twice with respect to $`m^2`$. In this way we obtain a quantity corresponding to the four-point vertex with two insertions at zero momentum shown in figure 1.
This quantity, which we denote by
$$F(\overline{p}^2,p_3^2,\mathrm{\Lambda })=_{q_E}\left(\frac{}{m^2}\right)^2\left[\frac{(2p_E+q_E)^\mu (2p_E+q_E)^\nu }{[(q_E+p_E)^2+m^2+\mathrm{\Lambda }^2]}D_{\mu \nu }(q_E,\mathrm{\Lambda })\right]$$
(56)
is ultraviolet convergent and (in covariant gauges) infrared finite even at $`\mathrm{\Lambda }=0`$ if $`p_E^2m^2`$. Still, the $`\mathrm{\Lambda }0`$ limit is singular in the massive axial gauge since the dominant contribution comes from the double pole part of the propagator which, due to equation (49), is infinity in the $`\mathrm{\Lambda }0`$ limit:
$$F(\overline{p}^2,p_3^2,\mathrm{\Lambda })\stackrel{\mathrm{\Lambda }0}{=}\frac{1}{2\mathrm{\Lambda }}_\stackrel{}{q}\frac{2}{[(\overline{q}+\overline{p})^2+m^2]^3}q_iq_j(2p+q)^i(2p+q)^j\mathrm{}.$$
(57)
Notice that this problem appears even in the Abelian theory. Moreover it is clear that the problem becomes worse in the non-Abelian case and at higher orders in perturbation theory. Still, in principle it is possible that cancellations allows to define unambigously physical quantities even in the $`\mathrm{\Lambda }0`$ limit, since the various Feynman diagrams should add in such a way that the final result at $`\mathrm{\Lambda }0`$ is finite and independent of the gauge fixing, i.e. the same that in covariant gauges. However, this is an highly non-trivial point and cannot be taken for granted at finite order in perturbation theory.
## 5 The planar gauge
For the reasons described above, we think that it is not convenient to insist on the axial gauge and from now on we switch to other gauge choices. The simplest examples where the difficulties of the axial gauge can be circumvented (at least in the standard massless formalism and at order $`O(g^4)`$ in perturbation theory) are the planar gauge and the light-cone gauge. We consider first the planar gauge.
### 5.1 Remarks on the planar gauge
Properly speaking, even the standard massless planar gauge choice with the CPV prescription is not completely controled and the consistency of the perturbative expansion in this gauge is an open problem; nevertheless it will be studied in this section for sake of comparison with the axial gauge and the light-cone gauge choices.
The reason of the difficulties can be traced back to the well known splitting formula
$$\frac{1}{[nk]}\frac{1}{[n(pk)]}=\frac{1}{np}\left(\frac{1}{[nk]}+\frac{1}{[n(pk)]}\right)\pi ^2\delta (nk)\delta (np).$$
(58)
Due to the presence of the delta function terms, in two-loops or higher orders computations involving gluon vertices there appear ill-defined objects of kind
$$\frac{1}{[kn]}\delta (kn)\frac{kn}{(kn)^2+\epsilon ^2}\frac{\epsilon /\pi }{(kn)^2+\epsilon ^2}$$
(59)
which have to be studied in detail. It is known that in the massless computation of the $`O(g^4)`$ Wilson loop there is a precise way to manage these terms and the final result is consistent with the covariant gauge . Nevertheless, the situation at order $`O(g^6)`$ is unknown and in general the definition of two-loops Feynman diagrams is problematic. This is the reason why we will switch to the light-cone gauge in the next section. However it is interesting to see that even at the level of the planar gauge the divergences found before in one-loop computations are absent, essentially because the double pole term proportional to $`p_\mu p_\nu `$ in (47) is absent. There is however a double pole in the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ part of the propagator, but since this term is multiplied by a $`\mathrm{\Lambda }^2`$ factor it is espected to vanish in the $`\mathrm{\Lambda }0`$ limit. This is indeed the case for infrared safe quantities and it will be explicitly verified in the examples we consider here. We stress that this is by no means a trivial statement, since in the computation of particular infrared divergent quantities the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ term in the denominator could be not subleading with respect to the other terms and actually could also give the dominant contribution. We will see in section 6 that this is indeed the case in the light-cone gauge. Moreover we will see in section 7 that the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ term contributes to the Wilson loop and is suppressed only if the $`\mathrm{\Lambda }0`$ limit is done properly, i.e. at finite $`T`$.
### 5.2 The $`x`$space propagator
We prove here that the Fourier transform of the propagator is well defined and in the $`\mathrm{\Lambda }0`$ limit reduces to the Fourier transform of the standard planar gauge propagator. This is obvious for the $`\overline{D}_{\mathrm{\Lambda },\mu \nu }`$ part of the propagator, therefore we have simply to prove that the Fourier transform of the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ term is vanishing at $`\mathrm{\Lambda }0`$. A direct computation gives
$$\underset{\mathrm{\Lambda }0}{lim}\mathrm{\Lambda }^2_{\overline{p},p_3}\frac{\mathrm{exp}(ip_3x_3+i\overline{p}\overline{x})}{[p_3]^2(p_3^2+\overline{p}^2+\mathrm{\Lambda }^2)}=\underset{\mathrm{\Lambda }0}{lim}\frac{\mathrm{\Lambda }^2}{8\pi ^2}\mathrm{ln}(\mathrm{\Lambda }^2\overline{x}^2)=0.$$
(60)
This result can be obtained by first doing the $`p_3`$ integral using $`1/[p_3^2]=[p_3]^1_{p_3}`$; the successive three-dimensional integral generates an infrared logarithimic singularity at $`\mathrm{\Lambda }0`$ which is killed by the $`\mathrm{\Lambda }^2`$ prefactor. Therefore we have solved the problem with the Fourier transform. However we stress again that in the computation of infrared sensitive quantities this term in general could gives an essential contribution to the final result. Furthermore, we will see in section 7 that this term is suppressed in the Wilson loop computation only if the $`\mathrm{\Lambda }0`$ and $`T\mathrm{}`$ limit are done in the correct order. Therefore the relation between the massive version of the planar gauge in the $`\mathrm{\Lambda }0`$ limit and the standard planar gauge in general is delicate.
### 5.3 Computation of a one-loop finite quantity
Now we can study what happens in the computation of the quark self-energy with two zero-momentum insertions, i.e. the analogous of the quantity $`F`$ defined in (56). If we split the Euclidean planar gauge propagator in three terms (in this section for notational simplicity we write $`p`$ for $`p_E`$)
$$\begin{array}{cc}\hfill D_{\mu \nu }(p,\mathrm{\Lambda })& =\frac{1}{p^2+\mathrm{\Lambda }^2}\left\{\delta _{\mu \nu }\frac{n_\mu p_\nu +n_\nu p_\mu }{[pn]}\frac{\mathrm{\Lambda }^2n_\mu n_\nu }{[pn]^2}\right\}\hfill \\ & =D_{\mu \nu }^{(a)}(p,\mathrm{\Lambda })+D_{\mu \nu }^{(b)}(p,\mathrm{\Lambda })+D_{\mu \nu }^{(c)}(p,\mathrm{\Lambda })\hfill \end{array}$$
(61)
we can split the computation in three different contributions:
$$F(\overline{p}^2,p_3^2,\mathrm{\Lambda })=F^{(a)}(p^2,\mathrm{\Lambda })+F^{(b)}(\overline{p}^2,p_3^2,\mathrm{\Lambda })+F^{(c)}(\overline{p}^2,p_3^2,\mathrm{\Lambda })$$
(62)
with
$$F^{(a)}(p^2,\mathrm{\Lambda })=_q\left(\frac{}{m^2}\right)^2\frac{(2p+q)^2}{(q^2+\mathrm{\Lambda }^2)((q+p)^2+m^2+\mathrm{\Lambda }^2)},$$
(63)
$$F^{(b)}(\overline{p}^2,p_3^2,\mathrm{\Lambda })=_q\left(\frac{}{m^2}\right)^2\frac{2(2p+q)n(2p+q)q}{[qn](q^2+\mathrm{\Lambda }^2)((q+p)^2+m^2+\mathrm{\Lambda }^2)},$$
(64)
$$F^{(c)}(\overline{p}^2,p_3^2,\mathrm{\Lambda })=_q\left(\frac{}{m^2}\right)^2\frac{\mathrm{\Lambda }^2(2pn+qn)^2}{[qn]^2(q^2+\mathrm{\Lambda }^2)((q+p)^2+m^2+\mathrm{\Lambda }^2)}.$$
(65)
The first contribution is the same as in Feynman gauge and can be computed using the standard Feynman parametrization. It gives
$$F^{(a)}(p^2,\mathrm{\Lambda })=\frac{1}{16\pi ^2}_0^1𝑑x(1x)^2\frac{p^2(1+4xx^2)+2[m^2(1x)+\mathrm{\Lambda }^2]}{[(p^2x+m^2)(1x)+\mathrm{\Lambda }^2]^2}$$
(66)
where the $`x`$integral can be performed in terms of elementary functions (logarithms). In particular in the on-shell limit $`p^2m^2,\mathrm{\Lambda }0`$ the $`x1`$ integration region dominates and we have
$$F^{(a)}(p^2,\mathrm{\Lambda }=0)=\frac{1}{4\pi ^2}\frac{1}{p^2+m^2}+O((p^2+m^2)^0)$$
(67)
if the $`\mathrm{\Lambda }0`$ limit is taken first. On the contrary, if the $`p^2m^2`$ limit is taken first, we have
$$F^{(a)}(p^2=m^2,\mathrm{\Lambda })=\frac{1}{16\pi }\frac{1}{m\mathrm{\Lambda }}+O\left(\frac{\mathrm{\Lambda }^0}{m^2}\right).$$
(68)
This is an explicit confirmation of the fact the the order of limits is a delicate question for infrared divergent quantities.
The second contribution is typical of planar gauge. Its general expression is rather complicate, but there is a simplification if we restrict the analysis to the limits $`|\overline{p}^2|p_3^2`$ or $`|\overline{p}^2|p_3^2`$. The first limit $`|\overline{p}^2|p_3^2`$ is trivial in the sense that the integral can be recast in a covariant-like form and its computation is no more difficult than the computation of the covariant term and gives a completely analogous contribution. In the second limit $`|\overline{p}^2|p_3^2`$ instead the peculiarities of the planar gauge are evident. In particular the quantity
$$F^{(b)}(p_3^2,\mathrm{\Lambda })=\underset{\overline{p}0}{lim}F^{(b)}(\overline{p}^2,p_3^2)$$
(69)
can be computed with the method of double Feynman parametrization discussed in appendix B. After some manipulation it reads
$$F^{(b)}(p_3^2,\mathrm{\Lambda })=_{q_3,\stackrel{~}{q}}_0^1𝑑x_{1x}^1𝑑z\left(\frac{}{m^2}\right)^2\frac{N(q_{}^2,\stackrel{~}{q}^2,p_3^2,x,z)}{D(q_{}^2,\stackrel{~}{q}^2,p_3^2,m^2,\mathrm{\Lambda }^2,x,z)}$$
(70)
with
$$N=2[p_3^4(1x^2)^2+p_3^2[2(3x^21)q_3^2\stackrel{~}{q}^2(1x^2)/z]+q_3^2[q_3^2+\stackrel{~}{q}^2/z]]$$
(71)
and
$$D=z^{3/2}[q_3^2+\stackrel{~}{q}^2+\mathrm{\Lambda }^2z+(p_3^2x+m^2)(1x)]^3.$$
(72)
Now the integrations in $`q_3,\stackrel{~}{q}`$ and $`z`$ are straightforward (see appendix B) and all the complexity is confined into the $`x`$integration. In the $`\mathrm{\Lambda }0`$ limit there is a strong simplification and it can be exactly performed in terms of elementary functions. The final result is
$$\begin{array}{cc}\hfill F^{(b)}(p_3^2,\mathrm{\Lambda }=0)& =\frac{1}{4\pi ^2p_3^2}\frac{p_3^2/m^21}{8\pi ^2(p_3^2+m^2)}\hfill \\ & \frac{\sqrt{p_3^2}\mathrm{ln}[(\sqrt{p_3^2+m^2}\sqrt{p_3^2})/(\sqrt{p_3^2}+\sqrt{p_3^2+m^2})]}{8\pi ^2(p_3^2+m^2)^{3/2}}.\hfill \end{array}$$
(73)
The third contribution is characteristic of the Wilsonian formulation and must vanish in the $`\mathrm{\Lambda }0`$ limit in order to recover the results of the standard massless planar gauge:
$$\underset{\mathrm{\Lambda }0}{lim}F^{(c)}(p_E)=0.$$
(74)
This is immediate to see in the limit $`|\overline{p}^2|p_3^2`$. In the opposite limit $`|\overline{p}^2|p_3^2`$ one could expect divergences at $`\overline{p}^2=m^2`$; but it is immediate to see that
$$F^{(c)}(\overline{p}^2=m^2,\mathrm{\Lambda })\frac{\mathrm{\Lambda }^2}{m^2}F^{(a)}(\overline{p}^2=m^2,\mathrm{\Lambda })$$
(75)
therefore $`F^{(c)}(\overline{p}^2=m^2,\mathrm{\Lambda })`$ is linearly vanishing at $`\mathrm{\Lambda }0`$. Thus the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ term can always be neglected, at least in this kind of one-loop computations. We will see that this is not always the case for the light-cone gauge.
## 6 Safeness of the light-cone gauge
For the point of view of the consistency of the perturbative expansion the light-cone gauge is expected to be the safest choice. There are various reasons for this expectation, which we will discuss now.
The first reason is the fact that, contrary to the planar gauge, the Euclidean propagator is (semi) positive definite. This can be immediately proved by solving the eigenvalue equation
$$det\left(\lambda \delta _\nu ^\mu D_\nu ^\mu (p_E,n_E;\mathrm{\Lambda })\right)=0$$
(76)
which after explicit computation reads
$$\lambda ^2\left(\lambda \frac{1}{p_E^2+\mathrm{\Lambda }^2}\right)^2=0$$
(77)
with solutions
$$\lambda _1=\lambda _2=\frac{1}{p_E^2+\mathrm{\Lambda }^2},\lambda _3=\lambda _4=0.$$
(78)
From this computation one sees that the eigenvalues $`\lambda _i=\lambda _i(p_E,n_E)`$ are non-negative. In addition, they depend only on the Lorentz-invariant combination $`p_E^2`$ and not on $`n_E`$. We emphasize that this is not true in non-covariant gauges others than the light-cone one. Therefore one expects a somehow less prononced Lorentz breaking in the light-cone gauge.
The second reason is that within the ML prescription the splitting formula (58) holds without the delta function term: as a consequence there are no ill defined terms in higher orders of perturbation theory as it happens with the CPV prescription in axial and planar gauges.
The third reason is that the massless limit of the light-cone gauge is expected to be less singular than in others gauges. This expectation comes from the that the Fourier transform of the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ term is quadratically vanishing as $`\mathrm{\Lambda }^20`$: this should be contrasted with the axial gauge case where the suppression of the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ term in only linear in $`\mathrm{\Lambda }`$ (see the discussion after Eq. (54)) and the planar gauge where this term vanishes as $`\mathrm{\Lambda }^2`$ times a logarithm. We will discuss in detail this latter point just below.
### 6.1 The $`x`$space propagator
In order to explicitly compute the Fourier transform of the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }(x)`$ term we have to study the Euclidean integral
$$D_\mathrm{\Lambda }^{nn}(x_E)=\mathrm{\Lambda }^2_{q_E}\frac{e^{iq_Ex_E}}{(q_En_E)^2(q_E^2+\mathrm{\Lambda }^2)}.$$
(79)
We proceed as follow. First, we introduce the angles $`\theta `$ and $`\theta ^{}`$ such as
$$q_1=q_{}\mathrm{cos}\theta ^{},q_2=q_{}\mathrm{sin}\theta ^{},q_3=q_{}\mathrm{sin}\theta ,q_4=q_{}\mathrm{cos}\theta $$
(80)
and we take $`x=(x_{},0,x_{},0)`$ (this is not restrictive); then the Fourier transform can be written
$$D_\mathrm{\Lambda }^{nn}(x_E)=\mathrm{\Lambda }^2\frac{q_{}dq_{}d\theta ^{}}{(2\pi )^2}\frac{dq_{}d\theta }{(2\pi )^2}\frac{e^{2i\theta iq_{}x_{}\mathrm{sin}\theta iq_{}x_{}\mathrm{sin}\theta ^{}}}{q_{}(q_{}^2+q_{}^2+\mathrm{\Lambda }^2)}.$$
(81)
$`D_\mathrm{\Lambda }^{nn}(x_E)`$ can be computed by first performing the angular integrations and then the momentum integrations in $`q_{}`$ and $`q_{}`$. The angular integrations can be done by using the following representation of Bessel functions
$$J_n(z)=\frac{1}{2\pi }_0^{2\pi }𝑑\theta e^{in\theta iz\mathrm{sin}\theta }$$
(82)
for $`n=0`$ and $`n=2`$ respectively. In this way we obtain
$$D_\mathrm{\Lambda }^{nn}(x_E)=\frac{\mathrm{\Lambda }^2}{(2\pi )^2}_0^{\mathrm{}}𝑑q_{}q_{}_0^{\mathrm{}}\frac{dq_{}}{q_{}}\frac{J_2(q_{}x_{})J_0(q_{}x_{})}{q_{}^2+q_{}^2+\mathrm{\Lambda }^2}.$$
(83)
Since at small $`z`$ the Bessel function $`J_2(z)\frac{1}{8}z^2`$ is quadratically vanishing whereas at large $`z`$ is exponentially suppressed, we see that the $`q_{}`$integral is infrared and ultraviolet finite. It can be exactly computed with the result
$$D_\mathrm{\Lambda }^{nn}(x_E)=\frac{\mathrm{\Lambda }^2}{8\pi ^2}_0^{\mathrm{}}𝑑q_{}q_{}\frac{\omega _{}^24x_{}^2+2\omega _{}^2K_2(\omega _{}x_{})}{\omega _{}^4}J_0(q_{}x_{}),$$
(84)
with $`\omega _{}^2q_{}^2+\mathrm{\Lambda }^2`$. Using the series expansion of the Bessel functions
$$K_2(z)=\frac{4z^2}{2z^2}+O(z^2\mathrm{log}z^2),J_0(z)=1\frac{1}{4}z^2+O(z^4),$$
one sees that the $`q_{}`$integral is also finite, even in the $`\mathrm{\Lambda }0`$ limit where it can be computed exactly and it gives
$$D_\mathrm{\Lambda }^{nn}(x_E)=\frac{\mathrm{\Lambda }^2}{16\pi ^2}\left[\frac{x_E^2}{x_{}^2}\mathrm{log}\frac{x_E^2}{x_{}^2}1+O(\mathrm{\Lambda }^2x_{}^2,\mathrm{\Lambda }^2x_{}^2)\right]$$
(85)
where $`x_E^2=x_{}^2+x_{}^2`$. Thus, $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }(x)`$ is quadratically vanishing in the $`\mathrm{\Lambda }0`$ limit. We also see that there could be problems in the limit $`x_{}/x_{}\mathrm{}`$ which is relevant for the Wilson loop computation. This point will be discussed in detail in section 7.
### 6.2 The structure of (soft) infrared divergences
It is possible to repeat the computation of section 5.3 in the light-cone gauge. The covariant contribution $`F^{(a)}(p,\mathrm{\Lambda })`$ is obviously the same, whereas the contributions $`F^{(b)}(p,\mathrm{\Lambda })`$ and $`F^{(c)}(p,\mathrm{\Lambda })`$ are different from the analogous one in planar gauge. Still, one can use the double Feynman parametrization method described in appendix B. In the limit $`p_{}^2/p_{}^20`$ the contribution $`F^{(b)}`$ can be rewritten in the form
$$F^{(b)}(p_{}^2,\mathrm{\Lambda })=_{q_{},\stackrel{~}{q}_{}}_0^1𝑑x_{1x}^1𝑑z\left(\frac{}{m^2}\right)^2\frac{N(q_{}^2,\stackrel{~}{q}_{}^2,p_{}^2,x,z)}{D(q_{}^2,\stackrel{~}{q}_{}^2,p_{}^2,m^2,\mathrm{\Lambda }^2,x,z)}$$
(86)
with
$$N(q_{}^2,\stackrel{~}{q}_{}^2,p_{}^2,x,z)=8q_{}^2p_{}^2+\mathrm{\Delta }N(q_{}^2,\stackrel{~}{q}_{}^2,p_{}^2,x,z)$$
(87)
and
$$D=z[q_{}^2+\stackrel{~}{q}_{}^2+\mathrm{\Lambda }^2z+(p_{}^2x+m^2)(1x)]^3.$$
(88)
Actually, we exactly computed the expression $`\mathrm{\Delta }N(q_{}^2,\stackrel{~}{q}_{}^2,p_{}^2,x,z)`$ in the numerator and it is possible to perform exactly the integrals in the general case, but the resulting expressions are lenghty and they will not be reported here. In order to analyze the on-shell divergences at $`p^2m^2`$ the only relevant term is the fist one displayed in the right hand side of (87). The integrations in $`q_{}`$ and $`q_{}`$ are straighforward and they give, after neglecting the $`\mathrm{\Delta }N`$ term,
$$F^{(b)}(p_{}^2,\mathrm{\Lambda })\frac{1}{4\pi ^2}_0^1𝑑x_{1x}^1\frac{dz}{z}\frac{p_{}^2(1x)^2}{[\mathrm{\Lambda }^2z+(p_{}^2x+m^2)(1x)]^2}.$$
(89)
If we take the $`\mathrm{\Lambda }0`$ limit first we obtain
$$F^{(b)}(p_{}^2,\mathrm{\Lambda }=0)\stackrel{p^2m^2}{=}\frac{1}{4\pi ^2}\frac{\mathrm{ln}[(p_{}^2+m^2)/p_{}^2]}{p_{}^2+m^2}+O\left(\frac{1}{p_{}^2+m^2}\right)$$
(90)
and this is dominant with respect to the divergence $`1/(p^2+m^2)`$ coming from the covariant contribution $`F^{(a)}`$. If we take the $`p^2m^2`$ limit first we obtain
$$F^{(b)}(p_{}^2=m^2,\mathrm{\Lambda })=\frac{1}{8\pi ^2}\frac{1}{\mathrm{\Lambda }^2}+O\left(\frac{\mathrm{\Lambda }^0}{m^2}\right),$$
(91)
which is still dominant with respect to the linear divergence coming from the covariant contribution $`F^{(a)}`$.
Now we can study in an analogous way the contribution of the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ part of the propagator, i.e. the $`F^{(c)}`$ term. It is obvious that this term is quadratically vanishing at $`\mathrm{\Lambda }0`$ if $`p^2m^2`$. Nevertheless, if $`p^2=m^2`$, this term could be divergent as $`\mathrm{\Lambda }0`$. This is the case indeed. Actually it is easy to understand that there is no divergence in the limit $`|p_{}^2|p_{}^2`$, using the same argument as in equation (75). However in the opposite limit $`|p_{}^2|p_{}^2`$ the infrared divergence is enhanced and actually dominates with respect to the covariant contribution. To see this point we write down the explicit expression of $`F^{(c)}`$ in the near on-shell region, obtained with the method of double Feynman parametrization. It reads
$$F^{(c)}(p_{}^2,\mathrm{\Lambda })=_{q_{},\stackrel{~}{q}_{}}_0^1𝑑x_{1x}^1𝑑z\left(\frac{}{m^2}\right)^2\frac{\stackrel{~}{N}(q_{}^2,\stackrel{~}{q}_{}^2,p_{}^2,x,z)}{\stackrel{~}{D}(q_{}^2,\stackrel{~}{q}_{}^2,p_{}^2,m^2,\mathrm{\Lambda }^2,x,z)}$$
(92)
with
$$\stackrel{~}{N}(q_{}^2,\stackrel{~}{q}_{}^2,p_{}^2,x,z)=24(1x)^2p_{}^4\mathrm{\Lambda }^2+\mathrm{\Delta }\stackrel{~}{N}(q_{}^2,\stackrel{~}{q}_{}^2,p_{}^2,x,z)$$
(93)
and
$$\stackrel{~}{D}=z[q_{}^2+\stackrel{~}{q}_{}^2+\mathrm{\Lambda }^2z+(p_{}^2x+m^2)(1x)]^4.$$
(94)
In the term $`\mathrm{\Delta }\stackrel{~}{N}(q_{}^2,\stackrel{~}{q}_{}^2,p_{}^2,x,z)`$ in the numerator we collected all the contributions which are subleading in the on-shell limit. After double derivation with respect to the mass $`m^2`$ and momentum integration we obtain
$$F^{(c)}(p_{}^2,\mathrm{\Lambda }0)\stackrel{p_{}^2m^2}{=}\frac{3}{2\pi ^2}_0^1𝑑x_{1x}^1\frac{dz}{z}\frac{\mathrm{\Lambda }^2p_{}^4(1x)^4}{[\mathrm{\Lambda }^2z+(p_{}^2x+m^2)(1x)]^4}.$$
(95)
In order to study the $`\mathrm{\Lambda }0`$ limit at $`p_{}^2=m^2`$ it is convenient do rescale the $`z`$ variable as
$$z=\stackrel{~}{z}(1x),_{1x}^1\frac{dz}{z}=_1^{1/(1x)}\frac{d\stackrel{~}{z}}{\stackrel{~}{z}}$$
(96)
We see that for $`x1`$ the upper limit of the $`\stackrel{~}{z}`$integral tends to infinity, therefore the $`z`$integration simplifies. Then the $`x`$integration is done taking in account that the $`x1`$ integration region dominates in the near on-shell region; finally one obtains
$$F^{(c)}(p_{}^2=m^2,\mathrm{\Lambda })\stackrel{\mathrm{\Lambda }0}{=}\frac{1}{6\pi ^2}\frac{m^2}{\mathrm{\Lambda }^4}.$$
(97)
It is important to notice that even if the $`\stackrel{~}{D}_\mathrm{\Lambda }`$ term dominates the soft divergence at $`m0`$, it can be always neglected in the analysis of hard or collinear divergences, i.e. if $`m=0`$. In this case in fact the contribution from the $`\stackrel{~}{D}_\mathrm{\Lambda }`$ term is quadratically vanishing. This point can be understood with a dimensional argument and explicitly checked by using the double Feynman parametrization formula and by noticing that at $`m^2=0`$ it is the $`x0`$ region of integration which dominates, not the $`x1`$ region. Then the $`z`$integration can be done by using
$$_{1x}^1\frac{dz}{z}f(x,z)xf(x,1),x0$$
(98)
and the $`x`$integration by using the tricks reported in appendix B. With similar techniques we checked that the gluon self-energy integral in the $`\mathrm{\Lambda }0`$ limit goes smoothly to the standard massless integral for off-shell Euclidean momenta $`p_{}^2p_{}^2`$.
## 7 The Wilson loop test
The Wilson loop is the simplest physical quantity where the effects and the problems of the infrared regularization can be studied. In particular our scope here is to study the subtilities of the $`\mathrm{\Lambda }0`$ limit and to test how the essential property of the gauge-invariance of the Wilson loop is recovered when the infrared cutoff is removed. In concrete in this section we compute the Wilson up to order $`O(g^2)`$ in perturbation theory. This is enough for elucidating various important features of massive axial, planar and light-cone gauges and it is a first step versus a more comprehensive computation at order $`O(g^4)`$ in perturbation theory.
For definiteness, we shall consider a rectangular Wilson loop $`\mathrm{\Gamma }_{LT}`$ of size $`2L\times 2T`$, with $`TL`$. We shall work in Euclidean space with coordinates $`x_1,x_2,x_3,x_4`$ and we shall take the loop in the plane $`x_2x_3`$, as shown in figure 2. We should notice that $`T`$ denotes a lenght in the spatial direction $`x_3`$, i.e. the direction of the gauge vector $`n_E^\mu =(0,0,1,0)`$, and not in the temporal direction $`x_4=ix_0`$. Therefore apparently this loop in not related to the interquark potential. Nevertheless if the theory is consistent all directions in the Euclidean space must be physically equivalent at $`\mathrm{\Lambda }0`$, therefore the final result must be the same that for the loop corresponding to the interquark potential where the side $`T`$ is in the temporal direction.
Formally the Wilson loop is defined by
$$W_{\mathrm{\Gamma }_{LT}}=\frac{1}{N_c}<\text{Tr}P\mathrm{exp}\left(ig_{\mathrm{\Gamma }_{LT}}A_\mu 𝑑x^\mu \right)>$$
(99)
where $`P`$ denotes the path ordering on the loop $`\mathrm{\Gamma }_{LT}`$, Tr is the trace in the fundamental representation of $`SU(N_c)`$ and the average is evaluated via a perturbative expansion of the Euclidean functional integral. In our case $`\mathrm{\Gamma }_{LT}`$ can be split in four pieces
$$\mathrm{\Gamma }_{LT}=\mathrm{\Gamma }^{(1)}+\mathrm{\Gamma }^{(2)}+\mathrm{\Gamma }^{(3)}+\mathrm{\Gamma }^{(4)},$$
(100)
parametrized as
$$\begin{array}{c}\hfill \mathrm{\Gamma }^{(1)}(s)=sT\left(\begin{array}{ccc}0& & \\ 0& & \\ +1& & \\ 0& & \end{array}\right),\mathrm{\Gamma }^{(2)}(s)=sL\left(\begin{array}{ccc}0& & \\ 1& & \\ 0& & \\ 0& & \end{array}\right),\\ \hfill \mathrm{\Gamma }^{(3)}(s)=sT\left(\begin{array}{ccc}0& & \\ 0& & \\ 1& & \\ 0& & \end{array}\right),\mathrm{\Gamma }^{(4)}(s)=sL\left(\begin{array}{ccc}0& & \\ +1& & \\ 0& & \\ 0& & \end{array}\right),\end{array}$$
(101)
where the parameter $`s`$ lives in the interval $`[1,1]`$. Equivalently, we can parametrize
$$\mathrm{\Gamma }_\mu ^{(1)}(t)=t\delta _{\mu 3},\mathrm{\Gamma }_\mu ^{(2)}(l)=l\delta _{\mu 2},\mathrm{\Gamma }_\mu ^{(3)}(t)=t\delta _{\mu 3},\mathrm{\Gamma }_\mu ^{(4)}(l)=l\delta _{\mu 2},$$
(102)
with $`t[T,T]`$ and $`l[L,L]`$. The Wilson loop can be easily computed by expanding in powers of the coupling constant,
$$W_{\mathrm{\Gamma }_{LT}}=W_{\mathrm{\Gamma }_{LT}}^{(0)}+gW_{\mathrm{\Gamma }_{LT}}^{(1)}+\frac{g^2}{2}W_{\mathrm{\Gamma }_{LT}}^{(2)}+\frac{g^3}{3!}W_{\mathrm{\Gamma }_{LT}}^{(3)}+O(g^4).$$
(103)
By using the properties of the generators in the fundamental representation of $`SU(N_c)`$, $`\text{Tr}\mathrm{\hspace{0.33em}1}=N_c`$, $`\text{Tr}T_a=0`$ and $`\text{Tr}(T_aT_b)=\frac{1}{2}\delta _{ab}`$, from (99) one obtains the explicit expression
$$W_{\mathrm{\Gamma }_{LT}}=1\frac{g^2}{4N_c}_{\mathrm{\Gamma }_{LT}}𝑑x^\mu 𝑑y^\nu <A_\mu (x)A_\nu (y)>^{(0)}+O(g^4).$$
(104)
Notice that higher orders corrections begins at order $`O(g^4)`$, not $`O(g^3)`$. The expectation value $`<A_\mu (x)A_\nu (y)>^{(0)}`$, computed at zero order in the coupling constant, coincides with the Fourier transform of the free propagator, therefore a a more explicit expression for the $`O(g^2)`$ contribution is
$$W_{\mathrm{\Gamma }_{LT}}^{(2)}=C_F_1^1𝑑s_1_1^1𝑑s_2\frac{\text{d}\mathrm{\Gamma }^\mu }{\text{d}s_1}\frac{\text{d}\mathrm{\Gamma }^\nu }{\text{d}s_2}D_{\mu \nu }(\mathrm{\Gamma }(s_1)\mathrm{\Gamma }(s_2)),$$
(105)
where the relation $`\delta ^{aa}=2N_cC_F`$ with $`C_F=(N_c^21)/(2N_c)`$ has been used. Specializing to the rectangular Wilson loop in figure 2 we have in general sixteen contribution to (105); however many terms gives the same contribution due to the symmetries $`D_{\mu \nu }(x)=D_{\nu \mu }(x)`$ and $`D_{\mu \nu }(x)=D_{\mu \nu }(x)`$ and finally one is left with the explicit expression
$$\begin{array}{c}\hfill \frac{W_{\mathrm{\Gamma }_{LT}}^{(2)}}{2C_F}=_T^T𝑑t_1_T^T𝑑t_2[D_{33}(0,2L,t_1+t_2,0)+D_{33}(0,0,t_1t_2,0)]\\ \hfill _L^L𝑑l_1_L^L𝑑l_2[D_{22}(0,l_1+l_2,2T,0)+D_{22}(0,l_1l_2,0,0)]\\ \hfill +4_T^T𝑑t_L^L𝑑lD_{23}(0,L+l,Tt,0).\end{array}$$
(106)
This formula can be further simplified in the $`T\mathrm{}`$ limit since various contributions are subleading. Moreover, due to the identity (which holds since $`\mathrm{\Gamma }_{LT}`$ is a closed path)
$$_{\mathrm{\Gamma }_{LT}}𝑑x^\mu 𝑑y^\nu D_{\mu \nu }(xy)=_{\mathrm{\Gamma }_{LT}}𝑑x^\mu 𝑑y^\nu D_{\mu \nu }^T(xy),$$
(107)
actually only the transverse part of the propagator contributes. At zero mass $`D_{\mu \nu }^T(p)`$ is the same in all gauges and this is the reason why the final result is gauge-independent; however at $`\mathrm{\Lambda }0`$ there is a dependence on the gauge vector $`n_\mu `$ coming from the $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ part of the propagator.
A naive expectation would suggest that $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ being proportional to $`\mathrm{\Lambda }^2`$ gives a vanishing contribution to the Wilson loop. This is in general should not be taken for granted. Nevertheless, suppose for a moment that this naive espectation is correct (this point will be analyzed in detail in next section). Then one can consider only the $`\overline{D}_{\mu \nu }`$ part of the propagator or even only the transverse part $`\overline{D}_{\mu \nu }^T`$. This object is independent of the gauge-fixing vector $`n_\mu `$ and therefore the Wilson loop is the same as in covariant gauges. In particular, we can effectively replace $`\overline{D}_{\mu \nu }`$ with $`\delta _{\mu \nu }/(p^2+\mathrm{\Lambda }^2)\delta _{\mu \nu }D_\mathrm{\Lambda }(p)`$ because they have the same transverse part. If we divide by $`2T`$ in order to eliminate the contributions subleading at $`T\mathrm{}`$ we obtain for the $`O(g^2)`$ analogous of the interquark potential
$$V_\mathrm{\Lambda }^{(2)}(2L)\underset{T\mathrm{}}{lim}\frac{g^2N_cW_{\mathrm{\Gamma }_{LT}}^{(2)}}{4T}$$
(108)
the expression
$$V_\mathrm{\Lambda }^{(2)}(2L)=\frac{g^2N_cC_F}{2T}_T^T𝑑t_1_T^T𝑑t_2D_\mathrm{\Lambda }(t_1+t_2,2L)+c$$
(109)
where $`c`$ is the infinite renormalization constant
$$c=\underset{T\mathrm{}}{lim}\frac{g^2N_cC_F}{2T}_T^T𝑑t_1_T^T𝑑t_2D_\mathrm{\Lambda }(t_1t_2,0).$$
(110)
This is the well known ultraviolet divergence of point-like charges and can be eliminated by fixing the potential to be zero at $`L\mathrm{}`$. With simple manipulations involving the representation of the delta function
$$2\pi \delta (p_3)=\underset{T\mathrm{}}{lim}_T^T𝑑x_3e^{ip_3x_3}=\underset{T\mathrm{}}{lim}\frac{2\mathrm{sin}(p_3T)}{p_3}$$
(111)
and the formula
$$\underset{T\mathrm{}}{lim}\frac{(2\mathrm{sin}p_3T)^2}{p_3^2}=\underset{T\mathrm{}}{lim}2T2\pi \delta (p_3),$$
(112)
from (109) one obtains
$$V_\mathrm{\Lambda }^{(2)}(2L)=g^2N_cC_F_{\overline{p}}\frac{\mathrm{exp}(ip_22L)}{p^2+\mathrm{\Lambda }^2}=g^2N_cC_F\frac{\mathrm{exp}(\mathrm{\Lambda }2L)}{4\pi 2L}.$$
(113)
This is the formula of the screened Coulomb potential of two colored charges at distance $`2L`$ and reduces to the standard one when the infrared cutoff is removed. We still stress that this formula is correct as far as we can neglect the contribution from the $`\stackrel{~}{D}_\mathrm{\Lambda }`$ term in the limit $`\mathrm{\Lambda }0`$.
### 7.1 Vanishing of the $`\stackrel{~}{D}_\mathrm{\Lambda }`$ term
Now we prove that the $`\stackrel{~}{D}_\mathrm{\Lambda }`$ term of the propagator gives a vanishing contribution to the Wilson loop. This is a non-trivial point since in this quantity there is a problem of commutativity between the limits $`\mathrm{\Lambda }0`$ and $`T\mathrm{}`$. This can be seen from the explicit expression of the $`\stackrel{~}{D}_\mathrm{\Lambda }`$ contribution to the interquark potential, which up to an unessential infinite constant reads, working for definiteness in the planar gauge,
$$\begin{array}{c}\hfill \stackrel{~}{V}_\mathrm{\Lambda }^{(2)}(2L)=\frac{g^2N_cC_F}{2T}_T^T𝑑t_1_T^T𝑑t_2D_{33}(0,2L,t_1+t_2,0)\\ \hfill =\frac{g^2N_cC_F}{2T}_{p_E}\frac{e^{ip_22L}\mathrm{\Lambda }^2}{[p_3]^2(p_3^2+\overline{p}^2+\mathrm{\Lambda }^2)}\frac{(2\mathrm{sin}p_3T)^2}{p_3^2}.\end{array}$$
(114)
We see that we cannot blindly use the formula (112) and compute the integral directly at $`T\mathrm{}`$. We are instead forced to consider finite $`T`$ and to make some subtle observation. The idea is that the $`p_3`$integral is dominated by the $`p_30`$ region and that actually it reduces to an integral of the kind
$$I_\alpha (T)=_{p_3}\frac{(2\mathrm{sin}p_3T)^2}{(p_3)^{2+2\alpha }}$$
(115)
with $`\alpha =1`$. Strictly speaking this is a divergent integral; nevertheless it can be defined in the region $`1/2<\alpha <1/2`$ where its value is
$$I_\alpha (T)=\frac{2}{\pi }\mathrm{\Gamma }(12\alpha )\mathrm{sin}(\alpha \pi )(2T)^{1+2\alpha }.$$
(116)
After analytic continuation to $`\alpha =1`$ we obtain
$$I_1(T)=\frac{4}{3}T^3.$$
(117)
A simpler way to derive this formula is to derive equation (117) three times with respect to $`T`$: then the sinus representation of the delta function is recovered and the integral is easily computed. In other words we can use the formula, which has to be interpreted in the sense of distributions,
$$\underset{T\mathrm{}}{lim}\frac{(2\mathrm{sin}p_3T)^2}{[p_3]^4}=\underset{T\mathrm{}}{lim}\frac{4}{3}T^32\pi \delta (p_3).$$
(118)
Then the three-dimensional integration is as in (113) and the final result is
$$\stackrel{~}{V}_\mathrm{\Lambda }^{(2)}(2L)=\frac{2g^2N_cC_F}{3}\mathrm{\Lambda }^2T^2\frac{\mathrm{exp}(\mathrm{\Lambda }2L)}{4\pi 2L}.$$
(119)
We see that this contribution is quadratically vanishing in the limit $`\mathrm{\Lambda }0`$, at finite $`T`$. The $`T\mathrm{}`$ limit cannot be taken before the $`\mathrm{\Lambda }0`$ limit. This is the crucial point of our analysis.
For what concerns the situation with other gauges, we observe that in the light-cone gauge we have to compute exactly the same integral as in (114) and therefore we have the same result. On the contrary, in the massive axial gauge we have to replace a factor $`1/[p_3]^2`$ with a factor $`(p_3^2+\mathrm{\Lambda }^2)^1`$ and using (48) we see that the final result is linearly vanishing with $`\mathrm{\Lambda }T`$. This means that the $`O(g^2)`$ Wilson loop test works for any gauge choice: but this is hardly a surprise since the $`O(g^2)`$ computation corresponds to the computation in an Abelian theory. It is well known that problems begin in the non-Abelian case and at higher orders in perturbation theory, starting from order $`O(g^4)`$, therefore we cannot conclude nothing about the final consistency of these Wilsonian gauge choices at this level. Nevertheless we think that this analysis of the free case is very instructive and allows to learn a lot about the possible origin of problems. In particular we have learned that in an $`O(g^4)`$ computation the only part of the propagator we have to control is the $`\stackrel{~}{D}_\mathrm{\Lambda }`$ part, which is quite simple and surely can be studied with a relatively little analitical effort. In other Wilsonian formulations of non-covariant gauges based on infrared cutoffs more complicate than a mass-like term the analysis is technically much more cumbersome but physically equivalent in what concerns the final results in the $`\mathrm{\Lambda }0`$ limit. The drawback of the generic cutoff is that the fine-tuning problem must be solved in order to fix the correct boundary conditions (renormalization prescriptions) such to have an infrared limit consistent with the gauge-symmetry; this difficult problem is avoided with the mass cutoff since with this choice the theory is Ward-identities-consistent to all scales. Finally we would stress the fact the planar gauge and in particular the light-cone gauge are expected to be much more regular than the axial gauge in the zero mass limit: this expectation is reflected in the present computation by the fact that the $`\stackrel{~}{D}_\mathrm{\Lambda }`$ term is quadratically suppressed in both planar and light-cone gauge and merely linearly suppressed in axial gauge.
## 8 Conclusions
In we pointed out that in algebraic non-covariant gauges it is possible to build up a Wilsonian formulation of gauge theories consistent with the Ward identities provided that the infrared cutoff is introduced as a formal “mass” term. However, we stressed that this infrared cutoff cannot be physically interpreted and must be removed in order to recover the essential property of gauge-independence of physical quantities. In this paper we have investigated the properties of the singular limit $`\mathrm{\Lambda }0`$ by explicitly computing various quantities. We have seen in general that the pure axial gauge choice is problematic since the Fourier transform of the propagator and even the simplest Feynman diagrams which are finite in covariant gauges are instead divergent at $`\mathrm{\Lambda }0`$, for any configuration of momenta. Moreover, we argued that other gauge choices, namely the planar gauge and the light-cone gauge are in a much better shape. We have however seen that the structure of infrared divergences is quite subtle even in these cases and that there are gauge-dependent quantities such as for on-shell configurations of momenta the contribution from the gauge-dependent term of the propagator $`\stackrel{~}{D}_{\mathrm{\Lambda },\mu \nu }`$ is not only non-vanishing, but even it is dominant with respect to the standard contribution. Therefore the $`\mathrm{\Lambda }0`$ limit for these quantities is delicate. However these quantities are unphysical and this fact should not be considered as suggesting an inconsistency of the theory. The only way to prove the consistency or the inconsistency of the approach is by considering true physical quantities like the interquark potential as obtained from a Wilson loop of size $`2L\times 2T`$ with $`T\mathrm{}`$. We have seen that the limits $`\mathrm{\Lambda }0`$ and $`T\mathrm{}`$ must be studied with great care, since they do not commute. In particular the crucial property of the exponentiation of the Wilson loop is recovered only if the $`\mathrm{\Lambda }0`$ limit is taken before the $`T\mathrm{}`$ limit. We have seen that the planar gauge and the light-cone gauge appear to be much more regular than the axial gauge in the infrared limit. This is explicitly realized in the Wilson loop $`O(g^2)`$ computation by the fact that the gauge dependent term is quadratically suppressed at small $`\mathrm{\Lambda }T`$ whereas in axial gauge this term is only linearly suppressed. We notice that in covariant gauges, by accident, the order of limits is not crucial at order $`O(g^2)`$, then this feature has not been recognized previously; nevertheless it is manifest in an $`O(g^4)`$ computation. The reason is that the infrared cutoff breaks the BRST-invariance and therefore the cancellation of the so called non-Abelian contributions to the Wilson loop no more works at $`\mathrm{\Lambda }0`$. Therefore the interquark potential cannot be perturbatively defined at $`\mathrm{\Lambda }0`$ and $`T\mathrm{}`$, but only at finite $`T`$. Finally we notice that the physical interquark potential, defined exactly at $`\mathrm{\Lambda }=0`$, is independent of the cutoff function choice when it is computed order by order in perturbation theory.
For the future, we plan to study in detail what happens in the $`O(g^4)`$ computation. Actually, at finite $`T`$, we expect that in planar and light-cone gauges the computation reduces smoothly to the standard massless computation, which is consistent, whereas the axial gauge should be very delicate and possibly inconsistent. Still, we should notice that even in the cases where the $`O(g^4)`$ Wilson loop test fails, in principle this only indicate a failure of the perturbative expansion and not necessarily of the full theory. Nevertheless one would feel much more confortable with an approach admitting a perturbative expansion. In such a perspective, the most promising possibility seems to be the light-cone gauge which certainly deserves additional investigations.
Note Added. During the completion of this work we received a communication from R. Soldati and A. Panza who explicitly computed the Wilson loop at order $`O(g^4)`$ in the massive axial gauge case and proved that the $`\mathrm{\Lambda }0`$ limit is singular. Therefore the axial gauge choice seems to be definitely pathological even in the Wilson renormalization group approach, at least at the perturbative level.
Acknowledgements
I acknowledge support from Padova University during the early stages of this work. I thank R. Soldati and A. Panza for useful discussions and for communicating to me the result of before publication.
## Appendix A Gauge-dependence of the pressure
In this appendix we provide a very simple and illuminating example which illustrates the gauge-dependence problem in presence of a non-zero mass cutoff $`\mathrm{\Lambda }`$.
Consider a free gas of photons in thermal equilibrium at temperature $`T=1/\beta `$ in a box of volume $`V`$. We can compute, as a typical quantity directly related to the partition function, the pressure of this gas in presence of the mass cutoff $`\mathrm{\Lambda }`$. We obtain different result in different gauges, but all these results collapse to the correct physical result in the physical limit $`\mathrm{\Lambda }0`$. This is trivial example, since the theory is free, nevertheless we believe it is very instructive. We recall that the pressure is a typical quantity which cannot be computed in perturbation theory in thermal field theory and where the non-perturbative powerfulness of the Wilson renormalization group approach could give an alternative way of facing the problem. In this sense it is an interesting quantity.
We begin the computation by recalling some elementary facts (see for instance ). In quantum field theory the thermodynamic pressure
$$p(\beta ,V)=\frac{1}{\beta V}\mathrm{ln}Z_{\beta V}$$
(120)
is defined in terms of the partition function
$$Z_{\beta V}=[d\varphi ]\mathrm{exp}\left(_0^\beta 𝑑\tau _Vd^3x_E(\varphi ,_\mu \varphi ;\mathrm{\Lambda }_0)\right)$$
(121)
where $`_E(\varphi ,_\mu \varphi ;\mathrm{\Lambda }_0)`$ is the bare Euclidean lagrangian of the theory, which is a function of the bare parameters depending on the ultraviolet cutoff $`\mathrm{\Lambda }_0`$. For instance for a free scalar field of mass $`\mathrm{\Lambda }`$ we have,
$$_E(\varphi ,_\mu \varphi ;\mathrm{\Lambda }_0)=\left(\frac{1}{2}_\mu \varphi ^\mu \varphi \right)_{\mathrm{\Lambda }_0}+\frac{1}{2}\mathrm{\Lambda }^2\varphi ^2+c_4(\mathrm{\Lambda },\mathrm{\Lambda }_0),$$
(122)
where the notation $`\left(\frac{1}{2}_\mu \varphi ^\mu \varphi \right)_{\mathrm{\Lambda }_0}`$ reminds that there is an ultraviolet cutoff inserted in the propagator in momentum space and the term $`c_4(\mathrm{\Lambda },\mathrm{\Lambda }_0)=\mathrm{\Lambda }_0^4[\stackrel{~}{c}_4+\stackrel{~}{c}_4^{}\mathrm{\Lambda }^2/\mathrm{\Lambda }_0^2+O(\mathrm{\Lambda }^4/\mathrm{\Lambda }_0^4)]`$ is a vacuum energy counterterm of dimension four, which will be fixed later on by imposing the normalization condition
$$\underset{\beta \mathrm{}}{lim}\mathrm{ln}Z_{\beta V}=0,$$
(123)
i.e. the partition function is fixed to be 1 at zero temperature. It is interesting to notice that in order to impose this normalization the ultraviolet regularization is needed even if the theory is free. Using the gaussian integration formula we obtain
$$Z_{\beta V}=𝒩\underset{\beta V}{det}(_E^2+\mathrm{\Lambda }^2)^{1/2}\mathrm{exp}(c_4(\mathrm{\Lambda },\mathrm{\Lambda }_0)\beta V)$$
(124)
where $`𝒩`$ is a temperature independent normalization factor to be fixed later whereas the determinant of the operator $`_E^2+\mathrm{\Lambda }^2`$ (acting on functions periodic in the imaginary time) is defined as
$$\underset{\beta V}{det}(_E^2+\mathrm{\Lambda }^2)^{1/2}\mathrm{exp}\left[\frac{1}{2}\text{Tr}_{\beta V}\mathrm{ln}(_E^2+\mathrm{\Lambda }^2)\right]$$
(125)
with
$$\text{Tr}_{\beta V}\mathrm{ln}(_E^2+\mathrm{\Lambda }^2)V\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}_\stackrel{}{p}\mathrm{ln}[(p_n^4)^2+\stackrel{}{p}{}_{}{}^{2}+\mathrm{\Lambda }^2],p_n^42\pi nT.$$
(126)
Using the stardard summation formula<sup>7</sup><sup>7</sup>7In order to use equation (127) we have to remove the ultraviolet cutoff in the energy variable $`p_4`$, otherwise the summation reduces to a finite sum with $`N\mathrm{\Lambda }_0/T`$ terms. In general the summation formula is defined up to a possibly divergent constant which can be reabsorved in the vacuum energy counterterm $`c_4(\mathrm{\Lambda },\mathrm{\Lambda }_0)`$.
$$\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\mathrm{ln}[(p_n^4)^2+X^2]=2\mathrm{log}[1+n(X)]+\beta X,$$
(127)
where $`n(X)=(\mathrm{exp}(\beta X)1)^1`$ is the Bose-Einstein distribution function, one obtains
$$\mathrm{ln}Z_{\beta V}=V_\stackrel{}{p}\mathrm{ln}[1+n(\omega _p(\mathrm{\Lambda }))]\frac{1}{2}\beta V_\stackrel{}{p}^{\mathrm{\Lambda }_0}\omega _p(\mathrm{\Lambda })\beta Vc_4+\mathrm{ln}𝒩,$$
(128)
with $`\omega _p(\mathrm{\Lambda })=\sqrt{\stackrel{}{p}{}_{}{}^{2}+\mathrm{\Lambda }^2}`$. The first term vanishes at $`\beta \mathrm{}`$. Imposing the normalization prescription (123) fixes the counterterm $`c_4(\mathrm{\Lambda },\mathrm{\Lambda }_0)`$ and the normalization factor $`𝒩`$ to be
$$c_4(\mathrm{\Lambda },\mathrm{\Lambda }_0)=\frac{1}{2}_\stackrel{}{p}^{\mathrm{\Lambda }_0}\omega _p(\mathrm{\Lambda }),𝒩=1,$$
(129)
therefore the pressure reduces to the well known expression
$$p(T,\mathrm{\Lambda })=T_\stackrel{}{p}\mathrm{ln}[1+n(\omega _p)]\stackrel{\mathrm{\Lambda }0}{=}\frac{\pi ^2}{90}T^4.$$
(130)
Now we can repeat the same computation in the gauge theory case which is less trivial. We will consider the general linear gauge (16) with $`\xi _20`$. Using the Gaussian integration formula we obtain
$$Z_{\beta V}(L,\xi _2)=𝒩det\xi _2^{1/2}\underset{\beta V}{det}^\mu L_\mu \underset{\beta V}{det}D_{\mu \nu }^{1/2}$$
(131)
where $`det\xi _2^{1/2}`$ comes from the integration of auxiliary fields, $`det_{\beta V}^\mu L_\mu `$ from the integration of ghost fields and $`det_{\beta V}D_{\mu \nu }^{1/2}`$ from the integration of gauge fields. With a lengthy but straighforward computation, one obtains the determinant of the Euclidean propagator in the general class of linear gauges as
$$detD_{\mathrm{\Lambda },\mu \nu }(p)=\frac{1}{(p^2+\mathrm{\Lambda }^2)^2}\frac{\xi _2}{(pL)^2+\mathrm{\Lambda }^2[L^2+\xi _2(p^2+\mathrm{\Lambda }^2)]}.$$
(132)
Suppose for a moment that $`\mathrm{\Lambda }=0`$: in this case we see that the partition function is gauge-independent since the $`\xi _2`$ term in the photon propagator determinant is cancelled by the contribution coming from the integration of the auxiliary fields whereas the $`L`$dependent part is cancelled by an analogous contribution from the ghost propagation, which in turn is fixed by requiring BRST-invariance of the total action. However, as can be explicitly seen, this cancellation mechanism of the gauge-dependence does not work in presence of an infrared regulator which breaks BRST-invariance. Therefore at finite $`\mathrm{\Lambda }`$ one obtains an unphysical dependence on the gauge-fixing parameters. In particular in the limit $`\xi _20`$ i.e. strictly imposing the condition $`L^\mu A_\mu =0`$ one obtains for the logarithm of the partition function the explicit expression
$$\begin{array}{cc}\hfill \mathrm{ln}Z_{\beta V}& =2\text{Tr}_{\beta V}\mathrm{ln}(_E^2+\mathrm{\Lambda }^2)^{1/2}+\text{Tr}_{\beta V}\mathrm{ln}((L)^2+L^2\mathrm{\Lambda }^2)^{1/2}\hfill \\ & \text{Tr}_{\beta V}\mathrm{ln}((L)^2)^{1/2}\beta Vc_4(\mathrm{\Lambda },\mathrm{\Lambda }_0)+\mathrm{ln}𝒩.\hfill \end{array}$$
(133)
The first term is the expected one, corresponding to the pressure of two bosonic degrees of freedom; however there are also terms which are explicitly $`L`$dependent at $`\mathrm{\Lambda }0`$ and therefore unphysical; the $`L`$dependence only cancels at $`\mathrm{\Lambda }0`$ where one recovers the correct result
$$p(T;L)\stackrel{\mathrm{\Lambda }0}{=}\frac{\pi ^2}{45}T^4L.$$
(134)
This is the generic situation, however it is interesting to study what happens in specific gauge choices since the gauge-dependence of the pressure can be less prononced than expected. For instance in the light-cone gauge $`L^2=n^2=0`$ we see that the $`n_\mu `$dependent terms cancel even at $`\mathrm{\Lambda }0`$. Actually the $`n_\mu `$dependence cancels in the large class of non-covariant gauges such as $`L_\mu (p)`$ does not depend on the $`p_4`$ variable (for instance this is the case for the planar gauge and the Coulomb gauge): in this case the $`L`$dependent terms in $`\mathrm{ln}Z_{\beta V}`$ have the form $`\beta Vf(L,\mathrm{\Lambda },\mathrm{\Lambda }_0)`$ and therefore can be reabsorved in the in the vacuum energy counterterm $`c_4(\mathrm{\Lambda },\mathrm{\Lambda }_0)`$. In other words, they are eliminated by the normalization condition (123) and do not contribute to the pressure even at $`\mathrm{\Lambda }0`$. However the dependence on the quantization direction should be expected in a two-loop computation; moreover it is evident in other observables, for instance in the Wilson loop computation, therefore in any case the infrared cutoff $`\mathrm{\Lambda }`$ cannot be physically interpreted as a true mass term, consistently with the standard lore that the only way to give a physical mass to a non-Abelian theory is via the Higgs mechanism. Nevertheless from this simple example one could extrapolate the conjecture (to be checked case by case perturbatively with higher order computations or non-perturbatively with a numerical analysis) that for particular observables there are classes of gauges more regular that others, in which the gauge-dependence is very mild even for non-zero $`\mathrm{\Lambda }`$. Clearly, this is a very interesting point in the spirit of phenomenological numerical analysis and should be investigated in the future.
## Appendix B One-loop integrals
In this appendix we give some generalities on the computation of Euclidean one-loop integrals in planar and light-cone gauges. We begin by fixing our notations on integrals: for axial and planar gauges we define
$$_x=d^4x,_q=\frac{d^4q}{(2\pi )^4},_{q_3}=\frac{dq_3}{2\pi },_{\overline{q}}=\frac{d^3\overline{q}}{(2\pi )^3},$$
(135)
with $`\overline{q}=(q^0,q^1,q^2)`$, whereas for the light-cone gauge we define
$$_q_{}=\frac{d^2q_{}}{(2\pi )^2},_q_{}=\frac{d^2q_{}}{(2\pi )^2},$$
(136)
with $`\stackrel{}{q}_{}=(q^1,q^2)`$, $`\stackrel{}{q}_{}=(q^3,q^0)`$. Euclidean vectors are obtained after Wick rotation $`p_4=ip_0`$; we shall use the notations
$$p_E=(\stackrel{}{p},ip_0),q_E=(\stackrel{}{q},iq_0),p_Eq_E\delta _{\mu \nu }p_E^\mu q_E^\nu =g_{\mu \nu }p^\mu q^\nu =pq$$
(137)
and
$$p_{}^2=p_3^2,p_{}^2=\overline{p}^2\text{(planar gauge)}$$
(138)
$$p_{}^2=p_3^2+p_4^2,p_{}^2=p_1^2+p_2^2\text{(light-cone gauge)}$$
(139)
In this appendix we will always work in Euclidean space even if for sake of notational convenience the index $`E`$ will be neglected.
For semplicity we will restrict our remarks to the computation of the one-loop self-energy of a scalar quark. In general this is a rather complicate function $`F=F(p_{},p_{},\mathrm{\Lambda })`$, but we can give analytical estimations in the two limits $`|p_{}^2|p_{}^2`$ and $`|p_{}^2|p_{}^2`$. In the first case by putting $`p_{}=0`$ we see that the spurious terms $`1/[qn]`$ cancel and then the one-loop integral can be put in an explicitly covariant form: therefore it can be computed through the usual Feynman parametrization. The case $`|p_{}^2|p_{}^2`$ instead is more cumbersome, nevertheless it can be easily implemented with a symbolic manipolation package. As a matter of fact, we prepared a set of routines based on the double Feynman parametrization method allowing to compute analytically all the integrals which are encountered in one-loop self-energy diagrams, including finite parts, and we checked that the $`\mathrm{\Lambda }0`$ limit reproduces the known results of the standard approach. However, for sake of brevity, here we simply sketch the analysis for the simplest examples: the generalization to more complicate cases is straighforward.
The general form of the Feynman integral one encounters in the evaluation of the quark self-energy (or its derivatives with respect to the mass) is given by the expression
$$F((pn)^2,p^2,\mathrm{\Lambda })=_q\frac{N(q,p,n)}{[qn](q^2+\mathrm{\Lambda }^2)((q+p)^2+m^2+\mathrm{\Lambda }^2)^{1+\alpha }}.$$
(140)
The integral (140) will be explicitly evaluated both in planar gauge and light-cone gauge in next subsections.
### B.1 Planar gauge integrals
Consider first the planar gauge case where $`qn=q_3`$. The most convenient technical tool we found to manage this kind of integrals is the double Feynman parametrization which consists in using the identities
$$\frac{1}{q_3}\frac{1}{q^2+\mathrm{\Lambda }^2}=_0^1𝑑y\frac{q_3}{[q_3^2+y(\overline{q}^2+\mathrm{\Lambda }^2)]^2}$$
(141)
and
$$\begin{array}{cc}& _q\frac{N(q,p,n)}{(q^2+A^2)^a((q+p)^2+m^2+\mathrm{\Lambda }^2)^b}=\hfill \\ & \frac{1}{B(a,b)}_q_0^1𝑑x\frac{N(qp(1x),p,n)x^{a1}(1x)^{b1}}{[q_3^2+Ax+(\overline{q}^2+p^2x+m^2+\mathrm{\Lambda }^2)(1x)]^{a+b}.}\hfill \end{array}$$
(142)
Then it is convenient to introduce the variable $`z=(y1)x+1`$ lying in the interval $`1xz1`$ and to rescale $`\stackrel{~}{q}=z^{1/2}\overline{q}`$; in this way equation (140) can be rewritten in the form
$$F=_{q_3,\stackrel{~}{q}}_0^1𝑑x_{1x}^1𝑑z\frac{[q_3N(qp(1x),p,n)]_{tr}(1x)^\alpha }{z^{3/2}[q_3^2+\stackrel{~}{q}^2+(p^2x+m^2)(1x)+\mathrm{\Lambda }^2z]^{3+\alpha }},$$
(143)
where we have introduced the translated and rescaled numerator
$$[q_3N(q,p,n)]_{tr}=[q_3N(q,p,n)]_{q_3q_3p_3(1x)}^{\overline{q}^2=\stackrel{~}{q}^2/z}.$$
(144)
After symmetrization in $`q_3`$ and three-dimensional angular average in the numerator, the momentum integrals can be performed by using the general formula
$$_{q_3,\stackrel{~}{q}}\frac{(q_3^2)^{M_1}(\stackrel{~}{q}^2)^{M_2}}{(q_3^2+\stackrel{~}{q}^2+A)^N}=\frac{B_1(M_1,M_2,N)}{8\pi ^3A^{NM_1M_22}}$$
(145)
where $`B_1(M_1,M_2,N)`$ denotes the product of beta functions
$$B_1=B(NM_1M_22,M_2+3/2)B(NM_11/2,M_1+1/2).$$
(146)
The integral in $`z`$ is trivial at $`\mathrm{\Lambda }=0`$ and a little complicate at $`\mathrm{\Lambda }0`$ but still expressible in terms of elementary functions; the integral in $`x`$ instead in non-trivial. Nevertheless one can explicitly check that it is finite and in general can be expressed in terms of special functions. In the $`\mathrm{\Lambda }0`$ limit the analysis strongly simplifies and one obtain the esplicit result reported in the text.
With a simple generalization of this method one can compute Feynman integrals where the double pole $`1/[qn]^2`$ appears, without encountering any problem. Moreover, the extension of this method to the integrals appearing in the gluon self-energy computation is straightforward.
### B.2 Light-cone gauge integrals
Consider now the light-cone gauge case where $`qn=iq_4+q_3`$. Still the double Feynman parametrization can be used, it is enough to replace identity (141) with
$$\frac{1}{iq_4+q_3}\frac{1}{q^2+\mathrm{\Lambda }^2}=_0^1𝑑y\frac{iq_4+q_3}{[q_4^2+q_3^2+y(q_{}^2+\mathrm{\Lambda }^2)]^2}.$$
(147)
Using (142), introducing the variable $`z=(y1)x+1`$ and rescaling $`\stackrel{~}{q}_{}=z^{1/2}q_{}`$ the Feynman integral (140) can be rewritten in the form
$$F=_{q_{},\stackrel{~}{q}_{}}_0^1𝑑x_{1x}^1𝑑z\frac{[(iq_4+q_3)N(q,p,n)]_{tr}(1x)^\alpha }{z[q_{}^2+\stackrel{~}{q}_{}^2+(p^2x+m^2)(1x)+\mathrm{\Lambda }^2z]^{3+\alpha }}$$
(148)
where we have defined the translated and rescaled numerator as
$$[(iq_4+q_3)N(q,p,n)]_{tr}=[(iq_4+q_3)N(q,p,n)]_{\stackrel{}{q}_{}\stackrel{}{q}_{}\stackrel{}{p}_{}(1x)}^{q_{}^2=\stackrel{~}{q}_{}^2/z}.$$
(149)
Now we can perform the angular average in two dimensions
$$<f(\stackrel{}{q}_{}\stackrel{}{p}_{})>_{\mathrm{\Omega }_2}\frac{1}{2\pi }_0^{2\pi }𝑑\theta f(q_{}p_{}\mathrm{cos}\theta ),$$
(150)
by using the general formulae
$$<(\stackrel{}{q}_{}\stackrel{}{p}_{})^{2n}>_{\mathrm{\Omega }_2}=\frac{\mathrm{\Gamma }(n+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(n+1)}q_{}^{2n}p_{}^{2n},<(\stackrel{}{q}_{}\stackrel{}{p}_{})^{2n+1}>_{\mathrm{\Omega }_2}=0.$$
(151)
In particular
$$<(\stackrel{}{q}_{}\stackrel{}{p}_{})^2>_{\mathrm{\Omega }_2}=\frac{1}{2}q_{}^2p_{}^2,<(\stackrel{}{q}_{}\stackrel{}{p}_{})^4>_{\mathrm{\Omega }_2}=\frac{3}{8}q_{}^4p_{}^4.$$
(152)
The integrals in $`q_{},\stackrel{~}{q}_{}`$ can be performed by using the general formula
$$_{q_{},\stackrel{~}{q}_{}}\frac{(q_{}^2)^{M_1}(\stackrel{~}{q}_{}^2)^{M_2}}{(q_{}^2+\stackrel{~}{q}_{}^2+A)^N}=\frac{B_2(M_1,M_2,N)}{16\pi ^2A^{NM_1M_22}}$$
(153)
where
$$B_2=B(NM_1M_22,M_2+1)B(NM_11,M_1+1).$$
(154)
The integral in $`z`$ is is trivial at $`\mathrm{\Lambda }=0`$ and a little complicate at $`\mathrm{\Lambda }0`$ but still expressible in terms of logarithms of $`\mathrm{log}(1x)`$; the integral in $`x`$ in non-trivial. Nevertheless one can explicitly check that it is finite and in general can be expressed in terms of polylogarithmic functions. In the $`\mathrm{\Lambda }0`$ limit and in the on-shell regime $`p^2m^2`$ the full expression strongly simplifies and we obtain the expression involving logarithms given in the text.
With a simple generalization of this method one can compute Feynman integrals where the double pole appears, without encountering any problem.
Moreover, the extension of this method to the integrals appearing in the gluon self-energy computation is straightforward.
### B.3 Feynman parameters integrals
Finally we report some useful trick to perform integrals on Feynman parameters in particular limits. Suppose one has to compute the integral of some function $`f(x,\mathrm{\Lambda })`$ which in the $`\mathrm{\Lambda }0`$ limit is dominate from the $`x0`$ region; this is the case for instance for
$$f(x,\mathrm{\Lambda })=\frac{P^\nu (x)}{(\mathrm{\Lambda }^2+p^2x)^n}$$
(155)
where $`P^\nu (x)`$ is a polynomial a degree $`\nu `$ in $`x`$, with $`\nu <n1`$. This is the typical expression we encounter in the computation of self-energy diagrams; then one can use the identity
$$_0^1𝑑xf(x,\mathrm{\Lambda })=_0^{\mathrm{}}𝑑xf(x,\mathrm{\Lambda })_1^{\mathrm{}}𝑑xf(x,\mathrm{\Lambda })$$
(156)
and neglect the second contribution which is subleading at $`\mathrm{\Lambda }0`$. Then the first integral can be performed by using the formula
$$_0^{\mathrm{}}𝑑x\frac{x^m}{(\mathrm{\Lambda }^2+p^2x)^n}=\frac{B(m+1,nm1)}{\mathrm{\Lambda }^{2n}}\left(\frac{\mathrm{\Lambda }^2}{p^2}\right)^{m+1},$$
(157)
which holds for $`m>1`$ and $`n>m+1`$. In the case in which the integral is dominate by the $`x1`$ region it is sufficient to change the variable $`x^{}=1x`$ and use the same trick.
|
warning/0005/math-ph0005007.html
|
ar5iv
|
text
|
# Untitled Document
LAVAL-PHY-99-21
Open problems for the superKdV equations
P. Mathieu<sup>1</sup> Work supported by NSERC (Canada) and FCAR (Québec) <sup>2</sup> Talk presented at AARMS-CRM Workshop on Backlund and Darboux Transformations: The geometry of Soliton Theory. June 4-9, 1999 (Halifax, Nova Scotia).
Département de Physique,
Université Laval,
Québec, Canada G1K 7P4
pmathieu@phy.ulaval.ca
Abstract
After a review of the basic results concerning the $`N=1,2`$ supersymmetric extensions of the Korteweg-de Vries equation, with a pedagogical presentation of the superspace techniques, we discuss some basic open problems mainly in relation with the $`N=2`$ extensions.
07/99
1. Introduction
Supersymmetry offers a powerful tool for widening the scope of integrability. The field of supersymmetric integrable systems turns out to be remarkably rich in addition to further displaying novel features such as conserved nonlocal ‘Poisson square roots’ of local conservation laws.<sup>3</sup> Some physical motivations for considering supersymmetric integrable systems are scattered in the footnotes 3, 4 and 7, while geometrical implications are alluded to in the conclusion. Not surprisingly, it started with the extension of the Korteweg-de Vries (KdV) equation although by now many other equations have been supersymmetrized.
The $`N=1,2`$ (where $`N`$ refers to the number of supersymmetries) integrable supersymmetric versions of the KdV equation have been found about 10 years ago .<sup>4</sup> The present restriction to $`N2`$ is partly motivated by the limitations of my own works. However there is also a physical motivation: these systems have found a remarquable application, albeit in their quantum formulation, in perturbed conformal field theory , in the context of which $`N=2`$ is the maximal number of supersymmetries that is of real interest. Nevertheless, in some context, the $`N=3`$ and $`4`$ extensions may be physically relevant and the corresponding extensions of KdV have been considered; see for instance . Our discussion will also be restricted to supersymmetric KdV extensions having an even Poisson brackets. A new $`N=2`$ super KdV equation with an odd Poisson structure has been found recently in but the resulting equations are somewhat less interesting in that the bosonic fields do not couple to the fermions in their time evolution. It would be of interest to see whether there are other ‘odd’ integrable extensions displaying fermionic interactions in the bosonic evolution equations. The key points of this development were 1- the realization that supersymmetrization could be restricted to the space variable only and 2- that the crucial KdV struture whose core needs to be preserved is the KdV second hamiltonian structure.<sup>5</sup> Here are some comments on the literature concerning the $`N=1`$ case. Already in 1984, Kupershmidt has presented a simple fermionic (but not supersymmetric - see below) extension of the KdV equation by placing the emphasis on its hamiltonian formulation (this system has actually two local hamiltonian structures). This was an important step toward the formulation of the right supersymmetric extension given that both hamiltonian operators were supersymmetric invariant (the hamiltonian themselves were not) and could then serve in the formulation of a genuine supersymmetric system. The supersymmetric KdV equation was initially found independently of the work of Manin-Radul on super KP hierarchy. It was realized afterwards that this general system has indeed a reduction to the supersymmetric KdV system. In Fourier components, the underlying hamiltonian operator is the Poisson bracket formulation of the Virasoro algebra , for which supersymmetric extensions were already known and could then be used to construct the appropriate supersymmetric extension of KdV. A number of developments have occured in the following years but recently the focus has moved toward the construction of extended super KdV hierachies<sup>6</sup> The most important physical application of these constructions concerns conformal field theory: the corresponding Poisson structures yield clasical versions of super $`W`$ algebras whose quantum form can be obtained via the quantization of the modified fields obtained through the Miura transformation (see e.g. ). (in the way the Boussinesq equation generalizes KdV, that is, via higher order Lax operators) . However a certain number of problems associated to the $`N=1,2`$ KdV systems have remained unsolved and the goal of this presentation is to identify some of them. For this it is necessary to present a brief review of the supersymmetric formulation of the KdV equation and, for the benefit of those readers unfamiliar with super technologies, some manipulations will be worked out in some detail.
2. Supersymmetrization of the KdV equation: N=1
We will formulate the supersymmetric extension of the KdV equation in the superspace formalism. That amounts to extend the $`x`$ variable to a doublet $`(x,\theta )`$ where $`\theta `$ is a Grassmannian (or anticommuting) variable: $`\theta ^2=0`$. Ordinary (i.e. commuting) fields $`f(x)`$ (functions of $`x`$ and $`t`$ in fact but the time dependence will generally be suppressed) will be replaced by superfields $`F(x,\theta )`$. Given that $`\theta ^2=0`$, these superfields have a very simple Taylor expansion in terms of $`\theta `$:
$$F(x,\theta )=f(x)+\theta \gamma (x)$$
$`(2.1)`$
$`f`$ and $`\gamma `$ are called the component fields. $`\gamma `$ is said to be the super-partner of $`f`$ and vice-versa. In the present case, $`F(x,\theta )`$ is a bosonic superfield: it has the same ‘statistics’ (i.e. commuting or anticommuting character) as the field appearing in the $`\theta `$ independent term (here $`f`$); on the other hand, $`\gamma `$ is anticommuting, i.e. it is a fermionic field. In particular, $`\gamma (x)\gamma (y)=\gamma (y)\gamma (x)`$ so that $`\gamma (x)^2=0`$; also for instance, $`\theta \gamma =\gamma \theta `$. The final ingredient that we need is the superderivative
$$D=\theta +_\theta $$
$`(2.2)`$
whose square is the usual space derivative: $`D^2=`$.
A supersymmetry transformation is nothing but a translation in superspace. Such a translation takes the form: $`xx\eta \theta `$ and $`\theta \theta +\eta `$, where $`\eta `$ is a constant anticommuting parameter, supposed, in the following, to be very small. Consider then the effect of the translation in the superfield:
$$\begin{array}{cc}\hfill F(x,\theta )F(x\eta \theta ,\theta +\eta )& =F(x,\theta )\eta \theta F(x,\theta )+\eta _\theta F(x,\theta )\hfill \\ & F(x,\theta )+\delta _\eta F(x,\theta )\hfill \\ & =f(x)+\theta \gamma (x)+\delta _\eta f(x)+\theta \delta _\eta \gamma (x)\hfill \end{array}$$
$`(2.3)`$
The second equality shows that $`\delta _\eta `$ is bosonic so that it commutes with $`\theta `$. We read off the component-field transformations to be
$$\delta _\eta f=\eta \gamma ,\delta _\eta \gamma =\eta f_x$$
$`(2.4)`$
This is called a supersymmetry transformation; it relates a bosonic field to a fermionic field and vice-versa. It has the remarquable virtue of linking a field transformation to a (super)space translation. Note that two successive supersymmetry transformations lead to
$$\delta _\eta \delta _\eta ^{}f=\eta ^{}\eta f_x,\delta _\eta \delta _\eta ^{}\gamma =\eta ^{}\eta \gamma _x$$
$`(2.5)`$
In other words, a translation in superspace, hence a supersymmetry transformation, is a sort of square root of an ordinary translation. Every local expression in the superfields and the superderivatives is manifestly supersymmetric invariant.
To supersymmetrize the KdV equation
$$u_t=u_{xxx}+6uu_x$$
$`(2.6)`$
one should then start by extending the $`u`$ field to a superfield. There are two ways of doing this: either as a fermonic superfield
$$u(x)\varphi (x,\theta )=\theta u(x)+\xi (x)$$
$`(2.7)`$
or as a bosonic superfield
$$u(x)U(x,\theta )=u(x)+\theta \lambda (x)$$
$`(2.8)`$
It turns out that the first choice is the one that gives the interesting extension.<sup>7</sup> The other possibility would not have a second hamiltonian structure associated to the super Virasoro algebra as a simple dimensional (i.e. degree counting) analysis shows (it requires the introduction of an anticommuting field of degree $`3/2`$ as $`\xi `$ and not of degree $`5/2`$ as $`\lambda `$). The degree counting is explained below. Notice that to a large extend, we try to reserve Greek letters for anticommuting variables or fields. The KdV equation is homogeneous with respect to the scaling gradation: in the normalization where deg $`=1`$, one finds that deg $`u=2`$. The identity $`D^2=`$ implies that deg $`D=1/2`$, so that deg $`\theta =1/2`$. For the superfield to be homogeneous, $`\xi `$ must have degree $`3/2`$.
Let us then proceed with a direct extension of the KdV equation, multiplying each term by $`\theta `$ and rewriting the result in terms of superfields :
$$\begin{array}{cc}\hfill u_t& \varphi _t\hfill \\ \hfill u_{xxx}& \varphi _{xxx}\hfill \\ \hfill 3uu_x& c\varphi D\varphi _x+(6c)\varphi _x(D\varphi )\hfill \end{array}$$
$`(2.9)`$
where $`c`$ is a free constant. We thus observe that the nonlinear term does not have a unique extension in terms of superfields. Therefore, this direct extension leaves us with a supersymmetric version of the KdV equation containing a free parameter:
$$\varphi _t=\varphi _{xxx}+c(\varphi D\varphi )_x+(62c)\varphi _x(D\varphi )$$
$`(2.10)`$
It turns out that this equation is integrable only if $`c=3`$ <sup>8</sup> Actualy the case $`c=0`$ is also integrable but its leads to a somewhat trivial system in which the fermionic fields decouple from the bosonic equation which reduces then to the usual KdV equation. Nevertheless, this equation happens to be relevant in supersymmetric extensions of matrix models that describes superstrings in $`d<3/2`$ dimensions, or equivalently, conformal field theories coupled to gravity .. We call the resulting equation the super KdV equation, or sKdV for short. Its component version reads<sup>9</sup> Notice that the product $`\xi \xi _{xx}`$ acts as a bosonic field: quite generally, a product of two fermions is a boson. It can be seen easily that passing an anticommuting variable in front of it does not induce an overall minus sign, e.g. $`\xi \xi _{xx}\theta =\xi \theta \xi _{xx}=\theta \xi \xi _{xx}`$. Notice moreover that this term is a total derivative: $`\xi \xi _{xx}=(\xi \xi _x)_x`$ since the extra resulting term is $`\xi _x\xi _x=0`$.
$$\begin{array}{cc}& u_t=u_{xxx}+6uu_x3\xi \xi _{xx}\hfill \\ & \xi _t=\xi _{xxx}+3(u\xi )_x\hfill \end{array}$$
$`(2.11)`$
It is not difficult to verify that the system (2.11) is invariant under the supersymmetry transformation $`\delta _\eta u=\eta \xi _x`$ and $`\delta _\eta \xi =\eta u`$. This is not so for the integrable fermionic extension proposed by Kupershmidt :
$$\begin{array}{cc}& u_t=u_{xxx}+6uu_x3\xi \xi _{xx}\hfill \\ & \xi _t=4\xi _{xxx}+6u\xi _x+3u_x\xi \hfill \end{array}$$
$`(2.12)`$
With (2.11) being called the super KdV equation, it would be appropriate to call (2.12) the Kuper-KdV equation.
The integrability of (2.10) can be established in various ways. The most direct argument is that it has a Lax representation:<sup>10</sup> Actualy, the Lax operator is not unique: the choice $`L=^2+\varphi D(D\varphi )`$ (the formal adjoint of $`^2\varphi D`$) leads to completely equivalent results.
$$L_t=[4L_+^{3/2},L]L=^2\varphi D$$
$`(2.13)`$
and the conservation laws are obtained as follows (the subscript gives the degree):
$$H_{2k+1}=𝑑x𝑑\theta \mathrm{sRes}L^{(2k+1)/2}$$
$`(2.14)`$
For super pseudodifferential operators, the $`+`$ projection and the super residue sRes are defined as follows
$$\mathrm{\Lambda }=\underset{k=\mathrm{}}{\overset{N}{}}\alpha _iD^i,\mathrm{\Lambda }_+=\underset{k=0}{\overset{N}{}}\alpha _iD^i\mathrm{sRes}\mathrm{\Lambda }=\alpha _1$$
$`(2.15)`$
In the above expression for the conservation laws, we have also introduced the superintegration. The integration over the $`\theta `$ variable is defined as follows:
$$𝑑\theta \mathrm{\hspace{0.17em}1}=0𝑑\theta \theta =1$$
$`(2.16)`$
The integration over $`\theta `$ is thus essentially equivalent to the differentiation with respect to $`\theta `$. With these rules, the superintegration of a superderivative vanishes (with the usal rule that the ordinary integral of a total derivative vanishes):
$$𝑑x𝑑\theta [D\varphi (x,\theta )]=𝑑x𝑑\theta (\theta \xi _x+u)=𝑑x\xi _x=0$$
$`(2.17)`$
For instance, the second conservation law is<sup>11</sup> Notice that for a fermionic variable, $`\xi \xi _x`$ is not a total derivative: a partial integration of $`𝑑x\xi \xi _x`$ leads to $`𝑑x\xi _x\xi `$ and the interchanges of the two terms generates another minus sign so that the original expression is recovered.
$$H_3=𝑑x𝑑\theta (\varphi D\varphi )=𝑑x(u^2\xi \xi _x)$$
$`(2.18)`$
Another way of establishing the integrability is to supersymmetrize the Gardner transformation . This extension is unique:
$$\varphi =\chi +ϵ\chi _x+ϵ^2\chi D\chi $$
$`(2.19)`$
with $`\chi =\theta w+\sigma `$. It maps a solution of the super Gardner equation<sup>12</sup> The component form of this superfield equation reads:
$$\begin{array}{cc}& w_t=w_{xxx}+6ww_x\sigma \sigma _{xx}+ϵ^2[6w^2w_x3(\sigma \sigma _xw)_x]\hfill \\ & \sigma _t=\sigma _{xxx}+3(\sigma w)_x+ϵ^2[3w(w\sigma )_x]\hfill \end{array}$$
Notice that $`\sigma _t`$ is not a total derivative. The usual Gardner equation is recovered by setting the fermionic field $`\sigma =0`$ and the sKdV equation is the limiting case where $`ϵ=0`$.
$$\chi _t=\chi _{xxx}+3(\chi D\chi )_x+ϵ^2\mathrm{\hspace{0.17em}3}(D\chi )(\chi D\chi )_x$$
$`(2.20)`$
into a solution of the sKdV equation. Since $`\chi _t`$ is a total superderivative, e.g.
$$(D\chi )(\chi D\chi )_x=\frac{1}{6}D[(D\chi )^3]+\frac{1}{2}[\chi (D\chi )^2]_x$$
$`(2.21)`$
$`𝑑x𝑑\theta \chi `$ is conserved and by inverting the super Gardner transformation (2.19), we recover an infinite number of conservation laws:
$$\chi =\underset{n=0}{\overset{\mathrm{}}{}}ϵ^nh_n[\varphi ]\frac{d}{dt}𝑑x𝑑\theta \underset{n=0}{\overset{\mathrm{}}{}}ϵ^nh_n[\varphi ]=0$$
$`(2.22)`$
(where $`h_n[\varphi ]`$ stands for a differential polynomial in $`\varphi `$). Now the crucial point is that the sKdV equation is independent of $`ϵ`$ so that each separate power of $`ϵ`$ must be separately conserved. This produces an infinite number of conservation laws, half of which can be shown to be nontrivial, having a leading term $`\varphi (D\varphi )^k`$; these are bound to be the $`H_{2k+1}`$ above . Note that these are all bosonic ($`\chi `$ is fermionic but the measure $`dxd\theta `$ is also fermionic).
Finally, we point out that the sKdV equation is bihamiltonian, the two hamiltonian operators being<sup>13</sup> The second hamiltonain structure has been found in the first two references and the first one in the last two. Here and below, the action of the derivatives is always delimited by parentheses, e.g., $`D\varphi =(D\varphi )\varphi D`$. ,
$$P_1=[D^3\varphi ]^1,P_2=D^5+3\varphi +(D\varphi )D+2\varphi _x$$
$`(2.23)`$
Notice that $`P_1`$ is a very complicated nonlocal hamiltonian operator, being essentially an infinite series: $`[D^3\varphi ]^1=D^3[1D^3\varphi ]^1`$. $`P_2`$ is the direct supersymmetrization of the KdV second hamiltonian structure: $`^3+4u+2u_x`$.
There is a remarquable feature of the super case that is not present for the usual KdV equation which is the presence of fermionic nonlocal conservation laws . The first few of them are
$$\begin{array}{cc}& J_{1/2}=𝑑x𝑑\theta (D^1\varphi )=𝑑x\xi \hfill \\ & J_{3/2}=𝑑x𝑑\theta (D^1\varphi )^2=𝑑xu(^1\xi )\hfill \\ & J_{5/2}=𝑑x𝑑\theta [(D^1\varphi )^36^1(\varphi D\varphi )]=𝑑x[3\xi (^1u)^26^1(u^2\xi \xi _x)]\hfill \end{array}$$
$`(2.24)`$
They Poisson commute with the local bosonic conservation laws $`H_n`$ but not among themselves :
$$\{J_{(4n+i)/2},J_{(4m+i)/2}\}=H_{2(n+m)+i}\mathrm{with}i=1,3\{J_{(4n+1)/2},J_{(4m+3)/2}\}=0$$
$`(2.25)`$
The $`J_i`$ are thus some sort of Poisson square roots of the usual conservation laws. The first fermionic nonlocal conservation law is actually local in terms of the component fields. It signals the presence of the supersymmetry invariance (and in particular $`\xi `$ is not a conserved density for the Kuper-KdV equation). The infinite sequence can be generated from the first two conservation laws by the application of the recursion operator $`P_1^1P_2`$. But there is a more spectacular way of expressing them that makes manifest their supersymmetric origin: in superspace, $`^2`$ not only has a square root but it also has a fourth root: $`(^2)^{1/4}=D`$. The fermionic nonlocal conservation laws are thus related to the super residues of the fourth root of odd powers of the Lax operators as
$$J_{k/2}=𝑑x𝑑\theta \mathrm{sRes}L^{k/4}(k\mathrm{odd})$$
$`(2.26)`$
The sKdV equation represents the first example of an integrable system for which nonlocal conservation laws arise in such a clean form. In that respect, we introduce the first open problem (OP):
OP-1: Find an integrable deformation that reproduces the fermionic nonlocal conservation laws.
3. Supersymmetrization of the KdV equation: N=2
The next extension to be considered is the addition of an extra supersymmetry which amounts to add an extra anticommuting space dimension. We thus extend $`x`$ to a triplet $`(x,\theta _1,\theta _2)`$, with $`\theta _1^2=\theta _2^2=0,\theta _1\theta _2=\theta _2\theta _1`$ and introduce two super derivatives:
$$D_1=\theta _1+_{\theta _1}D_2=\theta _2+_{\theta _2},D_1^2=D_2^2=,D_1D_2=D_2D_1$$
$`(3.1)`$
The superfields are now functions of $`(x,\theta _1,\theta _2)`$ (as well as $`t`$) and their Taylor expansion in terms of the anticommuting variables contain four terms. For instance the $`N=2`$ KdV superfield will be written as:
$$\mathrm{\Phi }(x,\theta _1,\theta _2)=\theta _2\theta _1u(x)+\theta _1\xi _1(x)+\theta _2\xi _2(x)+v(x)$$
$`(3.2)`$
$`\xi _1`$ and $`\xi _2`$ are two fermionic fields and $`v`$ is a new bosonic field of degree $`1`$ (in a supersymmetric theory, the number of bosonic and fermionic fields must be the same). Notice that $`\mathrm{\Phi }`$ is a bosonic superfield.
One could then proceed to the direct supersymmetrization of the KdV equation and get a multiparameter $`N=2`$ extension. However, a sounder approach, that reduces substantially the number of such free parameters, is to formulate the equation directly in terms of the $`N=2`$ supersymmetric version of the second hamiltonian structure (which is expected to be the core structure underlying integrability of nontrivial KdV extensions):
$$P_2=D_1D_2+2\mathrm{\Phi }(D_1\mathrm{\Phi })D_1(D_2\mathrm{\Phi })D_2+2\mathrm{\Phi }_x$$
$`(3.3)`$
e.g. as
$$\mathrm{\Phi }_t=P_2\frac{\delta }{\delta \mathrm{\Phi }}𝑑x𝑑\theta _1𝑑\theta _2[\mathrm{\Phi }(D_1D_2\mathrm{\Phi })+a\mathrm{\Phi }^3]$$
$`(3.4)`$
This hamiltonian is the direct generalisation of the KdV hamiltonian $`𝑑xu^2`$ and of the $`N=1`$ version $`𝑑x𝑑\theta \varphi (D\varphi )`$. However, the $`N=2`$ generalisation is not unique and this introduces a free parameter $`a`$. The resulting equation is
$$\mathrm{\Phi }_t=\mathrm{\Phi }_{xxx}+3(\mathrm{\Phi }D_1D_2\mathrm{\Phi })_x+\frac{(a1)}{2}(D_1D_2\mathrm{\Phi }^2)_x+3a\mathrm{\Phi }^2\mathrm{\Phi }_x$$
$`(3.5)`$
This is called the SKdV<sub>a</sub> equation (the capital S is used for $`N=2`$). This system is integrable for exactly three values of $`a`$ : $`a=2,1,4`$ . For $`a=2,4`$, the Lax representation is standard: $`L_t=[4L_+^{3/2},L]`$ with
$$\begin{array}{cc}& L_{a=4}=(D_1D_2+\mathrm{\Phi })^2\hfill \\ & L_{a=2}=^2+\underset{i,j=1,2}{}ϵ_{ij}D_i(D_1D_2+\mathrm{\Phi })D_j\hfill \end{array}$$
$`(3.6)`$
(with $`ϵ_{12}=ϵ_{21}=1`$) while for $`a=1`$ it is nonstandard : $`L_t=[4L_1^3,L]`$ with
$$L_{a=1}=^1[(D_1D_2\mathrm{\Phi })(D_2\mathrm{\Phi })D_1(D_1\varphi )D_2+\mathrm{\Phi }D_1D_2]$$
$`(3.7)`$
In all cases, there is an infinite number of conservation laws given by
$$H_{2k+1}=𝑑d\theta _1𝑑\theta _2\mathrm{SRes}L^{(2k+1)/2}$$
$`(3.8)`$
where the $`N=2`$ version of the residue of a pseudodifferential operator is the coefficient of $`D_1D_2^1`$.
Although there exists a Miura transformation, there are no known integrable (i.e., Gardner-type) deformations of it. This leads us to:
OP-2: Find an integrable deformation for SKdV<sub>-2,1,4</sub> that reproduces their conservations laws.
For $`a=2,4`$, there are two independent towers of fermionic nonlocal conservations laws : in each case, the first one is
$$𝑑x𝑑\theta _1𝑑\theta _2(D_i^1\mathrm{\Phi })=𝑑x\xi _i(i=1,2)$$
$`(3.9)`$
However, although the Lax operator has two distinct fourth roots, these have not yet been related to these fermionic conservation laws:
OP-3: For SKdV<sub>-2,4</sub>, find the relation between the fermionic nonlocal conservation laws and the Lax operator.
The existence of these fermionic conservation laws is natural in that there are two supersymmetries. On the other hand, for the $`a=1`$ case, the first fermionic laws (3.9) do not generate infinite towers: they are isolated.
OP-4: Why there are no infinite towers of fermionic conservation laws for SKdV<sub>1</sub>?
The SKdV<sub>-2,4</sub> equations are both bihamiltonian: their first hamiltonian operator is
$$P_1^{(a=4)}=,P_1^{(a=2)}=(D_1D_2^1D_1^1\mathrm{\Phi }D_1^1D_2^1\mathrm{\Phi }D_2^1)D_1^1\mathrm{\Phi }D_1^1$$
$`(3.10)`$
In that respect, the SKdV<sub>1</sub> stands as one of the rare example of classical integrable system which is not (known to be) bihamiltonian.
OP-5: Is SKdV<sub>1</sub> bihamiltonian?
Another very natural question is:
OP-6: Why is there exactly three integrable $`N=2`$ super KdV extensions?<sup>14</sup> There are in fact also three distinct SKdV hiearchies but they generalize (in the Lax sens) the SKdV<sub>2,4</sub> equations; the SKdV<sub>1</sub> equation appears as a sort of isolated point. Interesting technical observations in relation with the bosonic truncation of the Lax operators are presented in . Is there an underlying Lie algebraic interpretation for this threefold way (i.e, is this related to the existence of the three classical algebras)?
4. Concluding questions
There are further general questions that could be formulated in relation with super integrable systems. As stressed here, these equations are naturally formulated in superspace. Over the years, it became clear that a lot of structure is contained in the Painlevé test: for instance its truncation leads to Backlünd transformations and the Lax operator . Yet, no Painlevé analysis has been done directly in superspace.
OP-7: Is it possible to formulate the Painlevé analysis in superspace?
Moreover, little is known concerning the solutions of super integrable systems. The Darboux transformation in the $`N=1`$ case has been worked out in but nothing has been done at this point concerning the $`N=2`$ cases.<sup>15</sup> Some solutions for the SKdV<sub>a</sub> equations have been reported in .
Finally, unravelling the deep relations between geometry and soliton theory has been an important theme of this workshop; little is known on the super version of this connection. In particular, the super KdV equations have super Sine-Gordon relatives and these should lead to very interesting geometrical structures.
Acknowledgement
I would like to thank Z. Popowicz for his helpful comments on the recent literature and the organisers of the workshop for their kind invitation to present this work.
REFERENCES
relax1.P. Mathieu, Supersymetric extension of the Korteweg-de Vries equation, J. Math. Phys. 29 (1988) 2499-2506; Superconformal algebra and supersymmetric Korteweg-de Vries equation, Phys. Lett. B203 (1988) 287-291.
relax2.Yu. I. Manin and A.O. Radul, A supersymmetric extension of the Kadomstev-Petviashvili hierarchy, Comm. Math, Phys. 98 (1985) 65-77.
relax3.C.A. Laberge and P. Mathieu, $`N=2`$ superconformal algebra and $`O(2)`$ integrable fermionic extensions of the Korteweg-de Vries equation, Phys. Lett. B215 (1988) 718-722; P. Labelle and P. Mathieu, A new N = 2 supersymmetric Korteweg-de Vries equation, J. Math. Phys. 32 (1991) 923-927.
relax4. Z. Popowicz, The Lax formulation of the ‘new’ $`N=2`$ SUSY KdV equation, Phys. Lett. 174 (1993) 411.
relax5.B.A. Kupershmidt and P. Mathieu, Quantum Korteveg-de Vries like equations and perturbed conformal field theories, Phys. Lett. B227 (1989) 245-250; P. Mathieu, Integrability of perturbed superconformal minimal models, Nucl. Phys. B336 (1990) 338-348; P. Mathieu and M.A. Walton, Integrable perturbations of N = 2 superconformal minimal models, Phys. Lett. B254 (1991) 106-110.
relax6.C.M. Yung, The $`N=3`$ supersymmetric KdV hierarchy, (1993); S. Krivonos, A. Pashnev and Z. Popowicz,Lax Pairs for $`N=2,3`$ SUSY KdV equations and their extensions, Mod. Phys. Lett. A13 (1998) 1435; F. Delduc, L. Gallot and E. Ivanov, New super KdV system with $`N=4`$ SCA as the hamiltonian structure, Phys.Lett. B 396 (1997) 122-132.
relax7.Z. Popowicz, Odd bihamiltonian structures of new supersymmetric $`N=2,4`$ Korteweg-de Vries equation and odd susy Virasoro-like algebra, Phys. Lett B459 (1999) 150-158.
relax8.B. Kupershmidt, A super Korteweg-de Vries equation: An integrable system, Phys. Lett. A 102 (1984) 213-215.
relax9.J.L. Gervais, Infinite familiy of plynomial functions of teh Virasoro generators with vanishing Poisson brackets, Phys. Lett. B160 (1985) 277-279; B.A. Kupershmidt, Super Korteweg-de Vries equations associated to super extensions of the Virasoro algebra, Phys. Lett. A109 (1985) 417-423.
relax10.P. Mathieu, Extended classical conformal algebras and the second Hamiltonian structure of Lax equations, Phys. Lett. B208 (1988) 101-106.
relax11.C.M. Yung, The $`N=2`$ supersymmetric Boussinesq hierarchies, Phys. Lett. B309 (1993) 75-84; Z. Popowicz, Lax formulation of the $`N=2`$ superBoussinesq equation, Phys. Lett. B319 (1993) 478-484; Extensions of the $`N=2`$ SUSY $`a=2`$ Boussinesq hierarchy, Phys.Lett A 236 (1997) 455-461; S. Belluci, E. Ivanov, S. Krivonos and A. Pichugin, $`N=2`$ super Boussinesq hierarchy: Lax pairs and conservation laws, Phys. Lett. B312 (1993) 463-470; F.Delduc and L.Gallot, $`N=2`$ KP and KdV hierarchies in extended superspace, Comm. Math. Phys. 190 (1997) 395-410.
relax12.K. Becker and M. Becker, Nonperturbative solution of the super-Virasoro constraints, Mod. Phy. Lett. A8 (1993) 1205-1214.
relax13.C.S. Gardner, Korteweg-de vries equation and generalizations. IV. The Korteweg-de vries equation as a hamiltonian system, J. Math Phys. 19 (1971) 1548-1551.
relax14.I. Yamanaka and R. Sasaki, Super Virasoro algebra and solvable supersymmetric quantum field theories, Prog. Theor. Phys. 79 (1988) 1167-1184.
relax15.W. Oevel and Z. Popowicz, The bi-hamiltonian structure of the fully supersymetric Korteweg-de vries equation, Comm. Math. Phys. 139 (1991) 441-460.
relax16.J.M. Figueroa-O’Farrill, J. Mas and E. Ramos, Bihamiltonian structure of the supersymmetric SKdV hierarchy, Rev. Mod. Phys. 3 (1991) 479.
relax17.P.H.M. Kersten, Higher order supersymmetries and fermionic conservation laws of the supersymmetric extension of the KdV and the mKdV equation, Phys. Lett. A134 (1988) 25-30.
relax18.P. Dargis and P. Mathieu, Nonlocal conservation laws for supersymetric KdV equations, Phys. Lett. A176 (1993) 67-74.
relax19.L. Bonora, S. Krivonos and A. Sorin, Toward the construction of $`N=2`$ supersymmetric iontegrable hierarchies, Nucl. Phys. B477 (1996) 835-854.
relax20.J. Weiss, M. Tabor and G. Carnavale, The Painlevé property for partial differential equations, J. Math. Phys. 24 (1983) 522-526; J. Weiss, in Painlevé transcendents, ed. D. Levi and P. Winternitz, Plenum Press 1992 and references therein.
relax21.Q.P. Liu, Darboux transformations for supersymmetric Korteweg-de Vries equations, Lett. Math. Phys. 35 (1995) 115-122; Q.P. Liu and M. Manas Darboux transformations for SUSY integrable systems,in Supersymmetry and integrable models, Lect. Notes in Phys. 35, eds H. Aratyn et al, Springer Verlag (1995), 269-281.
relax22.M.A. Ayari, V. Hussin and P. Winternitz, Group invariant solutions for the $`N=2`$ super Korteweg-de Vries equation, J.Math. Phys. 40 (1999) 1951-1965.
|
warning/0005/cond-mat0005293.html
|
ar5iv
|
text
|
# Edge and bulk effects in the Terahertz-photoconductivity of an antidot superlattice
## I Introduction
The magnetotransport properties of a two-dimensional electron gas (2DEG) in high magnetic fields are successfully described in the edge channel or Landauer-Büttiker picture . According to this model the spatial current distribution in a 2DEG in high magnetic fields is strongly influenced by the filling factor of the 2DEG. At the edges of a Hall bar the Landau levels (LLs) of the unbounded 2DEG are bent upwards in energy due to the confining potential. The intersections of these upwards bent Landau levels with the Fermi energy form the so-called edge states . Electrons in edge states at opposite edges of the Hall bar flow into opposite directions. At integer filling factors the Fermi energy lies in the regime of localized states between two Landau levels of the 2D-bulk, and a sufficiently small electric current is carried by the edge states only. In thermal equilibrium all edge states at the same sample edge have the same chemical potential. Scattering between edge states of the same sample edge (inter-LL scattering) is likely to occur after a distance which is called the equilibration length $`l_{eq}`$. However, this kind of scattering event does not lead to a nonzero magnetoresistance $`\rho _{xx}`$ along the current direction as it cannot reverse an electron’s direction of motion. Over distances up to the equilibration length a nonequilibrium population between edge states (of the same edge) can be maintained. The transport regime in which no equilibration between edge states (of the same edge) occurs is called the adiabatic regime. At noninteger filling factors the Fermi energy lies within the topmost partially filled Landau level. This topmost (further also referred to as Nth) Landau level can form a bulk conductance channel that allows backscattering of electrons, which leads to a nonzero longitudinal magnetoresistance $`\rho _{xx}`$ . The Nth channel is almost perfectly decoupled from the lower N$`1`$ edge states . This means that the edge states propagate unaffected by the bulk channel just as they would do for integer filling factors.
Here we study the photoconductivity of nanostructured 2DEGs caused by intraband absorption in the quantum Hall regime. Intraband photoconductivity of 2DEGs has successfully been described by Neppl et al. using a bolometric model, which was originally employed to explain the mechanism of photoconductivity caused by intersubband resonance in 2DEGs. Later the same model was also applied to cyclotron-resonance-induced photoconductivity in the quantum Hall regime and to photoconductivity of systems with reduced dimensionality . In cyclotron resonance, the absorption of radiation excites single electrons from the N$`1`$st (bulk) LL below the Fermi energy to the Nth LL above the Fermi energy. Within a bolometric model the excited electrons thermalize via electron-electron interactions and a new quasi-equilibrium corresponding to a higher electron temperature is established. As the longitudinal resistance $`\rho _{xx}`$ depends on temperature, the increased electron temperature results in a change of the longitudinal resistance, $`\mathrm{\Delta }\rho _{xx}`$, which is measured in a photoconductivity experiment.
In more recent photoconductivity experiments, an enhanced cyclotron resonance amplitude in the adiabatic transport regime close to a current injecting contact was observed. The longitudinal resistance in the adiabatic regime at low temperatures and bias currents under illumination by a far-infrared laser was observed to show a sharp maximum at the cyclotron resonance. The authors concluded that the cyclotron resonance absorption process induces a nonequilibrium population of edge states. They infer that this also leads to inter-edge scattering (i.e. to scattering events that are able to reverse the electrons’ direction of motion) and thus to an increase in longitudinal resistance. This increase in resistance can be observed directly in photoconductivity and in magnetotransport .
We investigate the photoconductivity properties of an antidot superlattice which allows us to excite the cyclotron resonance and the fundamental magnetoplasmon mode by illumination with THz-radiation. We find that at noninteger filling factors mainly resonant electron heating contributes to photoconductivity in both absorption processes. Under adiabatic transport conditions at a filling factor of $`\nu =2`$, on the contrary, only the cyclotron resonance (CR) is observed. Here, in addition, the CR amplitude has a different dependence on the applied bias current than the magnetoplasmon and the cyclotron resonance amplitudes in the dissipative transport regime. This indicates that the photoresponse in the adiabatic transport regime is generated by a process different from electron heating which is mainly responsible for the photosignal in the dissipative regime.
## II Sample details and experimental technique
The sample investigated here is a square antidot superlattice with a period of 500 nm, fully covering the distance of $`39\mu m`$ between adjacent voltage probes on a $`39\mu m`$ wide Hall bar (Fig. 1).
The antidot-superlattice is prepared on a GaAs/AlGaAs heterostructure with a mobility (at $`T=4.2`$ K) of $`\mu =1.5310^5cm^2/Vs`$ and an electron density $`n_s=3.4510^{11}cm^2`$ in the unpatterned 2DEG. The 2DEG has a distance of 37 nm from the sample surface. The antidots are written by e-beam lithography and transferred into the 2DEG by shallow wet etching. In the patterned region the density is smaller by about 7$`\%`$ compared to the unpatterned 2DEG.
The antidots have a triangular shape (width = 300 nm, height = 200 nm) . However, this is not essential for the results presented here.
We study the photoresponse, in particular the change in resistance $`\mathrm{\Delta }R_{xx}=\mathrm{\Delta }U_{xx}/I`$ in response to the THz-illumination, by measuring the photo-induced voltage $`\mathrm{\Delta }U_{xx}`$ between two voltage probes in four-point geometry as shown in Fig. 1. We apply dc bias currents $`I`$ of up to 10 $`\mu A`$ between source and drain contacts. The bath temperature during the measurements is 4.2 K, and magnetic fields of up to 12 T are applied perpendicular to the plane of the 2DEG.
We use a broad-band mercury lamp to irradiate our samples. The light source is modulated by a Fourier-transform-spectrometer to spectrally analyze the measured photoresponse $`\mathrm{\Delta }U_{xx}`$. In the spectral range of interest to us, between 0 and $`200cm^1`$ (0 to 6 THz), the Hg lamp has an integrated intensity of a few micro-Watts.
## III Experimental results
In Fig. 2(a) typical data in the dissipative regime at a magnetic field of $`8.4T`$ ($`\nu 1.6`$) and a bias current of $`1\mu A`$ applied from the source to the drain contact are shown. As can be seen there, we detect the cyclotron resonance (CR) and a magnetoplasmon resonance (MP). The photoresponse $`\mathrm{\Delta }U_{xx}`$ was recorded between contacts 3 and 4. Fig. 2(b) shows a spectrum taken at the reference section on the same Hall bar (contacts 1 and 2). There only the cyclotron resonance is resolved.
We record the photoresponse spectra for a series of magnetic fields between 0 and 12 T and fit the spectral resonance position of the magnetoplasmon as a function of the applied magnetic field by the formula $`\omega ^2=\omega _0^2+\omega _c^2`$ to extrapolate the zero-magnetic-field plasmon frequency $`\omega _0`$. This $`\omega _0`$ is compared to the calculated frequency $`\omega _P`$ of a plasmon in a two-dimensional electron gas with a two-dimensional modulation of the charge density with periods $`a_x`$ and $`a_y`$ along the $`x`$ and $`y`$ directions, respectively. We determine $`\omega _P`$ from sample-specific parameters via the formula
$$\omega _P^2=\frac{N_se^2}{2m^{}ϵ_{eff}(k)ϵ_0}\sqrt{(n_x\frac{2\pi }{a_x})^2+(n_y\frac{2\pi }{a_y})^2}=:\frac{N_se^2}{2m^{}ϵ_{eff}(k)ϵ_0}k,$$
(1)
where in our case $`a_x=a_y=a=500nm`$. $`N_s`$ is the 2D electron density, $`m^{}`$ the effective mass, $`k`$ the wave-vector as defined by the above formula and $`ϵ_{eff}`$ the effective dielectric constant. For $`ϵ_{eff}`$ we use $`ϵ_{eff}(k)=ϵ_{GaAs}/(1+\frac{ϵ1}{ϵ+1}e^{2kd})`$,where the distance $`d`$ of the 2DEG from the sample surface equals $`d=37nm`$ in our samples. The effective mass $`m^{}=0.07`$ is deduced from the frequency of the cyclotron resonance, measured in the reference section. We fit our data by the $`(n_x,n_y)=(1,0)`$ mode and obtain very good agreement between theory and experiment. Therefore, the resonance beside the cyclotron resonance is identified as the fundamental (1,0) or the degenerate (0,1) magnetoplasmon mode in the superlattice.
Figure 3(a) shows photoconductivity spectra taken at a magnetic field of B = 8.0 T (corresponding to a filling factor of $`\nu =1.7`$ of the 2DEG) for bias currents applied between source and drain contacts ranging from $`+10\mu A`$ to $`10\mu A`$ in steps of $`2\mu A`$. As expected, with no bias current applied no photoconductivity is observed (curve $`I=0`$). For bias currents of $`I=\pm 2\mu A`$ both the cyclotron resonance (CR) and the magnetoplasmon (MP) are observed in the photoresponse $`\mathrm{\Delta }U_{xx}`$ and have a large amplitude already. For higher bias currents up to $`I=\pm 10\mu A`$ the photoresponse amplitude in the cyclotron resonance slightly increases and finally decreases again. The photoresponse amplitude in the magnetoplasmon absorption, on the other hand, decreases monotonically.
Fig. 3(b) shows data analogous to those in Fig. 3(a), but for a magnetic field of $`B=6.8T`$, corresponding to a filling factor of $`\nu =2`$. Under these conditions, up to high bias-currents of $`I=\pm 8\mu A`$ only the cyclotron resonance is observed, but not the magnetoplasmon. Only for very high bias currents of $`I=\pm 10\mu A`$ a magnetoplasmon-induced photoresponse is beginning to develop. The amplitude of the cyclotron resonance increases linearly with increasing bias current.
For a filling factor of $`\nu =1.9`$ (B = 7.2 T) (not shown) both the cyclotron resonance and the magnetoplasmon are observed in the photoresponse, but the magnetoplasmon only for bias currents higher than about $`I=\pm 3\mu A`$.
Summarizing, at integer filling factor $`\nu =2`$ of the 2DEG only the cyclotron resonance is observed in photoresponse. At $`\nu <2`$ both the magnetoplasmon and the cyclotron resonance are observed in photoresponse.
Figure 4 shows the amplitudes of the spectra from Figs. 3(a) and 3(b) as a function of the bias current between source and drain. Open circles correspond to the cyclotron resonance maximum, black dots to the magnetoplasmon signal. The dashed interpolating lines are guides to the eye only. For comparison, the differential longitudinal resistance $`dU_{xx}/dI_{bias}`$ as a function of the bias current, recorded in the same geometry as the photoresponse, is plotted (solid line).
## IV Discussion
Figure 4 demonstrates that the amplitude of the magnetoplasmon signal qualitatively follows the differential longitudinal resistance $`dU_{xx}/dI_{bias}`$ for different magnetic fields (filling factors). It will be shown in the following that therefore the magnetoplasmon signal can be explained by heating of the electron gas corresponding to the bolometric model .
In the bolometric model the photoresponse of a sample is regarded as a change in the voltage (or resistance or conductivity, depending on whether a voltage, resistance or conductivity is measured in the specific sample geometry) caused by an increased electron temperature. In our case, the longitudinal voltage is measured in a four point geometry (Fig. 1). Thus the photoresponse corresponds, in terms of the bolometric model, to a temperature-induced change in the longitudinal voltage $`\mathrm{\Delta }U_{xx}=dU_{xx}/dT\mathrm{\Delta }T`$. The increase in electron temperature, $`\mathrm{\Delta }T`$, is provided by the resonant absorption of the THz radiation. Since the absorption coefficient does not strongly depend on magnetic field or filling factor and since filling factor-induced oscillations of the electronic specific heat are rather weak, we can for simplicity assume that $`\mathrm{\Delta }T`$ does not depend on magnetic field or filling factor in the present, small signal limit. Thus, the photoresponse is determined by the dependence $`dU_{xx}/dT`$ of the voltage U on temperature.
On the other hand, a sufficiently high bias current $`I`$ also produces an increase in electron temperature. Thus the differential $`dU_{xx}/dI`$, is directly proportional to $`dU_{xx}/dT`$.
The amplitude of the magnetoplasmon-induced photoresponse as a function of bias-current $`I`$ resembles very much the differential longitudinal resistance $`dU_{xx}/dI`$ and thus $`dU_{xx}/dT`$. Thus the bolometric model can describe the photoresponse mechanism at the magnetoplasmon resonance as heating of the electron gas.
A nonzero longitudinal voltage $`U_{xx}(T)`$ is, on the other hand, a rough measure for the amount of conduction through bulk, dissipative electronic states. Thus the appearence of the magnetoplasmon in photoconductivity can be taken for an indicator of beginning bulk conductivity.
For $`\nu =1.7`$ (B = 8 T) the amplitude of the cyclotron resonance has a behaviour similiar to that of the magnetoplasmon (Fig. 4(a)), even though the features occur at higher bias currents. For $`\nu =2`$ or B = 6.8 T (Fig. 4(c)), in contrast, the amplitude of the cyclotron resonance signal increases linearly with increasing bias current while the amplitude of the magnetoplasmon response and the longitudinal resistance (solid line) stay almost zero up to bias currents of $`I=\pm 8\mu A`$. The suppression of the magnetoplasmon and the vanishing longitudinal resistance indicate that the 2D-bulk is still insulating. Thus only the edge-states contribute to the conductance at an external magnetic field of B = 6.8 T ($`\nu =2`$). Therefore, we attribute the cyclotron resonance part of the photoresponse at $`\nu =2`$ to backscattering of the topmost edge states. At B = 8 T probably both mechanisms are important, as will be discussed below.
Summarizing the discussion, the magnetoplasmon absorption leads to a photoresponse only in the dissipative transport regime, when the 2D-bulk is conducting, which also manifests itself in a nonzero (differential) longitudinal magnetoresistance $`\rho _{xx}=dU_{xx}/dI`$. The cyclotron resonance absorption leads to a photoresponse also in the adiabatic transport regime, when only the edge-states contribute to electric conductance.
As for the mechanism of backscattering we follow the argumentation of Diessel et al. and propose that the absorption of THz radiation at the cyclotron resonance in the regime of the QHE at $`\nu 2`$ leads to a nonequilibrium population of edge states and thus to an enhanced inter-edge scattering rate, i.e. rate of backscattering from one edge of the Hall bar through the insulating bulk region to the other edge.
In the case of an antidot superlattice as studied here, not only the extended edge states at the boundaries of the hall bar exist, but also localized states around the antidots. We propose that backscattering in our superlattice takes place in several steps via the current loops localized around the antidots.
We will now briefly discuss our data for different other filling factors (magnetic fields). At a filling factor of $`\nu =1.9`$ (B = 7.2 T) (Fig. 4(b)) the cyclotron resonance behaves similiar to $`\nu =2`$ (B = 6.8 T) (Fig. 4c). Magnetoplasmon amplitude and $`\rho _{xx}`$ are both almost zero for small bias-currents below 2$`\mu `$A and then increase in a way very similiar to each other. This is explained as follows: for low bias-currents below $`2\mu A`$ the sample is in an adiabatic state of conductance and only the cyclotron resonance absorption produces a photoresponse. For higher bias-currents bulk transport is beginning to be thermally activated, such that the sample is in a dissipative state now. Correspondingly also the magnetoplasmon absorption generates a photoresponse. At a magnetic field of B = 6.2 T, still in the plateau region of $`\rho _{xx}`$ around integer filling factor $`\nu =2`$ (not shown), the data are very similiar to those for $`\nu =2`$ (B = 6.8 T) (Fig. 4c). The reason for this is, that – up to bias-currents of $`10\mu A`$ – the 2DEG is still in an adiabatic state ($`\rho _{xx}`$ is still vanishing). Thus only the CR absorption generates a photoresponse. For a filling factor of $`\nu =1.6`$ (B = 8.6 T) (not shown), where $`\rho _{xx}`$ is nonzero the data resemble those for $`\nu =1.7`$ (B = 8 T) (where $`\rho _{xx}0`$ as well) (Fig. 4(a)). This is because the 2DEG is in a dissipative state. Thus both CR and MP absorption lead to a photoresponse.
Now there remains to be discussed why the magnetoplasmon is visible in the photoresponse in the dissipative transport regime only, but the cyclotron resonance occurs both in the dissipative and the adiabatic transport regimes, i.e. why only CR absorption is able to induce backscattering of edge-states.
The extent of the edge-states into the 2D-bulk is comparable to the edge depletion length, which is of the order of a few 100 nm.
The characteristic coherence length $`l`$ of a collective excitation with a frequency/wave vector-dependence $`\omega (q)`$ is generally equal to $`l=(d\omega /dq)\tau `$ with the scattering time $`\tau `$. For the magnetoplasmon in our samples we have $`l10\mu m`$. This means that the magnetoplasmon excitation has its full strength only a distance $`l10\mu m`$ away from the sample boundaries, i.e. deep in the 2D-bulk. At the location of the edge states, not much more than 100 nm away from the boundaries, the magnetoplasmon has almost vanished. When only the edge-states contribute to electric conduction, as is the case under adiabatic transport conditions, the magnetoplasmon in the 2D-bulk cannot be converted into a photoresponse. This conversion could only be done by the edge-states, but at the location of the edge-states the magnetoplasmon oscillation strength is already negligible.
The characteristic length for the cyclotron resonance is the magnetic length. Even though the cyclotron resonance as a collective phenomenon is affected by finite-size effects up to macroscopic dimensions , it is sensitive to the local potential landscape even in the nm-range . Thus absorption in the cyclotron resonance can be converted into a photoresponse also by the edge states.
## V Conclusion
We have examined cyclotron resonance- and magnetoplasmon-induced changes in the longitudinal voltage of an antidot superlattice in high magnetic fields by means of photoconductivity measurements. In the dissipative transport regime we detect both magnetoplasmon and cyclotron resonances in the photoresponse. In the adiabatic transport regime at integer bulk filling factor $`\nu =2`$ we detect only the cyclotron resonance in the photoresponse. To explain these experimental results, i.e. the different dependencies of the amplitudes of the photoresponse in the cyclotron resonance and in the magnetoplasmon resonance on the applied bias current at different filling factors, we suggest a model, wherein different mechanisms generate photoconductivity in the magnetoplasmon resonance and in the cyclotron resonance, respectively. According to this model electron heating is responsible for photoconductivity in the dissipative transport regime. In the regime of adiabatic transport at integer filling factor $`\nu =2`$ the photoresponse is generated by backscattering of the topmost edge state, and this backscattering can be caused by the cyclotron resonance only.
The reason for why only the CR can cause backscattering of edge states is that cyclotron and magnetoplasmon resonances have different characteristic lengths. The characteristic length of the cyclotron resonance is comparable to the lateral extent of the edge-states, thus CR absorption can induce backscattering of edge-states. The characteristic length of the magnetoplasmon is by almost an order of magnitude larger than the lateral extent of the edge-states, thus MP absorption cannot induce backscattering of edge-states.
###### Acknowledgements.
We would like to acknowledge Alik Chaplik and Achim Wixforth for valuable discussions and the Deutsche Forschungsgemeinschaft for financial support.
|
warning/0005/hep-ph0005172.html
|
ar5iv
|
text
|
# Spin structure and longitudinal polarization of hyperon in 𝑒⁺𝑒⁻ annihilation at high energies
## I Introduction
Spin effects in high energy fragmentation processes have attracted much attention recently. Study of such effects provide useful information for the spin structure of hadron and spin dependence of high energy reactions. There exist now two distinctively different pictures for the spin contents of the baryons: the static quark model picture using SU(6) symmetric wave function \[hereafter referred as SU(6) picture\], and the picture drawn from the data for polarized deep inelastic lepton-nucleon scattering (DIS) and SU(3) flavor symmetry in hyperon decay \[hereafter referred as DIS picture\]. It is natural to ask which picture is suitable to describe the relationship between the polarization of the fragmenting quark and that of the produced hadron. It has been pointed out in \[References\] that measurements of $`\mathrm{\Lambda }`$ longitudinal polarization in $`e^+e^{}`$ annihilation at high energies can provide useful information to answer this question. Calculations based on these two different pictures have been made in \[References\] and \[References\]. A comparison of the obtained results with the ALEPH data obtained at LEP, which was available at that time, was made. The result shows that the SU(6) picture seems to agree better with the data. This is rather surprising since the energy is very high at LEP and the initial quarks and anti-quarks produced at the annihilation vertices of the initial $`e^+e^{}`$ are certainly current quarks and current anti-quarks rather than the constituent quarks used in describing the static properties of hadrons using SU(6) symmetric wave functions. Other models are also proposed which can give a description of the available data. The available data are certainly still far from accurate and abundant enough to make a conclusive judgment of these different models. It is therefore important and urgent to make further checks of this conclusion by performing complementary measurements.
In this paper, we formulate the calculation method of longitudinal polarization of different hyperons in $`e^+e^{}`$ annihilation at high energies proposed in \[References\] and \[References\] in a systematic way so that they can be generalized to other hyperons and/or other reactions. We describe in detail the inputs and/or assumptions which have been used in the calculations. We then present the calculated results for different hyperons based on these two pictures. We present the results for the whole events as well as those for different particularly chosen event samples. We show that corresponding measurements can be used as complementary method to check the above mentioned conclusion. In particular, we discuss also how to distinguish such models from the kind of models proposed in \[References,References\]. In section 2, we outline the calculation method and present the results for $`\mathrm{\Lambda }`$ polarization. In section 3, we present our results for other hyperons.
## II The calculation method and the results for $`\mathrm{\Lambda }`$ in different event samples
We recall that, according to the standard model of electroweak interactions, quarks and anti-quarks produced at the annihilation vertices of $`e^+e^{}`$ are longitudinally polarized. Such longitudinal quark polarization can be transferred to the produced hadron which contains the initial quark thus lead to longitudinal polarization of hadron in the inclusive process $`e^+e^{}h+X`$. Measuring the polarizations of the produced hadrons, we can study the relation between the spin of the fragmenting quark and that of the produced hadron which contains that quark. In this aspect, the $`J^P=\frac{1}{2}^+`$ hyperons are most suitable candidates among all the hadrons. This is because these hyperons decay weakly so that their polarizations can easily be obtained from the angular distributions of their decay products. We now outline the calculation method of the longitudinal polarization $`P_{H_i}`$ of different hyperon $`H_i`$ in the inclusive process $`e^+e^{}H_i+X`$.
### II.1 The calculation method for different hyperons
Since the longitudinal polarization $`P_{H_i}`$ of the hyperon $`H_i`$ in the inclusive process $`e^+e^{}H_i+X`$ originates from the longitudinal polarization $`P_f`$ of the initial quark $`q_f^0`$ (where the subscript $`f`$ denotes its flavor) produced at the annihilation vertex of the initial state $`e^+e^{}`$, we should consider the $`H_i`$’s which have the following different origins separately.
(a) Hyperons which are directly produced and contain the initial quarks $`q_f^0`$’s originated from the annihilations of the initial $`e^+`$ and $`e^{}`$;
(b) Hyperons which are decay products of other heavier hyperons which were polarized before their decay;
(c) Hyperons which are directly produced but do not contain any initial quark $`q_f^0`$ from $`e^+e^{}`$ annihilation;
(d) Hyperons which are decay products of other heavier hyperons which were unpolarized before their decay.
It is clear that hyperons in groups (a) and (b) can be polarized. Those in group (a) are polarized since the polarization of the initial quark $`q_f^0`$ can be transferred to such hyperon $`H_i`$’s in the fragmentation process. We denote the probability for this polarization transfer by $`t_{H_i,f}^F`$ and call it polarization transfer factor from quark $`q_f`$ to hyperon $`H_i`$, where the superscript $`F`$ stands for fragmentation. Hyperons in group (b) can be polarized since such hyperons can inherit part of the polarization of the parent hyperons in the decay process. We denote the probability for this polarization transfer by $`t_{H_i,H_j}^D`$ and call it polarization transfer factor from hyperon $`H_j`$ to hyperon $`H_i`$, where the superscript $`D`$ stands for decay. In contrast, hyperons in groups (c) and (d) are unpolarized. Hence, we obtain that the polarization of the final hyperon is,
$$P_{H_i}=\frac{\underset{f}{}t_{H_i,f}^FP_fn_{H_i,f}^a+\underset{j}{}t_{H_i,H_j}^DP_{H_j}n_{H_i,H_j}^b}{n_{H_i}^a+n_{H_i}^b+n_{H_i}^c+n_{H_i}^d}.$$
(1)
Here $`P_f`$ is the polarization of the initial quark $`q_f^0`$; $`n_{H_i,f}^a`$ is the average number of the hyperons which are directly produced and contain the initial quark of flavor $`f`$; $`P_{H_j}`$ is the polarization of the hyperon $`H_j`$ before its decay; $`n_{H_i,H_j}^b`$ is average numbers of $`H_i`$ hyperons coming from the decay of $`H_j`$ hyperons which are polarized; $`n_{H_i}^a(\underset{f}{}n_{H_i,f}^a`$), $`n_{H_i}^b(\underset{j}{}n_{H_i,H_j}^b)`$, $`n_{H_i}^c`$ and $`n_{H_i}^d`$ are average numbers of hyperons in group (a), (b), (c) and (d) respectively. These different factors can be calculated in the following way.
Longitudinal polarization of the initial quark $`q_f^0`$ is determined by the standard model for electroweak interactions. The polarization comes from the weak interactions and the results for $`e^+e^{}\gamma ^{}/Zq_f^0\overline{q}_f^0`$ can be found e.g. in \[References\], i.e.,
$$P_f=\frac{A_f(1+\mathrm{cos}^2\theta )+B_f\mathrm{cos}\theta }{C_f(1+\mathrm{cos}^2\theta )+D_f\mathrm{cos}\theta },$$
(2)
where $`\theta `$ is the angle between the outgoing quark and the incoming electron, the subscript $`f`$ denotes the flavor of the quark, and
$$A_f=2a_fb_f(a^2+b^2)2(1\frac{m_Z^2}{s})Q_fab_f,$$
(3)
$$B_f=4ab(a_f^2+b_f^2)2(1\frac{m_Z^2}{s})Q_fa_fb,$$
(4)
$$C_f=\frac{(sm_Z^2)^2+m_Z^2\mathrm{\Gamma }_Z^2}{s^2}Q_f^2+(a^2+b^2)(a_f^2+b_f^2)2(1\frac{m_Z^2}{s})Q_faa_f,$$
(5)
$$D_f=8aba_fb_f4(1\frac{m_Z^2}{s})Q_fbb_f,$$
(6)
where $`m_Z`$ and $`\mathrm{\Gamma }_Z`$ are the mass and decay width of $`Z`$;
$$a=\frac{1+4\mathrm{sin}^2\theta _W}{2\mathrm{sin}2\theta _W},$$
(7)
$$b=\frac{1}{2\mathrm{sin}2\theta _W},$$
(8)
$$a_f=\{\begin{array}{cc}\hfill \frac{18\mathrm{sin}^2\theta _W/3}{2\mathrm{sin}2\theta _W},& \text{for }f=u,c,t,\hfill \\ \hfill \frac{1+4\mathrm{sin}^2\theta _W/3}{2\mathrm{sin}2\theta _W},& \text{for }f=d,s,b,\hfill \end{array}$$
(9)
$$b_f=\{\begin{array}{cc}\hfill \frac{1}{2\mathrm{sin}2\theta _W},& \text{for }f=u,c,t,\hfill \\ \hfill \frac{1}{2\mathrm{sin}2\theta _W},& \text{for }f=d,s,b,\hfill \end{array}$$
(10)
are the vector and axial vector coupling constants of electron and quark to $`Z`$ boson, and $`\theta _W`$ is the Weinberg angle. To see the size and the $`\theta `$-dependence of $`P_f`$, we show in Fig.1 the numerical results of $`P_f`$ as a function of $`\mathrm{cos}\theta `$ at LEP I energy. We see clearly that all the quarks produced at the $`e^+e^{}`$ annihilation vertices are longitudinally with significantly high polarizations. We see also that the magnitude of the polarization of the down-type quarks ($`d,s`$ and $`b`$) is large and varies little with the angle $`\theta `$, while that of the up-type quarks ($`u`$ and $`c`$) is a little bit smaller and has a relatively larger variation in the whole range of $`\mathrm{cos}\theta `$. The latter increases with $`\mathrm{cos}\theta `$ from 0.58 at $`\mathrm{cos}\theta =1`$ to 0.74 at $`\mathrm{cos}\theta =1`$. Averaging over $`\theta `$, we obtain,
$$P_f=\frac{P_f\sigma ^0d\mathrm{cos}\theta }{\sigma ^0d\mathrm{cos}\theta }=\frac{A_f}{C_f},$$
(11)
where $`\sigma ^0=C_f(1+\mathrm{cos}^2\theta )+D_f\mathrm{cos}\theta `$ is the angular variation of the unpolarized cross section. In Fig.2, we show $`P_f`$ as a function of the total $`e^+e^{}`$ center-of-mass energy $`\sqrt{s}`$. We see that, at the LEP I energy, i.e. $`\sqrt{s}=91GeV`$, the polarizations of quarks have the maximum negative value, i.e., $`P_f=0.67`$ for $`f=u,c`$, and $`P_f=0.94`$ for $`f=d,s,`$ and $`b`$. From the LEP I energy to the LEP II energy, the polarizations decrease a bit but it is still very large at the LEP II energy ($`\sqrt{s}=200GeV`$), where we have $`P_f=0.26`$ for $`f=u,c`$, and $`P_f=0.8`$ for $`f=d,s,`$ and $`b`$.
At LEP II energy, the $`W^+W^{}`$ events, i.e. $`e^+e^{}W^+W^{}q_{f1}\overline{q}_{f2}q_{f3}\overline{q}_{f4}`$, are significant. It takes about 10% of the whole events. The polarizations for the initial quarks in such events are different from those in $`e^+e^{}\gamma ^{}/Z^0q_f\overline{q}_f`$, and should be considered separately in the calculations. According to the standard model for electroweak interactions, the polarization of initial quark and that of initial anti-quark created in $`W^+`$ or $`W^{}`$ decay vertex are equal to -1.0 and 1.0, respectively. There exists also another type of events, i.e. $`e^+e^{}Z^0Z^0q_{f1}\overline{q}_{f1}q_{f2}\overline{q}_{f2}`$, which contribute at LEP II energy. But this contribution is very small (less than 1%). We will not consider such events in the following calculations.
The fragmentation polarization transfer factor $`t_{H_i,f}^F`$ from the initial quark $`q_f`$ to hyperon $`H_i`$ is equal to the fraction of spin carried by the $`f`$-flavor-quark divided by the average number of quark of flavor $`f`$ in the hyperon $`H_i`$. This fractional contribution to the hyperon spin from $`f`$-flavor-quark is different in the above-mentioned SU(6) or the DIS picture. The results in the SU(6) picture can easily be obtained from the wave functions. In the DIS picture, the fractional contribution of quarks of different flavors to the spin of a baryon in the $`J^P=\frac{1}{2}^+`$ octet is extracted from $`\mathrm{\Gamma }_1^p_0^1g_1^p(x)𝑑x`$ obtained in deep-inelastic lepton-proton scattering experiments and the constants $`F`$ and $`D`$ obtained from hyperon decay experiments. The way of doing this extraction is now in fact quite standard (see, for example, the Appendix in \[References\]). The results for $`\mathrm{\Lambda }`$, $`\mathrm{\Sigma }^0`$ and $`\mathrm{\Xi }`$ hyperons have been obtained in \[References\] and \[References\]. Using exactly the same way, we obtain also the results for $`\mathrm{\Sigma }^\pm `$. For completeness, we list the results for all the $`J^P=\frac{1}{2}^+`$ hyperons in Table 1.
In the calculations, we take also the contributions from the decay of $`J^P=\frac{3}{2}^+`$, i.e. the decuplet hyperons into account. The polarizations of such hyperons in the SU(6) picture can easily be calculated using the SU(6) symmetric wave functions. But, it is presently impossible to calculate them in the DIS picture since no DIS data is available for any one of the decuplet baryons. Therefore, in the calculation we take them into account in the same way as those in the SU(6) picture.
The decay polarization transfer factor $`t_{H_i,H_j}^D`$ is determined by the decay process and is independent of the process in which $`H_j`$ is produced. The results for different decay processes are different and will be given for each individual hyperon in the following sections of this paper.
After we obtain the results for $`P_f`$, $`t_{H_i,f}^F`$ and $`t_{H_i,H_j}^D`$, we can calculate $`P_{H_i}`$ if we know the average numbers $`n_{H_i,f}^a`$, $`n_{H_i,H_j}^b`$, $`n_{H_i}^c`$, and $`n_{H_i}^d`$ for hyperons $`H_i`$ from the different origins. These average numbers are determined by the hadronization mechanism and should be independent of the polarization of the initial quarks. Hence, we can calculate them using the a hadronization model which give a good description of the unpolarized data for multiparticle production in high energy reactions. Presently, such calculations can only be carried out using a Monte-Carlo event generator. We use the Lund string fragmentation model implemented by JETSET in the following.
### II.2 $`\mathrm{\Lambda }`$ polarization in the average events
Among all the $`J^P=\frac{1}{2}^+`$ hyperons, $`\mathrm{\Lambda }`$ is most copiously produced. Furthermore, the spin structure of $`\mathrm{\Lambda }`$ in the $`SU(6)`$ picture is very special, which makes it play a very special role in distinguishing the SU(6) and the DIS pictures. In the $`SU(6)`$ picture, spin of $`\mathrm{\Lambda }`$ is completely carried by the $`s`$ valence quark, while the $`u`$ and $`d`$ quarks have no contribution. Since the initial $`s`$ quark produced in the annihilation of the initial $`e^+e^{}`$ takes the maximum negative polarization, $`|P_\mathrm{\Lambda }|`$ obtained using the SU(6) picture is the maximum among all the different models. In contrast, in the DIS picture, the $`s`$ quark carries only about $`60\%`$ of the $`\mathrm{\Lambda }`$ spin, while the $`u`$ or $`d`$ quark each carries about $`20\%`$ (see Table 1). The resulting $`|P_\mathrm{\Lambda }|`$ should be substantially smaller than that obtained in the $`SU(6)`$ picture. Comparing the maximum with experimental results provide us a good test of the validity of the picture.
$`\mathrm{\Lambda }`$ is also the lightest hyperon, so the final $`\mathrm{\Lambda }`$’s produced in high energy reactions contain contributions from the decays of many different heavier hyperons. There are three octet, i.e. $`\mathrm{\Sigma }^0`$, $`\mathrm{\Xi }^0`$ and $`\mathrm{\Xi }^{}`$, and five decuplet, i.e., $`\mathrm{\Sigma }(1385)^{\pm ,0}`$ and $`\mathrm{\Xi }(1530)^{,0}`$, hyperons that can decay to $`\mathrm{\Lambda }`$. The polarization transfer in $`\mathrm{\Sigma }^0\mathrm{\Lambda }\gamma `$ has been studied in \[References\], the result for the polarization transfer factor $`t_{\mathrm{\Lambda },\mathrm{\Sigma }^0}^D`$ is $`1/3`$. $`\mathrm{\Xi }\mathrm{\Lambda }\pi `$ is a parity non-conserving decay and is dominated by S-wave. The polarization transfer $`t_{\mathrm{\Lambda },\mathrm{\Xi }}^D`$ for this process is equal to $`(1+2\gamma )/3`$, where $`\gamma =0.87`$ can be found in Review of Particle Properties \[References\]. The decay process from the decuplet hyperon to octet hyperon and a pseudo-scalar meson $`\pi `$ such as $`\mathrm{\Sigma }(1385)\mathrm{\Lambda }\pi `$ or $`\mathrm{\Xi }(1530)\mathrm{\Xi }\pi `$ is dominated by the $`P`$ wave, and the octet hyperon will get the same polarization as that of the initial decuplet hyperon, i.e., $`t^D=1`$. For explicity, we list all these results for $`t_{\mathrm{\Lambda },H_j}^D`$ in table.2. There are also some contributions from the decays of open charm or open beauty baryons. The polarization transfer factors in these decay processes are unfortunately unknown yet. This is a theoretical uncertainty in the calculations. However, according to the materials in Review of Particle Properties, we do know that each of these open charm or open beauty baryons can decay to $`\mathrm{\Lambda }`$ through many different channels. The contributions to $`\mathrm{\Lambda }`$ polarization in these different channels may be quite different. We expect that the net contribution from all these channels cannot be large. We will just neglect it in our calculations.
Using JETSET and PYTHIA, we obtain the average numbers $`n_{\mathrm{\Lambda },f}^a`$, $`n_{\mathrm{\Lambda },H_j}^b`$, $`n_\mathrm{\Lambda }^c`$, and $`n_\mathrm{\Lambda }^d`$ at $`\sqrt{s}=91`$ GeV and those at $`\sqrt{s}=`$200 GeV,respectively. We show them in Fig.3 and Fig.4. The results at $`\sqrt{s}=91`$ GeV are of course the same as those in \[References\] and \[References\]. From these figures, we see clearly that contribution to $`\mathrm{\Lambda }`$ from the events originating from the initial $`s`$ quarks play the most important role, in particular for large $`z`$. For example, for $`z>0.4`$, it gives about 70% of the whole $`\mathrm{\Lambda }`$’s, while those from the events initiated by $`u`$ or $`d`$ take only 10% respectively.
Using the results for the $`P_f`$’s, the $`t_{H_i,f}^D`$’s, and the $`t_{H_i,H_j}^D`$’s mentioned above, we obtain the longitudinal polarization of $`\mathrm{\Lambda }`$ as shown in Fig.5. A comparison of those results at LEP I with the ALEPH data and the OPAL data shows that the data of both groups agree better with the calculated results based on the $`SU(6)`$ picture. But, these available data are still far from accurate and enormous enough to make a decisive conclusion. Further complementary measurements are needed.
At the LEP II energy, $`W^+W^{}`$ events take about 10% of the whole events. We expect that such events give a even more significant contribution to $`\mathrm{\Lambda }`$ polarization, since the polarization of the initial quark at $`W^+`$ or $`W^{}`$ decay vertex is 100%, which is larger than those from the $`e^+e^{}\gamma ^{}/Z^0q_f^0\overline{q}_f^0`$. Since there are two initial quarks and two initial anti-quarks in $`W^+W^{}`$ events, the energy of each of them should be much smaller than those in the $`e^+e^{}\gamma ^{}/Z^0q_f^0\overline{q}_f^0`$ events. Hence, such events should contribute mainly in relatively smaller $`z`$ regions. Adding all the different together, we obtain $`P_\mathrm{\Lambda }`$ at LEP II energy shown in Fig.5. We see that, compared with those obtained at the LEP I energy, $`|P_\mathrm{\Lambda }|`$ at LEP II energy is a little larger at small $`z`$ region, and is a little smaller at large $`z`$ region. This is consistent with the above-mentioned qualitative expectations and can be checked experimentally.
### II.3 $`\mathrm{\Lambda }`$ polarization in different subsamples of events
In the two models we discussed above, $`\mathrm{\Lambda }`$ polarization comes solely from the hyperons which contain the initial quark created at the $`e^+e^{}`$ annihilation vertex. The contributions are very much different for the hyperons which contain the initial $`s`$ from those which contain the initial $`u`$ or $`d`$ quarks. There is also a big differences between the contribution from the hyperons which contain the initial quarks and that from those which do not. It is thus clear that we can get a further check of the two pictures if we study the $`\mathrm{\Lambda }`$ polarization in events which originate from the initial $`u`$ or $`d`$ quark or those which originate from the initial $`s`$ quark separately. We should also get a very sensitive check to different pictures if we study only those $`\mathrm{\Lambda }`$’s which contain the initial quarks.
In this connection, it should also be mentioned that there exist another different type of models for spin transfer in fragmentation processes. In contrast to that described in subsection II.1, in these models, no distinction is made between the hyperons which contain the initial quarks or those which do not contain the initial quarks. Some of them even do not distinct those which are directly produced or those which come from heavier hyperon decays. In other words, in these models, there is no distinction between hyperons from the groups (a), (b), (c) and (d). It is simply assumed that a “reciprocity relation” is valid between the fragmentation function for the final hyperons and the corresponding quark distribution functions in the hyperons. They are taken as proportional to each other, with a proportional constant which is common for quarks of different flavors. The hyperon polarization can easily be calculated in such models if the spin-dependent quark distribution functions are known. The obtained results depend obviously very much on the spin-dependent quark distributions, which are used as theoretical input. Since the spin dependent quark distributions in hyperons are still poorly known yet, the obtained results can be very much different from each other if different sets of quark distributions are used. But, since the (polarized) fragmentation function from a given quark $`q_f^0`$ to a given hyperon $`H_i`$ depends only on the fractional momentum $`z`$ of the quark carried by the produced hyperon, the obtained polarization at a given $`z`$ should be completely the same if we study only those hyperons which contain the initial quarks or all of them in events originating from a given type of initial quark.
It is unfortunately impossible to make a complete separation of hyperons in these different groups in experiments. However, it should be possible to separate the events into different subsamples, in each of which the hyperons from one group dominate. Measuring $`\mathrm{\Lambda }`$ polarization in such subsamples of events should give further complementary checks of the different models mentioned above. Using the Monte-Carlo event generator JETSET , we can study the various possibilities in this direction. We note that, if a $`\mathrm{\Lambda }`$ originates from the initial $`s`$ quark, the leading particle in the opposite direction should contain the $`\overline{s}`$ produced in the $`e^+e^{}`$ annihilation vertex. Hence, we choose an event sample according to the following criteria:
(i) $`\mathrm{\Lambda }`$ is the leading in one direction;
(ii) the leading particle in the opposite direction is $`K^+`$.
We expect that such $`\mathrm{\Lambda }`$’s should mainly have the origin (a) mentioned in the subsection II.1. In Fig.6, we show the results obtained from JETSET for the average numbers of $`\mathrm{\Lambda }`$’s of different origins. We see that such leading $`\mathrm{\Lambda }`$’s indeed mainly originate from the initial $`s`$ quark. The contribution from the initial $`u`$ or $`d`$ quark is indeed substantially small. It takes only about 3% of such $`\mathrm{\Lambda }`$’s. In particular, in $`z>0.5`$ region there is almost no contribution from the initial $`u`$ or $`d`$ quark at all.
Using Eq.(1), we calculated $`P_\mathrm{\Lambda }`$ for the events under the conditions (i) and (ii) using the SU(6) and the DIS pictures. The comparison of the obtained results with those obtained in the average events are shown in Fig.7. We see that there is indeed a significant difference between the results obtained for such particularly chosen events and those for the average events in particular in the region of $`z>0.3`$. Checking such differences by measuring $`P_\mathrm{\Lambda }`$ for such events can be helpful in distinguishing the validity of different pictures for the spin transfer in fragmentation processes.
## III Longitudinal polarization of other $`J^P=\frac{1}{2}^+`$ hyperons
The production rates for other octet hyperons are smaller than that for $`\mathrm{\Lambda }`$ so the statistic errors should be larger for the polarizations of these hyperons. On the other hand, decay contributions from heavier hyperons to these hyperons are also much less significant than that in case of $`\mathrm{\Lambda }`$. Hence, the contaminations from the decay processes are much smaller. These conclusions can easily be checked using a Monte-Carlo event generator for $`e^+e^{}`$ annihilation into hadrons. In Fig.8, we show the results obtained from JETSET for these hyperons compared with those for $`\mathrm{\Lambda }`$. We see that, the production rate for $`\mathrm{\Sigma }^+`$ or $`\mathrm{\Xi }^0`$ is less than $`20\%`$ of that for $`\mathrm{\Lambda }`$. On the other hand, we see also that the contribution from heavier hyperon decays is also much smaller. For example, for $`\mathrm{\Sigma }^+`$’s, the decay contribution takes only about $`7\%`$ of the total rate. The situations for $`\mathrm{\Sigma }^{}`$, $`\mathrm{\Xi }^0`$, and $`\mathrm{\Xi }^{}`$ are similar to that for $`\mathrm{\Sigma }^+`$. Most of them are directly produced. Hence, the theoretical uncertainties in the calculations for these hyperons are much smaller. The study of polarizations of these hyperons should provide us with good complementary tests of different pictures. In this section, we calculate the longitudinal polarizations of these hyperons, i.e., $`\mathrm{\Sigma }^+`$, $`\mathrm{\Sigma }^{}`$, $`\mathrm{\Xi }^0`$ and $`\mathrm{\Xi }^{}`$, in $`e^+e^{}`$ annihilation at LEP I and LEP II energies.
The calculations are quite similar to those for $`\mathrm{\Lambda }`$ production. The general procedure is the same and has been outlined in last section. The polarization transfer factor $`t_{H_i,f}^F`$ are given in table 1. The situation for decay contributions is quite simple: There are only contribution from the corresponding decuplet hyperon decays. More precisely, $`\mathrm{\Sigma }`$ contains only decay contribution from $`\mathrm{\Sigma }(1385)`$ and $`\mathrm{\Xi }`$ has the decay contribution from $`\mathrm{\Xi }(1530)`$. Both decay processes are strong decays and are dominated by the P-wave. The polarization transfer is the same as that in $`\mathrm{\Sigma }(1385)\mathrm{\Lambda }\pi `$, i.e. $`t_{\mathrm{\Sigma },\mathrm{\Sigma }(1385)}^D=t_{\mathrm{\Xi },\mathrm{\Xi }(1530)}^D=1`$.
Before we present the numerical results of the calculations, we would like to note the following qualitative expectations.
First, although the spin structure of $`\mathrm{\Sigma }^+`$ and that of $`\mathrm{\Sigma }^{}`$ or that of $`\mathrm{\Xi }^0`$ and that of $`\mathrm{\Xi }^{}`$ are symmetric under the exchange of $`u`$ and $`d`$ (c.f. Table 1), their polarizations should be quite different from each other. This is because the polarization of the initial $`u`$ and that of the initial $`d`$ produced at the $`e^+e^{}`$ annihilation vertices are quite different. This can be seen clearly in Fig.1, where we see that the magnitude of the polarization of $`u`$ quark is smaller than that of $`d`$ quark.
Second, from Table 1, we see that the contributions from the two different flavors in these hyperons ($`\mathrm{\Sigma }^{+,}`$, $`\mathrm{\Xi }^{0,}`$) are quite different: they have different signs and different magnitudes, and the differences are also different in the SU(6) or the DIS picture. This implies that their contributions to hyperon polarizations are also quite different. They should have different signs and different magnitudes. These differences make the situation very interesting. We should have different expectations for different hyperons. Even the signs of polarizations of some hyperons in the two different pictures can be different from each other.
Using JETSET and PYTHIA, we calculated the different contributions to $`\mathrm{\Sigma }^+`$, $`\mathrm{\Sigma }^{}`$, $`\mathrm{\Xi }^0`$ and $`\mathrm{\Xi }^{}`$ from all the different sources discussed above at LEP I and LEP II, respectively. The results at LEP I energy are shown in Fig.9. Those at LEP II energy are similar and are not shown. From this Fig.9, we see that the $`\mathrm{\Sigma }^+`$’s mainly come from the initial $`s`$ and the initial $`u`$-quark, and that the $`\mathrm{\Sigma }^{}`$’s come mainly from the initial $`s`$ and the initial $`d`$-quark, and that these two contributions are comparable with each other in particular for $`\mathrm{\Sigma }^{}`$’s. In contrast, the source for $`\mathrm{\Xi }^{0,}`$ is very pure: They come predominately from the initial $`s`$-quarks. All the others are negligible.
Using Eq.(1) and the results shown in Fig.9, we calculated the hyperon polarizations, $`P_{\mathrm{\Sigma }^+}`$, $`P_\mathrm{\Sigma }^{}`$, $`P_{\mathrm{\Xi }^0}`$ and $`P_\mathrm{\Xi }^{}`$. The obtained results are shown in Fig.10. From these results, we see clearly that the polarizations are different for these different hyperons, and that there are considerably large differences between the results using the $`SU(6)`$ picture and those using the DIS picture. In particular, we see that $`P_{\mathrm{\Sigma }^+}`$ in the $`SU(6)`$ picture has opposite sign to that in the DIS picture. But their magnitudes are relatively small. The magnitude of $`P_{\mathrm{\Xi }^0}`$ and that of $`P_\mathrm{\Xi }^{}`$ are considerably larger, in particular in the large $`z`$ region. This is because the contribution to $`\mathrm{\Xi }^0`$ or $`\mathrm{\Xi }^{}`$ is very clear. They come mainly from the initial $`s`$-quark. There is thus little theoretical uncertainty in calculating $`P_\mathrm{\Xi }`$. Hence, although the statistics may be much worse than $`\mathrm{\Lambda }`$, there are still good reasons to study $`P_\mathrm{\Xi }`$ in $`e^+e^{}\mathrm{\Xi }+X`$ at LEP energies.
We thank Gösta Gustafson, Li Shi-yuan, Wang Qun, Xie Qu-bing and other members in the theoretical particle physics group of Shandong University for helpful discussions. This work was supported in part by the National Science Foundation of China (NSFC) and the Education Ministry of China.
|
warning/0005/hep-th0005191.html
|
ar5iv
|
text
|
# Chapter 1 Pauli Terms Must Be AbsentIn Dirac Equation
## Chapter 1 Pauli Terms Must Be AbsentIn Dirac Equation
Kurt Just and James Thevenot
footnotetext: AMS Subject Classification: 81E10, 81Q05, 81S05.
## Abstract
It should be of interest, whether Dirac’s equation involves all 16 basis elements of his Clifford algebra $`Cl_D.`$ These include the 6 ‘tensorial’ $`\sigma ^{\mu \nu }`$ with which the ‘Pauli terms’ are formed. We find that these violate a basic axiom of any \*-algebra, when Dirac’s $`\mathrm{\Psi }`$ is canonical. Then the Dirac operator is spanned only by the 10 elements $`1,i\gamma _5,\gamma ^\mu ,\gamma ^\mu \gamma _5`$ (which don’t form a basis of $`Cl_D`$ because the $`\sigma ^{\mu \nu }`$ are excluded).
Keywords: Quantum field theory, Dirac equation, Clifford algebra.
### 1 Motivation and conclusions
In Dirac’s equation
$$i/\mathrm{\Psi }=\mathrm{\Psi }\text{with}/:=\gamma ^\mu \frac{}{x^\mu }$$
(1.1)
the Bose field $``$ is a member of the Clifford algebra $`Cl_D.`$ Hence it can be written as
$$=𝒮^++i\gamma _5𝒮^{}+\gamma ^\mu 𝒱_\mu ^++\gamma ^\mu \gamma _5𝒱_\mu ^{}+\sigma ^{\mu \nu }𝒯_{\mu \nu }.$$
(1.2)
Here $`𝒮^\pm ,𝒱^\pm ,𝒯_{\mu \nu }`$ are matrices which act on the flavors and colors of $`\mathrm{\Psi }`$ (the Dirac field for leptons and quarks). In the excellently verified Standard Model, the matrices
$$𝒮:=𝒮^++i\gamma _5𝒮^{}\text{and}𝒱_\mu :=𝒱_\mu ^++\gamma _5𝒱_\mu ^{}$$
(1.3)
contain all the fields of Higgs and Yang-Mills. Vice versa, the Standard Model requires (1.2) to contain the 10 basis elements $`1,i\gamma _5,\gamma ^\mu ,\gamma ^\mu \gamma _5Cl_D,`$ but not the further $`\sigma ^{\mu \nu }`$ or $`\sigma ^{\mu \nu }\gamma _5`$ (6 of which are linearly independent). Thus we encounter the question, to what extent (1.2) can involve $`𝒯_{\mu \nu }=𝒯_{\nu \mu }.`$ For this problem it is irrelevant whether $`𝒯_{\mu \nu }`$ is a separate ‘tensor potential’ or a multiple of Maxwell’s $`F_{\mu \nu }:=A_{\mu ,\nu }A_{\nu ,\mu }`$ (as proposed by Pauli ) or of its dual $`\epsilon _{\mu \nu \rho \sigma }F^{\rho \sigma }.`$ Hence we always call $`\sigma ^{\mu \nu }𝒯_{\mu \nu }`$ the ‘Pauli term’ of (1.2). In rigorous, but not trivial ways, we find that $`𝒯_{\mu \nu }`$ must be absent for a very basic reason: For the members $`𝐚,𝐛,\mathrm{}`$ of any \*-algebra and their conjugates $`𝐚^{},𝐛^{},\mathrm{},`$ one postulates $`(𝐚^{}𝐛)^{}=𝐛^{}𝐚.`$ This would be violated by any Pauli term. Hence we must demand
$$𝒯_{\mu \nu }=0\text{in order to keep}(𝐚^{}𝐛)^{}=𝐛^{}𝐚.$$
(1.4)
In other words, our fields will not generate a \*-algebra, unless (1.2) is restricted by (1.4). This clear result has not been found in the literature, because a familiar reciprocity is generally misnamed a theorem, whereas we prove it to be a condition, which excludes (1.2) unless it satisfies (1.4).
Showing in Sections 2 and 3 the adopted foundations, we indicate the proof of (1.4) briefly in Section 4, and more elaborately in Appendix A. It rests on a reciprocity condition, which is discussed in Appendix B. Calling this a ‘relation’ , one generally suggests that it holds without restricting the Bose fields to be prescribed. That misnomer may have caused the absence of (1.4) in the literature.
### 2 Gauge theory from Clifford algebra
For the vector field from (1.3) we must admit the gauge transformation
$$𝒱_\mu e^{i\omega }\left(𝒱_\mu i_\mu \right)e^{i\omega }𝒱_\mu +\omega _{,\mu }+i[𝒱_\mu ,\omega ].$$
(2.5)
In order to state that $`𝒱_\mu `$ and $`\omega `$ are hermitian matrices, we write them
$$𝒱_\mu (x)=𝐭_Y𝒱_\mu ^Y(x)=𝒱_\mu (x)^{}\text{and}\omega (x)=𝐭_Y\omega ^Y(x)=\omega (x)^{}.$$
(2.6)
While the constant matrices $`𝐭_Y`$ act on flavors and colors, they contain all coupling constants and the $`\gamma _5:=i\gamma ^0\gamma ^1\gamma ^2\gamma ^3`$ from Dirac’s Clifford algebra $`Cl_D.`$ We generate this $`Cl_D`$ by $`\gamma ^{(\mu }\gamma ^{\rho )}=\eta ^{\mu \rho }`$ from
$$\gamma _\mu ^{}=\gamma _\mu ^1=\gamma ^\mu =\overline{\gamma ^\mu },\text{where}\overline{\mathrm{\Gamma }}:=\gamma _0\mathrm{\Gamma }^{}\gamma _0Cl_D.$$
(2.7)
A transformation similar to the homogeneous part of (2.5) follows for the $`𝒮`$ of (1.3). In order to prove (1.4), however, we must initially use (1.2) with (1.3) in the form
$$=𝒮+\gamma ^\mu 𝒱_\mu +\sigma ^{\mu \nu }𝒯_{\mu \nu },\text{where}𝒯_{\mu \nu }0.$$
(2.8)
This together with $`(𝐚^{}𝐛)^{}=𝐛^{}𝐚`$ will in Section 4 yield a contradiction, which then proves (1.4). That proof holds in Quantum Induction where the canonical relations
$$[\mathrm{\Psi }(x),\mathrm{\Psi }(0)^{}]_+\delta (x^0)=\delta (x)\text{and}[\mathrm{\Psi }(x),\mathrm{\Psi }(0)^T]_+\delta (x^0)=0$$
(2.9)
together with Dirac’s equation (1.1) are fundamental. Its possible validity under presumptions different from (1.1) through (1.3) is discussed in Appendix B.
### 3 Short distance representation
For the bilocal, time ordered Dirac matrix
$$b(x,z):=(4\pi )^2T\mathrm{\Psi }(x+z)\overline{\mathrm{\Psi }}(xz),$$
(3.10)
(1.1) and (2.9) provide the differential equation
$$\{/^x+/^z+2i(x+z)\}b(x,z)=2\pi ^2\delta (z)=i/^zz/_{}^3,$$
(3.11)
where $`z/_{}^3:=(z^2iϵ)^2z/`$ with $`ϵ+0.`$ The representation used here for $`\delta (z):=\delta (z^0)\delta (z^1)\delta (z^2)\delta (z^3)`$ follows directly from the familiar
$$\mathrm{}(z^2iϵ)^1=(2\pi )^2i\delta (z).$$
Writing (3.10) as
$$b(x,z)=iz/_{}^3+(C^2+C^1+r^0)(x,z),$$
(3.12)
let us anticipate that the $`C^h`$ can be made homogeneous in the sense that
$$C^h(x,\lambda z)=\lambda ^hC^h(x,z)\text{for}h=2,1\text{and}\lambda C/.$$
(3.13)
Also using the ‘Taylor representation’
$$(x+z)=(x)+z^\mu _{,\mu }(x)+R(x,z)z/,\text{where}R(x,0)=0,$$
(3.14)
we can split (3.11) into
$$/^zC^2(x,z)=2(x)z/_{}^3,$$
(3.15)
$$/^zC^1(x,z)=2z^\mu (x)_{,\mu }z/_{}^3\{2i(x)+/^x\}C^2(x,z),$$
(3.16)
$$/^zr^0(x,z)=\mathrm{}2iR(x,z)z_{}^2.$$
(3.17)
Here the dots symbolize infinitely many unknown terms; they will be irrelevant because they are not more singular than $`z/_{}^1`$ (for $`z0`$ at $`z^20)`$. As one easily verifies, (3.15) with (2.8) can be solved by
$$z^4C^2(x,z)=2z/z^\mu \overline{𝒱}_\mu (x)z^2𝒮(x)^{}+z/\sigma ^{\mu \nu }z/𝒯_{\mu \nu }(x)^{}.$$
(3.18)
This clearly satisfies (3.13) and due to Appendix A is the only solution of (3.15) which does so. Since it makes the right side of (3.16) homogeneous in $`z`$ of the order $`h=2,`$ we can choose also $`C^1(x,z)`$ to obey (3.13). On the right side of (3.17), we have omitted terms which due to (3.18) are as homogeneous as $`z/_{}^1`$ (or less singular).
Thus (3.17) can be solved by an $`r^0`$ which (due to $`R(x,0)=0)`$ satisfies
$$z^\mu r^0(x,z)0\text{for}z0.$$
(3.19)
Therefore, we can include in $`r^0`$ those terms from $`C^2`$ and $`C^1`$ which are left arbitrary by (3.15) and (3.16) because they are independent of $`z.`$ In order to learn much more about $`r^0(x,z)`$ than (3.19) shows, one would need ‘outer’ boundary conditions (at large $`z).`$ These could be obtained from heat kernels (HK); but here the ‘inner’ condition (by (2.9) giving to (3.11) its right side) has been sufficient.
No obstruction to solving (3.11) has been encountered in (3.18) or in Appendix A. Neither would any arise if we would extend our recursion for (3.12) or invoke HK. Taking the Dirac adjoint of (3.10), however, we obtain the ‘reciprocity’ condition
$$\overline{z^4b(x,z)}=z^4b(x,z)\text{due to}(\mathrm{\Psi }_+\mathrm{\Psi }_{}^{})^{}=\mathrm{\Psi }_{}\mathrm{\Psi }_+^{}.$$
(3.20)
The latter exemplifies a general rule for \*-algebras.
### 4 Reciprocity as a condition
The $`z^4`$ in (3.20) removes the denominators of the terms in (3.12), so that the hermitian conjugation affects only their numerators. Hence these must (for $`h=2,1)`$ satisfy
$$\overline{z^4C^h(x,z)}=z^4C^h(x,z)\text{and}\overline{z^4r^0(x,z)}=z^4r^0(x,z),$$
(4.21)
because all three are linearly independent. In (4.21) we did not mention the leading $`iz/_{}^3`$ of (3.12), because it satisfies (3.20) trivially. Since (2.8) equals its adjoint $`\overline{}=\overline{𝒮}+\gamma ^\mu 𝒱_\mu ^{}+\sigma ^{\mu \nu }\overline{𝒯_{\mu \nu }},`$ that reciprocity is also verified for (3.18).
Instead of (4.21) with $`h=1,`$ however, in Appendix A we find
$$\overline{z^4C^1(x,z)}z^4C^1(x,z)\text{unless}𝒯_{\mu \nu }=0.$$
(4.22)
Hence it is misleading when one calls (3.20) a reciprocity ‘theorem’ , as if it were fulfilled for arbitrarily prescribed Bose fields (2.8). Since (3.18) satisfies (4.21) even with $`𝒯_{\mu \nu }0,`$ we also see that (1.4) cannot be derived as long as one only examines that solution of (3.15). In other words, we do not know a way of reaching the conclusion (1.4) without noting that the solution of (3.16) provides (4.22).
By (1.4), however, the $`C^1(x,z)`$ solving (3.16) is simplified greatly; and the further parts of (3.12) are shortened still more drastically. Hence superfluous work is done by those who pursue higher terms of (3.12) without inserting (1.4) quickly. They solve the differential equation (3.11) correctly but in excessive generality. Thus they miss the fact that (3.10) must also satisfy (3.20), which is a non-trivial condition not expressed by (3.11).
### Acknowledgments
For comments we are thankful to S. A. Fulling, K. Kwong, Z. Oziewicz, W. Stoeger and E. Sucipto.
### A Appendix: Linear Differential Equations
#### A.1 Exactly homogeneous solutions
In (3.11) we have used
$$i/^zz/_{}^3=2\pi ^2\delta (z)\text{for}z/_{}^3:=(z_{}^2)^2z/,$$
(A.23)
where $`z_{}^2=(z^2iϵ)^1`$ with $`ϵ+0.`$ These obviously provide
$$/^zz/_{}^3z^\mu =\gamma ^\mu z/_{}^3,/^zz_{}^2=2z/_{}^3,$$
$$/^zz/_{}^3\sigma ^{\mu \nu }z/=\gamma ^\rho z/_{}^3\sigma ^{\mu \nu }\gamma _\rho =2\sigma ^{\mu \nu }z/_{}^3,$$
(A.24)
which are derived most easily, when one uses (A.23) as often as possible. Thus (3.18) makes
$$/^zC^2(x,z)=2\gamma ^\mu z/_{}^3\overline{𝒱}_\mu (z)+2z/_{}^3𝒮(x)^{}+2z/_{}^3\sigma ^{\mu \nu }𝒯_{\mu \nu }(x)^{},$$
(A.25)
so that (3.15) with (2.8) is fulfilled.
All these calculations of course do not make sense on the cone $`z^2=0.`$ Hence throughout this paper we assume $`z^20`$ , as we must clearly do in (3.12) through (3.17). This is also true for the limits with $`z0,`$ as needed in (3.19) and in the proof of (A.23). Hence such limit transitions can proceed on any path which ends at $`z=0,`$ except that it must not touch the cone $`z^2=0.`$
While (3.15) with (2.8) is due to (A.23) satisfied by (3.18), its most general solution follows when we add any matrix $`H`$ which fulfills the homogeneous Dirac equation $`/H=0.`$ For choosing this $`H`$ we need the
Singularity Theorem: Every Poincaré covariant member $`H`$ of Dirac’s Clifford algebra $`Cl_D,`$ which solves
$$/H(z)=0\text{in a neighborhood of}z=0,$$
(A.26)
becomes for $`z0`$ with $`z^20`$ either more singular than $`z/^3`$ or less than $`z/^1,`$ hence yields either
$$\underset{z0}{lim}z/^3H(z)=\mathrm{}\text{or}\underset{z0}{lim}z^\mu H(z)=0.$$
(A.27)
One proves this easily when $`H`$ depends only on $`z.`$ The general proof is lengthy and scarcely of interest to physicists; hence we shall show it whenever requested.
Obvious solutions of (A.26) and (A.27) are all $`H`$ which do not depend on $`z.`$ Already the solution of (3.15) by (3.18) says that (3.13) with $`h=2`$ can be satisfied. The theorem (A.27) proves that (3.18) yields the only $`C^2(x,z)`$ which does so (such that (3.15) and (3.13) make (3.18) necessary). It also shows that the leading term of (3.12) is determined uniquely. In the same way, (A.27) says that the solution of (3.16) and (3.13) will be unique when such a $`C^1(x,z)`$ can be found at all.
#### A.2 Relevant short distance terms
Inserting (2.8) and (3.18) in (3.16), we obtain
$`/^zC^1(x,z)`$ $`=2z^\mu \gamma ^\rho z/_{}^3(\overline{𝒱}_{\rho ,\mu }\overline{𝒱}_{\mu ,\rho }2i\overline{𝒱}_\rho \overline{𝒱}_\mu )`$
$`+z_{}^2\gamma ^\rho (𝒮_{,\rho }^{}+2i𝒱_\rho 𝒮^{})+2z/_{}^3z^\rho (𝒮_{,\rho }^{}2i𝒮^{}\overline{𝒱}_\rho )`$
$`+2iz_{}^2𝒮𝒮^{}+2z/_{}^3\gamma ^{\mu \nu }z/𝒮𝒯_{\mu \nu }`$
$`+2z^\rho \gamma ^{\mu \nu }z/_{}^3(2𝒯_{\mu \nu }\overline{𝒱}_\rho +i𝒯_{\mu \nu ,\rho })2z_{}^2\gamma ^{\mu \nu }𝒯_{\mu \nu }𝒮^{}`$
$`+2i\gamma ^{\mu \nu }z/_{}^3\gamma ^{\rho \sigma }z/𝒯_{\mu \nu }𝒯_{\rho \sigma }`$
$`+\gamma ^\rho z/_{}^3\gamma ^{\mu \nu }z/(2𝒱_\rho 𝒯_{\mu \nu }i𝒯_{\mu \nu ,\rho })`$ (A.28)
with $`\gamma ^{\mu \nu }:=\gamma ^{[\mu }\gamma ^{\nu ]}=i\sigma ^{\mu \nu }=\gamma ^\mu \gamma ^\nu \eta ^{\mu \nu }.`$ Here all the fields $`𝒮,𝒱_\mu ,𝒯_{\mu \nu }`$ and their partial derivatives are localized at $`x`$; hence this parameter has been suppressed in the notation. Although (A.28) is for $`C^1`$ a linear differential equation in $`z`$ of the first order, deriving a solution was tedious; but after such a $`C^1`$ has been found, only differentiations are needed to verify its validity. Only a few readers, however, would perform this extremely easy but time-consuming task; thus let us merely say that the $`C^1(x,z)`$ solving the differential equation (A.28) is twice as lengthy as this.
It implies (4.22) and thus contradicts (4.21), unless all 5 local field polynomials
$`𝒵_{\rho \sigma \tau \mu }^1`$ $`:=[𝒯_{[\rho \sigma },𝒯_{\tau ]\mu }]_+,`$
$`𝒵_{\rho \sigma \tau }^2`$ $`:=𝒯_{[\rho \sigma }\overline{𝒱}_{\tau ]}+𝒱_{[\tau }𝒯_{\rho \sigma ]},`$
$`𝒵_\rho ^3`$ $`:=[𝒱^\mu ,𝒯_{\mu \rho }]`$ (A.29)
$`𝒵_{\rho \sigma }^4`$ $`:=𝒯_{\rho \sigma }𝒮^{}+𝒮𝒯_{\rho \sigma },`$
$`𝒵_{\rho \sigma \tau }^5`$ $`:=𝒯_{\rho \sigma ,\tau }+i\overline{𝒱}_\tau 𝒯_{\rho \sigma }i𝒯_{\rho \sigma }𝒱_\tau `$
satisfy
$$𝒵_{\mathrm{}}^n(x)=0.$$
(A.30)
These derivations have been rigorous, because we did not admit any approximation (rarely possible in physics). Our solution (1.4) is clearly the only which holds irrespective of $`𝒮`$ and $`𝒱_\mu .`$ If (A.30) were mathematically solvable by any $`𝒯_{\mu \nu }0`$ (a case we can’t examine exactly), it would restrict $`𝒮`$ and $`𝒱_\mu `$ in complicated and extremely unphysical ways.
Within the permitted size of this paper, we can’t show the complete proof for the necessity of (A.30); the known extension to quantum fields would enlarge it enormously. A result as simple as (1.4), however, should find a much shorter proof. Hence we would prefer to delay the publication until that simplicity is achieved. If we would not show the result (1.4) and indicate our lengthy derivation from (3.20), however, it would be hard to get anyone interested in such a problem.
### B The Reciprocity Violation
Let us finally collect further remarks about our result (1.4) and its absence from the literature:
* The reciprocity condition (3.20) has been called a relation, as if it had been proved with (1.1) containing the most general Bose field (1.2).
* We had to perform extensive computer algebra for the derivation of $`C^1(x,z)`$ from its differential equation (A.28). Still more would be required if one were to make the compact result from HK explicit. Neither is needed any longer, because it is easy to verify a known solution of any differential equation, no matter how hard its integration had been. Because of (a), however, nobody found this tedious search worthwhile.
* In the mathematics of HK, the boundary conditions at large $`z`$ are presently more interesting than the behavior of (3.12) at small $`z.`$
* It is fashionable, instead of the differential equation (3.11) to solve a related integral equation, with an ‘inner’ boundary condition given by the right side of (3.11) and a purely mathematical condition at some outer boundary (where $`z`$ is large or infinite). For our problem, no choice of the latter makes sense because the result (A.30) depends only on (3.11) and its boundary condition at $`z=0.`$ Why should we use a physically irrelevant integral equation for a conclusion which is completely determined by a differential equation together with a single, well justified boundary condition?
* Methods of HK have been initiated for classical field theories, where the reciprocity arises from a symmetry of their Green functions.
* In many treatments by HK, not only the Bose field $``$ but also Dirac’s $`\mathrm{\Psi }`$ is non-quantized (or not even mentioned). Then (3.11) is regarded as an equation for a classical Green function $`b(x,z),`$ not related to any quantum field such as (3.10). In that approach, one hardly sees whether a reciprocity (3.20) should be desired.
* Under infinite renormalizations, (1.1) and (2.9) and therefore (3.12) break down . Hence we can’t prove (1.4) in familiar settings (although it might be true even there).
* Any significant $`𝒯_{\mu \nu }0`$ would damage the excellent verification of the Standard Model by the magnetic moment of the electron. This agreement had formerly been regarded as a brilliant confirmation of renormalized QED. Under the present philosophy of ‘effective’ actions, however, it is an unimportant result of imprecise measurements.
* Instead of (3.10), one often uses $`\beta (e,t):=(4\pi )^2\text{T}\mathrm{\Psi }(e+t)\overline{\mathrm{\Psi }}(e)=b(e+\frac{1}{2}t,\frac{1}{2}t)`$ with the eccentric coordinates $`e=xz`$ and $`t=2z.`$ These simplify (especially under gravity) the derivation of (3.11), but make the analysis of (3.20) complicated.
* The singularity at $`z^2=0`$ makes (3.12) dependent on the time ordering of (3.10). Hence (3.20) can’t simply be written $`\overline{b}(x,z)=b(x,z)`$ because an hermitian conjugation reverses the time order.
* Wherever basic ‘tensor potentials’ $`𝒯_{\mu \nu }`$ have found any attention, one has coupled them to each other or further Bose fields , leaving their interaction with Dirac’s $`\mathrm{\Psi }`$ open.
* Whenever ‘tensor couplings’ are mentioned in phenomenology , it is unclear whether they are fundamental or caused by bound states or by ‘radiative’ corrections.
* The differential or integral equation for (3.10) can ‘mathematically’ be solved without any concern about the reciprocity (3.20) needed in physics. However, (3.11) without (3.20) does not exhaust the contents of (3.10).
* The leading terms $`iz/_{}^3`$ and $`C^2(x,z)`$ of (3.12) satisfy (3.20) even when (2.8) has $`𝒯_{\mu \nu }0.`$ Hence the contradiction between (4.21) and (4.22) is not recognized until one also determines $`C^1(x,z).`$
* A further reason for the usual rejection of (1.4) may be that such a clear result deserves a simple derivation. Instead our proof has required the lengthy (but straightforward) deduction of the differential equation (A.28) and its explicit solution. That $`C^1(x,z)`$ is twice as long as (A.28) and therefore not shown here; but we hope that others can simplify our arguments.
* In order to examine (3.20) completely , the reciprocity (4.21) should also be checked for the ‘remainder’ $`r^0(x,z).`$ Analyzing those parts of it which in $`z`$ are homogenous of the orders $`h=0`$ and $`h=1,`$ we have not found any restriction beyond $`𝒯_{\mu \nu }=0`$ (which simplifies those parts enormously). Our local approach (without outer boundary conditions) cannot extend that result to all orders. This should be taken as an incentive to treat the condition (3.20) globally, but not as excuse for discarding our result (4.22). Doing so would correspond to ignoring the singular part of a Laurent series until all its orders are known.
* For many authors, the Higgs field does not contribute to $`,`$ because they attach it in isospinors to $`\mathrm{\Psi }.`$
* Many authors prefer two-component spinors instead of Dirac’s $`\mathrm{\Psi }.`$
* Some authors use other notations for Dirac matrices, for instance $`\alpha ,\beta `$ instead of $`\gamma ^\mu `$ or explicit 4 x 4 squares.
* For Dirac’s $`\gamma ^\mu ,`$ one sometimes uses representations in which $`\gamma ^\mu `$ is not its own adjoint or the $`\beta `$ in $`\overline{\mathrm{\Psi }}=\mathrm{\Psi }^{}\beta `$ differs from $`\gamma ^0.`$
* From (2.8) we derived (1.4) by ‘reductio ad absurdum’, which not every mathematician appreciates.
* Readers may dislike (1.4), because beautiful theories are no longer expected to be simple but to offer rich mathematical structures.
* Instead of the Dirac equation for physics (which is of first order in Minkowski space), mathematicians prefer the elliptic equation given by its iteration in Euclidean space.
* One often uses Dirac’s Clifford algebra without any basis, hence not separating the scalar, vectorial and tensorial parts of $``$.
Kurt Just
Department of Physics
University of Arizona
Tucson AZ 85721
E-mail: just@physics.arizona.edu
James Thevenot
Department of Physics
University of Arizona
Tucson AZ 85721
E-mail: jimthev@physics.arizona.edu
Received: September 30, 1999; Revised: February 6, 2000.
|
warning/0005/math0005033.html
|
ar5iv
|
text
|
# The Anisotropic Averaged Euler Equations
## 1 Introduction
A fundamental problem in turbulent fluid dynamics is the difficulty in resolving the many spatial scales that are activated by the complicated nonlinear interactions. It is a challenge to produce models that capture the large scale flow, while correctly modeling the influence of the small scale dynamics. While there are many efforts in this direction, the goal of the present paper is to introduce a new method that is based on the combination of two basic ideas: the use of an ensemble averaging that represents a spatial sampling of material particles over small spatial scales, and the use of asymptotic expansions together with this averaging on the level of the variational principle.
Our approach is conceptually similar to the method of Reynolds averaging and Large Eddy Simulation techniques, but has the advantage of 1) not needing an additional closure model and 2) automatically providing a small scale corrector to the macroscopic flow field.
Our methodology has some interesting connections with the method of Optimal Prediction introduced by Chorin et al. , which will be explored in future publications.
In the body of the paper we shall comment on a comparison between our approach and that of Chen et al. and Holm , which produces different equations.
### 1.1 A Brief Review of the Euler and Isotropic Averaged Euler Equations
#### A Brief History.
There has been much recent interest in the averaged Euler equations for ideal fluid flow. In this paper we will focus on the geometry and analysis of a related set of equations, which we call the anisotropic averaged Euler equations. The original averaged Euler equations appear as a special isotropic case of the more general equations.
The isotropic averaged Euler equations on all of $`^n`$ first appeared in the context of an approximation to the Euler equations in Holm et al. \[1998a\] and some of its variational structure was developed in Holm et al. \[1998b\]; this variational structure retains the quadratic form of the variational structure for the original Euler equations, so that the equations can be viewed as describing a certain geodesic flow in a sense similar to the work of Arnold and Ebin and Marsden .
Remarkably, these equations are mathematically identical to the well-known inviscid second grade fluids equations introduced by Rivlin and Erickson . The geometric analysis of these equations, including well-posedness and other analytic properties, was developed in Shkoller and Marsden et al. . These references also discuss the relation to the second-grade fluid literature.
In Oliver and Shkoller , the link with the vortex blob method was established; therein, it was shown that the vortex blob numerical algorithm generates unique global weak solutions to the averaged Euler equations. These weak solutions induce a weak coadjoint action on the vector space of vorticity functions, modeled as the space of Radon measures. The existence of such a weak coadjoint action makes rigorous the formal constructions of Marsden and Weinstein on the geometry of point-vortex and vortex blob dynamics.
The works of Chen et al. and Holm formulated equations for the slow time dynamics of fluid motion by averaging over fast time fluctuations about the mean; that approach, founded on a Reynolds decomposition translated over the Lagrangian parcel, and the resulting system of equations, is different from our approach and from the results that we shall present. We give a few more details on the comparison in the body of the paper.
#### The Euler Equations as Geodesics and Notation.
It is well-known how to view the Euler equations as geodesics on the group of diffeomorphisms and that this view has concrete analytical advantages, due to the work of Arnold and Ebin and Marsden . In particular, this work shows that the equations define a smooth vector field (a spray) on the group of diffeomorphisms, that is, in Lagrangian (or material) representation. The reduction of the equations from material to spatial (Eulerian) representation may be viewed by the classical and general technique of Euler-Poincaré reduction (see Marsden and Ratiu and Holm et al. \[1998b\] for an exposition and further references) and this view is a helpful guide to understanding other fluid theories as well.
The geometric view of fluid mechanics, along with a careful understanding of the averaging process, will be basic to the present paper, so we briefly review the salient features of the theory for the reader’s convenience, and to establish notation.
Let $`(M,g)`$ be a $`C^{\mathrm{}}`$ compact, oriented $`n`$-dimensional Riemannian manifold with $`C^{\mathrm{}}`$ boundary (possibly empty). Of course open regions with smooth boundary in the plane or space are key examples. The Riemannian volume form associated with the metric $`g`$ is denoted $`\mu `$.
The Euler equations for the velocity field $`u`$ of an ideal, incompressible, homogeneous fluid moving on $`M`$ (such as a region $`\mathrm{\Omega }`$ in $`^2`$ or $`^3`$) are
$$\frac{u}{t}+(u)u=p$$
(1.1)
with the constraint $`\mathrm{div}u=0`$ and the boundary condition that $`u`$ is tangent to the boundary, $`M.`$ The pressure $`p`$ is determined by the incompressibility constraint. The nonlinear term $`(u)u`$ is interpreted in the context of manifolds to be $`_uu`$, the covariant derivative of $`u`$ along $`u`$. In Euclidean coordinates, these equations are given as follows (using the summation convention for repeated indices):
$$\frac{u^i}{t}+u^j\frac{u^i}{x^j}=\frac{p}{x^i},$$
and on a Riemannian manifold (or in curvilinear coordinates in Euclidean space), the Euler equations take the following coordinate form:
$$\frac{u^i}{t}+u^j\frac{u^i}{x^j}+\mathrm{\Gamma }_{jk}^iu^ju^k=g^{ij}\frac{p}{x^j},$$
where $`g_{ij}`$ are the components of the Riemannian metric $`g`$, $`g^{ij}=[g_{ij}]^1`$, and $`\mathrm{\Gamma }_{jk}^i`$ are the associated Christoffel symbols. Using covariant derivative notation, these coordinate equations read
$$\frac{u^i}{t}+u^ju_{;j}^i=g^{ij}p_{,j}.$$
We let the flow of the time dependent vector field $`u(t,x)`$ be denoted by $`\eta (t,x)`$ so that
$$\frac{}{t}\eta (t,x)=u(t,\eta (t,x)),$$
with $`\eta (0,x)=x`$ for all $`x`$ in $`M`$. For each $`t`$, we denote the map $`\eta (t,)`$ by $`\eta _t`$ so that $`\eta _0=e`$, the identity map. Thus, the map $`x\eta _t(x)`$ gives the particle placement field for the fluid. Because of the incompressibility, each map $`\eta _t`$ is volume preserving and is a diffeomorphism.
We shall be working with vector fields $`u`$ of Sobolev class $`H^s`$ for $`s>(n/2)+1`$ and, correspondingly, $`\eta _t𝒟_\mu ^s`$, the group of $`H^s`$-volume preserving diffeomorphisms. If there is any danger of confusion, we shall write $`𝒟_\mu ^s(M)`$ to indicate the underlying manifold $`M`$. See Ebin and Marsden and Shkoller for some basic properties of Hilbert class diffeomorphism groups for manifolds with boundary.
Arnold’s theorem on the Euler equations may be stated as follows: A time dependent velocity field $`u`$ satisfies the Euler equations iff the curve $`\eta _t`$ is a geodesic of the right invariant $`L^2`$-metric on $`𝒟_\mu ^s`$.
This $`L^2`$-metric is defined as follows. The tangent space to $`𝒟_\mu ^s`$ at the identity is identified with the space $`𝔛_{\mathrm{div}}^s`$, the space of $`H^s`$ divergence free vector fields on $`M`$ that are tangent to the boundary $`M`$. The right invariant $`L^2`$-metric is defined to be the weak Riemannian metric on $`𝒟_\mu ^s`$ whose value at the identity is
$$u,w_{L^2}=_Mu(x),w(x)_x\mu (x),$$
where we write the pointwise inner product as $`u(x),w(x)_x=g(x)(u(x),w(x))`$, and the pointwise norm $`|u(x)|^2=u(x),u(x)_x`$.
As we shall explain shortly, with the maturation of Euler-Poincaré theory, Arnold’s theorem becomes an easy consequence of more general and rather simple results.
#### Lie Derivative and Vorticity Form.
As is well-known, the Euler equations can be written in terms of Lie derivatives as
$$\frac{u^{\mathrm{}}}{t}+\mathrm{\pounds }_uu^{\mathrm{}}=𝐝\left(\frac{1}{2}|u|^2p\right)=𝐝p^{},$$
(1.2)
where $`u^{\mathrm{}}`$ is the one-form associated to the vector field $`u`$ via the metric, and $`\mathrm{\pounds }_uu^{\mathrm{}}`$ denotes the Lie derivative of the one-form $`u^{\mathrm{}}`$ along $`u`$. Taking the exterior derivative of (1.2) gives the familiar advection equation for vorticity:
$$\frac{\omega }{t}+\mathrm{\pounds }_u\omega =0,$$
where $`\omega =𝐝u^{\mathrm{}}`$ is the vorticity, thought of as a two-form. In 2D, $`\omega `$ is identified with a scalar and is traditionally thought of as the 2D-curl of the velocity field, while in 3D, $`\omega `$ may be identified (using the volume-form $`\mu `$) with a vector field which is traditionally obtained by taking the curl of $`u`$.
The vorticity equation is the infinitesimal version of the following advection property:
$$\omega _t=(\eta _t)_{}\omega _0.$$
Of course in two dimensions, this gives the usual advection of vorticity as a function, while in three (or higher) dimensions, the advection is understood in terms of advection of two-forms.
The Euler equations have both an interesting Hamiltonian structure in terms of Poisson brackets (a Lie-Poisson bracket) and a variational structure. In this paper we shall be working primarily with the variational structure; the Hamiltonian structure, along with references to the literature may be found in Marsden and Weinstein , Arnold and Khesin and Marsden and Ratiu .
#### Lagrangian and Variational Form.
The Lagrangian is given by the total kinetic energy of the fluid; in spatial representation, this Lagrangian is
$$L(u)=\frac{1}{2}_M|u(x)|^2\mu .$$
(1.3)
The corresponding (unreduced) Lagrangian on $`T𝒟_\mu ^s`$ is given by
$$(\eta ,\dot{\eta })=\frac{1}{2}_Mg(\eta (x))(\dot{\eta }(x),\dot{\eta }(x))\mu .$$
(1.4)
Hamilton’s principle on $`𝒟_\mu ^s`$ applied to the Lagrangian $``$ gives geodesics on this group. Euler-Poincaré reduction techniques (see Marsden and Ratiu ) show that this variational principle reduces to the following principle in terms of Eulerian velocities:
$$\delta _a^bL(u)𝑑t=0,$$
which should hold for all variations $`\delta u`$ of the form
$$\delta u=\dot{w}+[u,w],$$
where $`w`$ is a time dependent vector field (representing the infinitesimal particle displacement) vanishing at the temporal endpoints<sup>1</sup><sup>1</sup>1The constraints on the allowed variations of the fluid velocity field are commonly known as “Lin constraints”. This itself has an interesting history, going back to Ehrenfest, Boltzmann, Clebsch, Newcomb and Bretherton, but there was little if any contact with the heritage of Lie and Poincaré on the subject.. Here, $`[w,u]`$ denotes the usual Jacobi–Lie bracket of vector fields. One readily checks that this reduced principle yields the standard Euler equations. This simple computation is the heart of Arnold’s theorem.
#### Analytical Issues.
While the Eulerian (spatial) representation has been emphasized in most analytic studies of the Euler equations, fluid motion viewed on the Lagrangian (material) side has some distinct advantages. For example, Ebin and Marsden proved that the flow, solving the Euler equations, on the volume-preserving diffeomorphism group $`𝒟_\mu ^s`$, $`s>n/2+1`$, is smooth in time. They derived a number of interesting consequences from this result, including theorems on the convergence of solutions of the Navier-Stokes equations to solutions of the Euler equations as the viscosity goes to zero when $`M`$ has no boundary. In addition, Marchioro and Pulvirenti analyzed the Lagrangian flow map to establish sharp well-posedness of the 2D Euler equations and prove convergence of the vortex blob algorithm. In many cases, the Lagrangian framework is, in fact, the more natural setting to study the behavior of solutions, and we shall emphasize this point of view.
### 1.2 The Averaged Euler Equations
#### The Isotropic Averaged Euler Equations.
Let $`\alpha `$ be a positive constant. In Euclidean space and in Euclidean coordinates, the isotropic averaged Euler equations (inviscid second-grade fluids equations)<sup>2</sup><sup>2</sup>2These are also known as the Euler-$`\alpha `$ equations. read:
$$\frac{v^i}{t}+u^j\frac{v^i}{x^j}\alpha ^2\left[\frac{u^j}{x^i}\right]\mathrm{\Delta }u_j=\frac{p}{x^i},$$
where $`v=u\alpha ^2\mathrm{\Delta }u`$ and $`\mathrm{\Delta }`$ denotes the componentwise Laplacian, and there is a summation over repeated indices (in Euclidean coordinates, as is common, we make no distinction between indices up or down). While there are several choices, the no slip boundary conditions $`u=0`$ are often used for this model.
#### Rate of Deformation Tensor.
One of the interesting things that comes out of a careful derivation of the equations is the natural occurence of the rate of deformation tensor, which is defined by
$$\mathrm{Def}u=\frac{1}{2}\left(u+(u)^T\right)$$
which we write in coordinates as:
$$\left(\mathrm{Def}u\right)_j^i=\frac{1}{2}\left(u_{;j}^i+u_{;i}^j\right).$$
We also let $`\mathrm{Def}u^{\mathrm{}}=\frac{1}{2}\left[u^{\mathrm{}}+(u^{\mathrm{}})^T\right]`$ which we write in coordinates as
$$D_{ij}=\left(\mathrm{Def}u^{\mathrm{}}\right)_{ij}=\frac{1}{2}\left(u_{i;j}+u_{j;i}\right).$$
Note that this is exactly the Lie derivative of the metric tensor; that is, $`\mathrm{Def}u^{\mathrm{}}=\mathrm{\pounds }_ug`$, which is sometimes called the Killing tensor.
#### Smoothness Properties.
Results on smoothness of the Lagrangian flow map for the averaged Euler equations were given in Shkoller on compact boundaryless Riemannian manifolds, and in Marsden et al. on compact Euclidean domains. The problem of how to formulate this system on compact Riemannian manifolds with boundary was solved in Shkoller ; the equations take the form
$$_t(1\alpha ^2\mathrm{\Delta }_r)u+_u(1\alpha ^2\mathrm{\Delta }_r)u\alpha ^2(u)^t\mathrm{\Delta }_ru=\mathrm{grad}p,$$
together with the constraint $`\mathrm{div}u=0,`$ and with appropriate initial conditions $`u(0)=u_0`$, as well as boundary conditions. The symbol $`\mathrm{\Delta }_r`$ is the operator $`\mathrm{Def}^{}\mathrm{Def}`$ acting on divergence-free vector fields, where $`\mathrm{Def}^{}`$ is the $`L^2`$ formal adjoint of the (rate of) deformation operator $`\mathrm{Def}`$. Explicitly,
$$\mathrm{\Delta }_r=(d\delta +\delta d)+2\mathrm{Ric}.$$
(1.5)
As with the usual Euler equations, the function $`p`$ is determined from the incompressibility condition.
#### Lie Derivative Form—The Isotropic Equations.
The averaged Euler equation can be neatly written in terms of Lie derivatives:
$$_tv^{\mathrm{}}+\mathrm{\pounds }_uv^{\mathrm{}}=dp,$$
(1.6)
where $`v^{\mathrm{}}=(1\alpha ^2\mathrm{\Delta }_r)u^{\mathrm{}}`$.
#### The Anisotropic Averaged Euler Equations.
These equations, which are the main subject of the present paper, and which will be derived in §3.2, will now be stated. The basic variables that are evolving in the anisotropic averaged Euler equations are the macroscopic velocity field $`\mathrm{u}`$ and a symmetric tensor field $`F`$ on $`M`$; the tensor field $`F`$ will be interpreted as the contravariant spatial fluctuation tensor and it will keep track of the anisotropy of the fluid deviations from the macroscopic flow. These equations also depend on a choice of length scale $`\alpha `$.
It is convenient to define the linear operator $`𝒞:𝔛_{\mathrm{div}}^sH_0^1H^{s2}`$, $`s1`$, by
$$𝒞u:=\mathrm{Div}\left[C:u^{\mathrm{}}\right],$$
where $`\mathrm{}`$ is the map from vector fields to one-forms associated with the metric $`g`$, and the fourth-rank symmetric positive tensor $`C`$ is the symmetrization of the tensor $`Fg^1`$, given in local coordinates by
$$C^{ijkl}=\frac{1}{4}\left(F^{lj}g^{ik}+f^{kj}g^{il}+F^{li}g^{jk}+F^{ki}g^{jl}\right).$$
With this notation, the anisotropic averaged Euler equations on manifolds are
$`_t(1\alpha ^2𝒞)u`$ $`+_u(1\alpha ^2𝒞)u\alpha ^2[u]^t𝒞u+2\alpha ^2F:[(\mathrm{Def}u^{\mathrm{}})^2]^{\mathrm{}}`$
$`4\alpha ^2\mathrm{Div}\left(\left(\mathrm{Def}u\right)^2F\right)=\mathrm{grad}p,`$
together with the advection equation
$$_tF+\mathrm{\pounds }_uF=0,$$
the incompressibility constraint $`\text{div }u=0`$, initial data $`u(0)=u_0`$ and $`F(0)=F_0`$, and the Dirichlet boundary condition $`u=0`$.
#### Lie Derivative Form—Anisotropic Equations.
The anisotropic averaged Euler equations can also be written using Lie derivatives as
$$_tv^{\mathrm{}}+\mathrm{\pounds }_uv^{\mathrm{}}+[2\alpha ^2F:[(\mathrm{Def}u^{\mathrm{}})^2]^{\mathrm{}}4\alpha ^2\mathrm{Div}(\left(\mathrm{Def}u\right)^2F)]^{\mathrm{}}=dp,$$
(1.7)
where $`v^{\mathrm{}}=(1\alpha ^2𝒞^{\mathrm{}})u^{\mathrm{}}`$, where $`𝒞^{\mathrm{}}u^{\mathrm{}}=\text{Div}[C:u^{\mathrm{}}]^{\mathrm{}}`$.
#### Coordinate Form.
In local coordinates, the anisotropic averaged Euler equations become
$`_t(u^i\alpha ^2[C^{ijkl}u_k,_j],_l)+(u^i\alpha ^2[C^{ijkl}u_k,_j],_l),_mu^m\alpha ^2u_m,_i[C^{mjkl}u_k,_j],_l`$
$`+2\alpha ^2F^{kj}[D_{km}g^{mn}D_{nj}],_i4\alpha ^2[F^{kj}D_{im}g^{mn}D_{nj}],_k`$
$`=p,_i`$
together with the advection equation
$$_tF^{ij}+F^{ij},_ku^kF^{kj}u^i,_kF^{ik}u^j,_k=0,$$
with the constraint $`u_{,i}^i=0`$, given initial conditions $`u^i(0)=u_0^i`$, and with the no-slip conditions $`u^i=0`$ on the boundary. If the metric $`g`$ is not the Euclidean metric $`\delta _{ij}`$, then the partial derivatives above should be interpreted as arising from the Levi-Civita covariant derivative associated to $`g`$.
### 1.3 Outline of the Main Results.
The main results of the present work are as follows:
1. We derive, in a systematic way, the first order averaged Lagrangian given in coordinates by
$$L_1^\alpha (u,F)=\frac{1}{2}_M\left\{g_{ik}u^iu^k+2\alpha ^2g^{ik}F^{jl}D_{ij}D_{kl}\right\}[\text{det }g]^{\frac{1}{2}}𝑑x.$$
and, using the calculus of variations, derive the associated anisotropic averaged Euler equations as the corresponding Euler-Poincaré equations. The Euler-Poincaré technique was also used in Holm , but the Lagrangian and associated equations are different. In particular, the principles and philosophy governing the derivation of the Lagrangian are completely different.
2. We show that the equations are well posed; in fact, we show more, namely that the corresponding Lagrangian flow map is smooth in time in the appropriate Sobolev topology.
3. Another important achievement is that while the macroscopic velocity field $`u`$ is computed on spatial scales larger than $`\alpha `$, we are able to obtain a corrector for this macroscopic field to order $`\alpha ^2`$. This is done in §4.2 and is similar to what one does in the theory of homogenization.
## 2 The Derivation
### 2.1 Introduction
This section presents a new method for constructing models of hydrodynamics which takes into account the fundamental idea that a fluid particle is not a point, but rather a collection of points forming a representative sampling. Our approach is founded upon a certain type of Lagrangian ensemble averaging performed at the level of the variational principle. A similar idea on the level of the equation itself, as opposed to the variational principle, was used by Barenblatt and Prostokishin for deriving models of damage propagation.
#### Naive Averaging Does not Work.
We first explain why the naive approach to spatially averaging a quadratic Lagrangian or Hamiltonian does not suffice. As a simple example, consider the Lagrangian on scalar functions on $`^n`$ given by $`L(u)=\frac{1}{2}_^nu^2(x)𝑑x`$ and for a given positive constant $`\alpha `$, define a new averaged Lagrangian by
$$L^\alpha (u)=_^n\frac{1}{|B(x,\alpha )|}_{B(x,\alpha )}u^2(z)𝑑z𝑑x$$
which is obtained from $`L`$ by averaging the original Lagrangian over balls of radius $`\alpha `$. Here $`B(x,\alpha )`$ denotes the ball of radius $`\alpha `$ about the point $`x`$ in $`^n`$ and $`|B(x,\alpha )|`$ denotes its volume.
Taylor expanding the integrand about $`x`$ and then integrating by parts yields cancellation of all but the zeroth-order term, thus reproducing exactly the original Lagrangian $`L`$. This is to be expected since the quadratic nonlinearity is rather weak, and since absolutely no information concerning the local spatial structure of the continuum is being provided. The latter issue is of fundamental importance and is the foundation upon which we shall build our theory.
### 2.2 The Averaging Construction.
To implement our construction, we will average over an ensemble of Lagrangian fluctuation maps. We will now proceed to develop this formalism.
#### Fuzzying the Lagrangian Flow.
Let $`(M,g)`$ be a $`C^{\mathrm{}}`$ compact, oriented $`n`$-dimensional Riemannian manifold with $`C^{\mathrm{}}`$ boundary (possibly empty). We consider a two-parameter family of volume-preserving diffeomorphisms $`\xi ^{ϵ,\theta }`$ of $`M`$ depending on a “radial” component $`ϵ[\alpha /2,\alpha /2]`$, $`\alpha >0`$, and an “angular” component $`\theta S_+^{n1}`$, where $`S_+^{n1}`$ denotes the upper hemisphere of the unit sphere $`S^{n1}`$ in $`^n`$. In case $`M`$ has nonempty boundary, we embed $`M`$ into its double $`\stackrel{~}{M}`$ and consider this two-parameter family defined on $`\stackrel{~}{M}`$; in this case, $`\xi ^{ϵ,\theta }`$ need not leave $`M`$ invariant. This fact will be important later for certain ellipticity properties.
The parameterization is chosen such that
$$\begin{array}{c}\xi ^{0,\theta }(t,x)=x,\\ \text{dist}(x,\xi ^{ϵ,\theta }(x))<|ϵ|\end{array}$$
for all $`ϵ[\alpha /2,\alpha /2]`$, all $`t`$, and $`\theta S_+^{n1}`$. We define the infinitesimal fluctuation vector by
$$\xi ^{}(\theta ,t,x)=\frac{d}{dϵ}|_{ϵ=0}\xi ^{ϵ,\theta }(t,x),$$
a vector field depending on the parameter $`\theta `$ and time $`t`$.
For each time $`t`$, the Lagrangian flow map $`\eta _t`$, where $`\eta _t(x)=\eta (t,x)`$, associated with a solution of the Euler equations is a volume-preserving diffeomorphism of $`M`$ which maps fluid particles $`xM`$ to $`\eta _t(x)M`$. Motivated by the idea that a particle in a continuum should really be regarded as a representative of a sample of particles over a region, we define the $`\xi _t^{ϵ,\theta }`$-perturbed particle placement field by
$$\eta _t^{ϵ,\theta }(x)=(\xi _t^{ϵ,\theta })^1\eta _t(x)$$
(2.1)
for all $`ϵ[\alpha /2,\alpha /2]`$ and $`\theta S_+^{n1}`$. The family of maps $`\eta _t^{ϵ,\theta }`$ is called the fuzzy flow. For each $`ϵ`$, $`\theta `$, and $`t`$, the map $`\eta _t^{ϵ,\theta }:MM`$ is a volume-preserving diffeomorphism of the fluid container. Note that at $`ϵ=0`$, $`\eta _t^{0,\theta }=\eta _t`$ for all $`\theta S_+^{n1}`$.
We take $`\eta _t𝒟_\mu ^s(M)`$ and $`\xi _t^{ϵ,\theta }𝒟_\mu ^{\mathrm{}}(\stackrel{~}{M})`$ so that $`\eta _t^{ϵ,\theta }𝒟_\mu ^s(\stackrel{~}{M})`$. See §1.1 for the definition of the group $`𝒟_\mu ^s`$.
#### Decomposition of the Spatial Velocity Field.
Our goal is to derive the Eulerian velocity field $`u_t^{ϵ,\theta }`$ corresponding to the $`\xi _t^{ϵ,\theta }`$-perturbed particle placement field $`\eta _t^{ϵ,\theta }`$, and define a new Lagrangian by averaging the velocity $`u_t^{ϵ,\theta }`$ over the radial parameter $`ϵ`$ and the angular coordinate $`\theta `$. We shall proceed with this averaging process as follows: we begin by defining the Eulerian vector fields associated with our three Lagrangian maps. Let
$`_t\eta (t,x)`$ $`=u(t,\eta (t,x)),`$
$`_t\xi ^{ϵ,\theta }(t,x)`$ $`=w^{ϵ,\theta }(t,\xi ^{ϵ,\theta }(t,x)),`$
$`_t\eta ^{ϵ,\theta }(t,x)`$ $`=u^{ϵ,\theta }(t,\eta ^{ϵ,\theta }(t,x)).`$
Differentiating the Lagrangian decomposition (2.1) with respect to time $`t`$, we obtain the spatial velocity decomposition
$$u^{ϵ,\theta }(t,x)=\left\{\xi _{}^{ϵ,\theta }{}_{}{}^{}(uw^{ϵ,\theta })\right\}(t,x),$$
(2.2)
where the notation $`\xi _{}^{ϵ,\theta }{}_{}{}^{}`$ denotes the pullback by the map $`\xi ^{ϵ,\theta }`$. We can also write this decomposition using the push-forward notation via the relation $`\left(\xi _t^{ϵ,\theta }\right)^{}=\left(\xi _t^{ϵ,\theta }\right)_{}^{1}{}_{}{}^{}`$, so that the action on a vector field $`v`$ is given by
$$\left(\xi _t^{ϵ,\theta }\right)^{}v=T(\xi _t^{ϵ,\theta })^1v\xi _t^{ϵ,\theta },$$
where we use the symbol $`T`$ to denote the tangent map (which is locally represented by the matrix of partial derivatives). Thus, the decomposition (2.2) may be equivalently written as <sup>3</sup><sup>3</sup>3This decomposition can also be written as $`u^{ϵ,\theta }(t,x)=\text{Ad}_{(\xi _t^{ϵ,\theta })^1}(u_tw_t^{ϵ,\theta })`$, where Ad is the adjoint action of the volume-preserving diffeomorphism group on divergence-free vector fields.
$$u^{ϵ,\theta }(t,x)=T(\xi _t^{ϵ,\theta })^1(x)\left(u(t,\xi _t^{ϵ,\theta }(x))w^{ϵ,\theta }(t,\xi _t^{ϵ,\theta }(x))\right),$$
(2.3)
where, again, $`u^{ϵ,\theta }(t,x)`$ is the Eulerian spatial velocity field corresponding to the fuzzy flow $`\eta _t^{ϵ,\theta }`$.
#### Comments on the Nature of the Decomposition.
The Lagrangian decomposition (2.1) which “fuzzies” the Lagrangian flow map yields the decomposition (2.3) for the corresponding Eulerian variables which is of a hybrid Lagrangian-Eulerian type. The Lagrangian characteristics of this decomposition are encompassed in the presence of the purely Lagrangian fluctuation maps $`\xi _t^{ϵ,\theta }`$, and it is indeed the presence of this Lagrangian term in (2.3) which allows us to proceed with an asymptotic expansion which is both philosophically and mathematically different from the “naive” expansion we discussed earlier. We should also emphasize that without this Lagrangian aspect, the decomposition (2.3) would reduce to the usual additive (Reynolds) decomposition of spatial velocity fields into their mean and fluctuating parts, which does not reflect the fuzzyness of the Lagrangian flow map.
Our approach should also be contrasted with the approach taken by Chen et al. and Holm . In those papers, the decomposition
$$\eta ^\sigma (t,x)=\eta (t,x)+\sigma (t,\eta (t,x)),$$
is made, where $`\sigma (t,x)`$ is a fluctuation vector field, and $`\eta ^\sigma (t,x)`$ is a perturbed Lagrangian trajectory of the reference element $`x`$. This decomposition is intrinsically problematic, in that a material vector field $`\sigma (t,\eta (t,))`$ is being added to a volume-preserving diffeomorphism $`\eta (t,)`$. As a consequence, the perturbed trajectory $`\eta ^\sigma (t,x)`$ does not come from a volume-preserving diffeomorphism of the fluid container, that is, $`\eta ^\sigma (t,)`$ is not a volume-preserving map.
#### The Averaged Lagrangian.
We define the averaged Lagrangian $`L^\alpha `$ by
$`L^\alpha (u)`$ $`={\displaystyle \frac{1}{2}}{\displaystyle _M}{\displaystyle \frac{1}{\alpha }}{\displaystyle _{\alpha /2}^{\alpha /2}}{\displaystyle _{S_+^{n1}}}u^{ϵ,\theta }(t,x),u^{ϵ,\theta }(t,x)𝑑ϵ\nu (\theta )\mu (x)`$
$`={\displaystyle \frac{1}{2}}{\displaystyle _M}{\displaystyle \frac{1}{\alpha }}{\displaystyle _{\alpha /2}^{\alpha /2}}{\displaystyle _{S_+^{n1}}}\left|\xi _{}^{ϵ,\theta }{}_{}{}^{}(uw^{ϵ,\theta })(t,x)\right|^2𝑑ϵ\nu (\theta )\mu (x),`$
where $`\mu `$ is the Riemannian volume form on $`M`$, and $`\nu `$ is the induced Riemannian volume form on $`S_+^{n1}`$, the upper hemisphere of the unit sphere $`S^{n1}`$ in $`^n`$.
#### Comments on the Nature of the Fuzzying Operation.
By using the upper hemisphere $`S_+^{n1}`$ and integrating from $`\alpha /2`$ to $`\alpha /2`$, we are tacitly assuming that there is a hyperplane of symmetry in the $`\theta `$-parameter space. This is not a restriction even near the boundary, as the hyperplane of symmetry can always be chosen orthogonal to the boundary and the maps $`\xi _t^{ϵ,\theta }`$ can be chosen to be symmetric about this hyperplane with respect to the radial parameter $`ϵ`$.
The reader should keep in mind that the variables $`\theta `$ and $`ϵ`$ parameterize possible families of maps and are not to be confused with spatial spheres in the flow itself. We are averaging over these families of maps and not literally over spatial regions. A representative of the family of fluctuation maps $`\xi ^{ϵ,\theta }`$ in the two dimensional case and near a boundary is shown in Figure 2.1.
The internal structure behind the fuzzyness of the macroscopic Lagrangian flow<sup>4</sup><sup>4</sup>4For a general continuum, the information about the structure of the representative sampling would be encoded in the fluctuation maps. For example, this might include defects or microstructure. is completely encoded in the fluctuation maps $`\xi _t^{ϵ,\theta }`$.
The zeroth-order assumption that these maps are simply the identity map leads to the zeroth-order Lagrangian $`L_0^\alpha `$ which is exactly equal to the Lagrangian $`L`$ given in (1.3) and thus produces the usual Euler equations of hydrodynamics as the continuum model. We proceed to obtain the first order correction to this model which accounts for the spatial fluctuations.
#### Asymptotic Expansion.
We Taylor expand $`u^{ϵ,\theta }`$ in $`ϵ`$ about $`ϵ=0`$ to obtain
$$u^{ϵ,\theta }(t,x)=\xi _{t}^{ϵ,\theta }{}_{}{}^{}(u_tw_t^{ϵ,\theta })(x)=u_t(x)+ϵ\mathrm{\pounds }_{\xi ^{}(\theta ,t,x)}u_t(x)ϵ\dot{\xi }^{}(\theta ,t,x)+O(ϵ^2),$$
(2.4)
where the overdot means the time derivative. This follows from the definition of the Lie derivative, the fact that $`w^{ϵ,\theta }=_t\xi _t^{ϵ,\theta }\xi _t^{ϵ,\theta }`$, and that $`\xi _t^{ϵ,\theta }|_{ϵ=0}=e`$. Using the zero-torsion condition on the Levi-Civita covariant derivative, $`\mathrm{\pounds }_\xi ^{}u=_\xi ^{}u_u\xi ^{}`$, and suppressing the dependence on $`t`$ and $`x`$, we get
$$u^{ϵ,\theta }=u+ϵ(u\xi ^{}(\theta )\xi ^{}(\theta )u\dot{\xi }^{}(\theta ))+O(ϵ^2)$$
or, in index notation,
$$u_{}^{ϵ}{}_{}{}^{i}=u^i+ϵ\left(u_{;j}^i\xi ^j(\theta )\xi _{;j}^i(\theta )u^j\dot{\xi }^i(\theta )\right)+O(ϵ^2),$$
where
$$u=u^i_i,\xi ^{}=\xi ^i_i\text{and}u\xi ^{}=u_{;j}^i\xi ^j_i.$$
In order to proceed, we make the first-order Taylor Hypothesis that the infinitesimal fluctuation vector $`\xi ^{}`$ is frozen, as a one-form, into the fluid so that its Lie transport vanishes; namely,
$$(\dot{\xi }_t^{})^{\mathrm{}}+\mathrm{\pounds }_u(\xi ^{})^{\mathrm{}}=0.$$
(2.5)
We again express the Lie derivative of the $`1`$-form field $`(\xi ^{})^{\mathrm{}}`$ in terms of the covariant derivative to obtain, in index notation,
$$\dot{\xi }_i+u_{;i}^j\xi _j+\xi _{i;j}u^j=0.$$
From this hypothesis, the $`O(ϵ)`$ term in the Taylor expansion (2.4) is
$`u_{i;j}\xi _{i;j}u^j+u_{;i}^j\xi _j+\xi _{i;j}u^j`$ $`=u_{i;j}\xi ^j+u_{;i}^j\xi _j`$
$`=u_{i;j}\xi ^j+u_{j;i}\xi ^j=2\mathrm{Def}u^{\mathrm{}}\xi ^{}(\theta ).`$
It follows that
$$u^{ϵ,\theta }=(\xi ^{ϵ,\theta })^{}(uw^ϵ)=u+2ϵ\mathrm{Def}u\xi ^{}(\theta )+O(ϵ^2).$$
(2.6)
Substitution of (2.6) into the averaged Lagrangian $`L^\alpha `$ yields
$`L^\alpha (u)={\displaystyle \frac{1}{2}}{\displaystyle _M}{\displaystyle \frac{1}{\alpha }}{\displaystyle _{S_+^{n1}}}{\displaystyle _{\alpha /2}^{\alpha /2}}[|u(x)|^2+2ϵu(x),\mathrm{Def}u(x)\xi ^{}(x,\theta )`$
$`+\mathrm{\hspace{0.33em}4}ϵ^2|\mathrm{Def}u(x)\xi ^{}(x,\theta )|^2+O(ϵ^3)]dϵ\nu (\theta )\mu (x).`$ (2.7)
An important point about this expression is the following: There is no contribution from the term $`u,O(ϵ^2)`$ to the energy at order $`O(ϵ^2)`$. In fact, the $`O(ϵ^2)`$ term in (2.6) has the form $`O(ϵ^2)=ϵ^2\left(\dot{\xi ^{\prime \prime }}+R\right)`$, where
$$\dot{\xi ^{\prime \prime }}:=\frac{d}{dt}\frac{d^2\xi }{dϵ^2}|_{ϵ=0}.$$
However, $`\xi ^{\prime \prime }`$ is an independent field and must have its own dynamics specified. We assume that this dynamics is chosen so that $`\dot{\xi }^{\prime \prime }+R`$ is $`O(ϵ)`$ and so the a priori $`O(ϵ^2)`$-term in (2.6) is in fact $`O(ϵ^3)`$.
Integrating (2.7) in $`ϵ`$, rescaling $`\alpha \sqrt{\alpha /6}`$, and defining the symmetric rank-$`2`$ contravariant spatial fluctuation tensor (indices up) $`F`$ by
$$F(x)=_{S_+^{n1}}\xi ^{}(x,\theta )\xi ^{}(x,\theta )\nu (\theta ),$$
we obtain the first-order averaged Lagrangian
$`L_1^\alpha (u,F)`$ $`={\displaystyle \frac{1}{2}}{\displaystyle _M}{\displaystyle _{S_+^{n1}}}\left[|u|^2+2\alpha ^2|\mathrm{Def}u\xi ^{}(x,\theta )|^2\right]\nu (\theta )\mu (x)`$
$`={\displaystyle \frac{1}{2}}{\displaystyle _M}\left[|u(x)|^2+2\alpha ^2F(x)\mathrm{Def}u(x),\mathrm{Def}u(x)\right]\mu (x)`$
$`={\displaystyle \frac{1}{2}}{\displaystyle _M}\{|u(x)|^2+2\alpha ^2[g(x)F(x)]:[\mathrm{Def}u(x)\mathrm{Def}u(x)]\}\mu (x).`$ (2.8)
In coordinate notation, the first-order averaged Lagrangian takes the form
$$L_1^\alpha (u,F)=\frac{1}{2}_M\left\{g_{ik}u^iu^k+2\alpha ^2g_{ik}F^{jl}[\mathrm{Def}u]_j^i[\mathrm{Def}u]_l^k\right\}[\text{det }g]^{\frac{1}{2}}𝑑x.$$
The first-order averaged Lagrangian $`L_1^\alpha `$ is a function of the macroscopic Eulerian velocity field $`\mathrm{u}`$ and the contravariant spatial fluctuation tensor $`\mathrm{F}`$.
#### The Isotropic Case.
In the case that the fluctuation tensor is isotropic so that
$$F(x)=g^1(x),$$
the isotropic first-order averaged Lagrangian $`L_{1,\text{iso}}^\alpha `$ is given by
$$L_{1,\mathrm{iso}}^\alpha (u)=\frac{1}{2}_M\left[|u(x)|^2+2\alpha ^2|\mathrm{Def}u(x)|^2\right]\mu (x).$$
In this special case, the Lagrangian depends only on the Eulerian velocity field $`u`$ and no semi-direct product theory is required; in fact, the standard Euler-Poincaré theory for reduced Lagrangian variational principals may be invoked to obtain the isotropic averaged Euler equations as
$$\begin{array}{c}_t(1\alpha ^2\mathrm{\Delta }_r)u+_u(1\alpha ^2\mathrm{\Delta }_r)u\alpha ^2(u)^t\mathrm{\Delta }_ru=\mathrm{grad}p,\\ \text{div }u=0,u(0)=u_0,\end{array}$$
(2.9)
where $`\mathrm{\Delta }_r=(d\delta +\delta d)+2\mathrm{Ric}`$ (see Shkoller ). As we stated above, the equations (2.9) are precisely the equations of inviscid second-grade non-Newtonian fluids, and exactly coincide with Chorin’s vortex blob algorithm when a particular choice of smoothing kernel is used (see Oliver and Shkoller ). In the case that $`M`$ is a manifold without boundary, the incompressibility of the fluid allows us to replace the term $`2\alpha ^2|\mathrm{Def}u|^2`$ with $`\alpha ^2|u|^2`$ in (2.8), and still obtain the identical evolution equations as in (2.9); however, for domains with boundary it is essential to retain the strain tensor $`\mathrm{Def}u`$ in the Lagrangian so as to obtain the natural boundary conditions which ensure ellipticity of the operator $`(1\alpha ^2\mathrm{\Delta }_r)`$.
## 3 The Variational Principle and Semidirect products
We shall next explain the sense in which the Lagrangian $`L_1^\alpha (u,F)`$ defined in (2.8), a function of spatial variables, can be obtained from a Lagrangian defined in material variables. This will be done via an Euler-Poincaré procedure, which involves the group $`𝒟_{\mu ,D}^s\mathrm{}C^{\mathrm{}}(T^{2,0}(M))`$, the semi-direct product of the volume-preserving diffeomorphism group $`𝒟_{\mu ,D}^s`$ (with Dirichlet boundary conditions) and the smooth sections of the vector bundle $`T^{2,0}(M)`$, consisting of second-rank contravariant symmetric tensors. Before proceeding to our specific example, we shall digress briefly to explain the general theory.
### 3.1 Lagrangian Semidirect Product Theory.
#### The General Set Up.
Let $`V`$ be a vector space and assume that the Lie group $`G`$ acts linearly on the right on $`V`$ (and hence $`G`$ also acts on its dual space $`V^{}`$). In the case that the vector space $`V`$ consists of sections of a vector bundle $`E`$, $`V^{}`$ will denote the sections of the dual bundle $`E^{}`$. The semidirect product $`S=G\mathrm{}V`$ is the Cartesian product $`S=G\times V`$ whose group multiplication is given by
$$(g_1,v_1)(g_2,v_2)=(g_1g_2,v_2+v_1g_2),$$
(3.1)
where the action of $`gG`$ on $`vV`$ is denoted simply as $`vg`$. The Lie algebra of $`S`$ is the semidirect product Lie algebra, $`𝔰=𝔤\mathrm{}V`$, whose bracket is
$$[(\psi _1,v_1),(\psi _2,v_2)]=([\psi _1,\psi _2],v_1\psi _2v_2\psi _1),$$
(3.2)
where we denote the induced action of $`𝔤`$ on $`V`$ by concatenation, as in $`v_2\psi _1`$. For $`vV`$ and $`aV^{}`$, define the bilinear operator $`va𝔤^{}`$ by
$$va,\eta =a\eta ,v.$$
#### The Objects in Our Case.
We choose $`G`$ to be the topological group $`𝒟_{\mu ,D}^s`$. While this is not a Lie group, right multiplication is a smooth operation, and this is the crucial feature we shall make use of. The tangent space at the identity $`T_e𝒟_{\mu ,D}^s`$ is equal to $`𝔛_{\mathrm{div},D}^s`$, the $`H^s`$ vector fields on $`M`$ vanishing on the boundary and with zero divergence, and plays the role of the Lie algebra $`𝔤`$.
We set $`V=H^s(T^{2,0}(M))`$, the $`H^s`$ sections of the vector bundle $`T^{2,0}(M)`$ consisting of contravariant symmetric two tensors (indices down). Thus, the vector space $`V^{}`$ is $`H^s(T^{0,2}(M))`$, the $`H^s`$ sections of the covariant two tensors (indices up). The duality is with respect to the induced Riemannian metric on $`T^{2,0}(M)`$. The topological group $`𝒟_{\mu ,D}^s`$ acts on the vector space $`H^s(T^{2,0}(M))`$ by pull-back; hence, this action takes values in $`H^{s1}(T^{2,0}(M))`$. Since the group is volume preserving, the induced right action on $`V^{}`$ is also by pull-back. We have the map $`(\eta ,F)\eta ^{}F`$.
It follows that the infinitesimal action is by the Lie derivative which also maps $`H^s`$ sections into sections of class $`H^{s1}`$. Thus, according to the above definition, the diamond operator is computed as follows: Let $`KH^{s1}(T^{2,0}(M))`$, $`FV^{}=H^s(T^{0,2}(M))`$ and let $`u𝔤=𝔛_{\mathrm{div},D}^s`$. We define the operator
$$_F:𝔛_{\mathrm{div},D}^sH^{s1}(T^{2,0}(M)),𝓛_F(u)=\mathrm{\pounds }_uF.$$
Then the adjoint operator (with respect to the Riemannian metric on $`H^s(T^{2,0}(M))`$) $`𝓛_F^{}:H^{s1}(T^{0,2}(M))𝔛_{\mathrm{div},D}^s`$ and is defined by
$$KF,u=\mathrm{\pounds }_uF,K=u,𝓛_F^{}K.$$
Thus, we have
$$KF=𝓛_F^{}K.$$
#### Semidirect Euler-Poincaré Reduction.
Assume we have a right $`G`$-invariant function $`:TG\times V^{}`$. For $`a_0V^{}`$, let $`_{a_0}:TG`$ be given by $`_{a_0}(v_g)=(v_g,a_0)`$, so $`_{a_0}`$ is right invariant under the lift to $`TG`$ of the right action of $`G_{a_0}`$ on $`G`$, where $`G_{a_0}`$ is the isotropy group of $`a_0`$. Define $`L:𝔤\times V^{}`$ by
$$L(v_gg^1,a_0g^1)=(v_g,a_0).$$
For a curve $`g(t)G,`$ let $`\xi (t):=\dot{g}(t)g(t)^1`$ and let $`a(t)=a_0g(t)^1`$, which is the unique solution of the equation $`\dot{a}(t)=a(t)\xi (t)`$ with initial condition $`a(0)=a_0`$.
In our setting, $`a_0=F_0C^{\mathrm{}}(T^{2,0}(M))`$,
$$v_g=u_\eta \{vH^s(M,TM)v\eta ^1𝔛_{\mathrm{div},D}^sH_0^1(TM),\eta 𝒟_{\mu ,D}^s\},$$
and $`L(u_\eta \eta ^1,\eta ^{}F_0)=(u_\eta ,F_0)`$ where $`L`$ is given by (2.8).
###### Theorem 3.1.
The following are equivalent:
1. Hamilton’s variational principle
$$\delta _{t_1}^{t_2}_{a_0}(g(t),\dot{g}(t))𝑑t=0$$
(3.3)
holds, for variations $`\delta g(t)`$ of $`g(t)`$ vanishing at the endpoints.
2. $`g(t)`$ satisfies the Euler–Lagrange equations for $`_{a_0}`$ on $`G`$.
3. The constrained variational principle
$$\delta _{t_1}^{t_2}L(\xi (t),a(t))𝑑t=0$$
(3.4)
holds on $`𝔤\times V^{}`$, using variations of the form
$$\delta \xi =\dot{\eta }[\xi ,\eta ],\delta a=a\eta ,$$
(3.5)
where $`\eta (t)𝔤`$ vanishes at the endpoints.
4. The Euler–Poincaré equations hold on $`𝔤\times V^{}`$
$$\frac{d}{dt}\frac{\delta L}{\delta \xi }=\mathrm{ad}_\xi ^{}\frac{\delta L}{\delta \xi }\frac{\delta L}{\delta a}a.$$
(3.6)
### 3.2 Computation of the Anisotropic Averaged Euler Equations
It is convenient to define the linear operator $`𝒞:𝔛_{\mathrm{div}}^sH_0^1H^{s2}`$, $`s1`$, mapping divergence-free vector fields to vector fields, by
$$𝒞u:=\mathrm{Div}\left[C:u^{\mathrm{}}\right],$$
where $`\mathrm{}`$ is the map from vector fields to one-forms associated with the metric $`g`$, and again the fourth-rank symmetric positive tensor $`C`$ is the symmetrization of the tensor $`Fg^1`$, given in local coordinates by
$$C^{ijkl}=\frac{1}{4}\left(F^{lj}g^{ik}+f^{kj}g^{il}+F^{li}g^{jk}+F^{ki}g^{jl}\right).$$
The functional derivatives of $`L_1^\alpha `$ with respect to $`u`$ and $`F`$ are given by
$$\frac{\delta L_1^\alpha }{\delta u}=(1\alpha ^2𝒞)u$$
and
$$\frac{\delta L_1^\alpha }{\delta F}=2\alpha ^2\left[\mathrm{Def}u\right]^2.$$
We can then compute that
$$\frac{\delta L_1^\alpha }{\delta F}F=2\alpha ^2F:[(\mathrm{Def}u^{\mathrm{}})^2]^{\mathrm{}}4\alpha ^2\mathrm{Div}(\left(\mathrm{Def}u\right)^2F)$$
Letting $`t=(\mathrm{Def}u)^2`$, in index notation, we get
$$\left[\frac{\delta L_1^\alpha }{\delta F}F\right]_k=2\alpha ^2F^{ij}t_{ij;k}4\alpha ^2\left[F^{ij}t_{kj}\right]_{;i}.$$
Using Theorem 3.1, we derive the following result.
###### Theorem 3.2.
The Euler-Poincaré equations on Riemannian manifolds, associated to the Lagrangian $`L_1^\alpha `$ given by (2.8), are the following anisotropic averaged Euler equations:
$`_t(1\alpha ^2𝒞)u`$ $`+_u(1\alpha ^2𝒞)u\alpha ^2[u]^t𝒞u+2\alpha ^2F:[(\mathrm{Def}u^{\mathrm{}})^2]^{\mathrm{}}`$
$`4\alpha ^2\mathrm{Div}\left(\left(\mathrm{Def}u\right)^2F\right)=\mathrm{grad}p`$ (3.7)
together with the advection equation
$$_tF+\mathrm{\pounds }_uF=0,$$
(3.8)
the incompressibility constraint $`\text{div }u=0`$, initial data $`u(0)=u_0`$ and $`F(0)=F_0`$, and no-slip conditions $`u=0`$ on the boundary.
#### Anisotropic Averaged Euler Equations in General Coordinates.
In general coordinates on a manifold, the averaged Euler equations read
$`_t(u^i\alpha ^2[C^{ijkl}u_k,_j],_l)+(u^i\alpha ^2[C^{ijkl}u_k,_j],_l),_mu^m\alpha ^2u_m,_i[C^{mjkl}u_k,_j],_l`$
$`+2\alpha ^2F^{kj}[D_{km}g^{mn}D_{nj}],_i4\alpha ^2[F^{kj}D_{im}g^{mn}D_{nj}],_k`$
$`=p,_i`$
where, as earlier, $`D_{ij}=\frac{1}{2}\left(u_{i;j}+u_{j;i}\right)`$ is the rate of deformation tensor and indices are raised and lowered using the metric tensor (which of course need not be diagonal in general coordinates), and $`C^{ijkl}=\frac{1}{4}\left(F^{lj}g^{ik}+f^{kj}g^{il}+F^{li}g^{jk}+F^{ki}g^{jl}\right)`$. In Euclidean space, one need only set the components of the metric tensor $`g_{ij}`$ to the Kronecker delta $`\delta _{ij}`$.
#### Comments on the Form of the Equations.
In 2D, identifying $`F`$ with the vector $`(F^{11},F^{12},F^{22})`$, equation (3.8) takes the form
$$\frac{D}{dt}\left[\begin{array}{cccc}F^{11}& & & \\ F^{12}& & & \\ F^{22}& & & \end{array}\right]=\left[\begin{array}{cccc}2u_{,1}^1& 2u_{,2}^1& 0& \\ u_{,1}^2& 0& u_{,2}^1& \\ 0& 2u_{,1}^2& 2u_{,1}^1& \end{array}\right]\left[\begin{array}{cccc}F^{11}& & & \\ F^{12}& & & \\ F^{22}& & & \end{array}\right],$$
(3.9)
where $`D/dt`$ denotes $`_t+(u)`$. Notice that the matrix on the right-hand-side of (3.9) is traceless; a similar form holds in 3D as well. This is not surprising, since by virtue of the incompressibility of the Lagrangian flow and the fact that $`F_t=\eta _t^{}F_0`$, we have that
$$det(F_t)=det(F_0),$$
for all $`t`$ for which the solution exists. As consequence, the operator $`(1\alpha ^2𝒞)`$ remains uniformly elliptic, if $`F_0`$ is strictly positive.
#### The Circulation Theorem.
Let $`\gamma :S^1M`$ be a loop and let $`\gamma _t=\eta _t\gamma `$ denote the evolution of the loop moving with the fluid.
###### Theorem 3.3.
For a solution of the anisotropic averaged Euler equations, we have
$$\frac{d}{dt}_{\gamma _t}(1\alpha ^2𝒞^{\mathrm{}})u^{\mathrm{}}=2\alpha ^2_{\gamma _t}[2\mathrm{Div}(\left(\mathrm{Def}u\right)^2F)F:[(\mathrm{Def}u^{\mathrm{}})^2]^{\mathrm{}}]^{\mathrm{}}.$$
This follows directly from the Lie derivative form of the equations given in (1.7). We note that if one were to make use of the general Kelvin-Noether theorem given in Holm et al. \[1998b\], one would arrive at the same result.
## 4 Analytic Properties
In this section we prove well-posedness and other properties of the solutions by showing that these equations are given by a smooth vector field in material representation in the appropriate Sobolev topologies. This is in line with what is known about the Euler equations, as described in the introduction. We also discuss the corrector for the equations.
### 4.1 Well-posedness of Classical Solutions
We shall prove existence, uniqueness, and smooth dependence on initial data on finite time intervals for solutions of the anisotropic averaged Euler equations. For simplicity, we shall restrict the fluid domain $`M`$ to be a compact subset of Euclidean space with smooth boundary, although our methods can be applied to Riemannian manifolds.
We begin by collecting some preliminary results. Set $`𝒱^s=H_0^1H^s`$ and $`𝒱_\mu ^s=H_0^1𝔛_{\mathrm{div}}^s`$. Also, let $`𝒟_D^s`$ denote the $`H^s`$ class diffeomorphisms which fix the boundary, and again let $`𝒟_{\mu ,D}^s`$ denote the diffeomorphisms in $`𝒟_D^s`$ which preserve the volume $`\mu `$.
###### Lemma 4.1.
For $`u𝒱_\mu ^s`$, $`s>1`$,
$$_t𝒞u=𝒞(_tu)+\mathrm{Div}\left[_uFu+uuF+uFu^t\right],$$
and
$`_u𝒞u`$ $`=𝒞(_uu)+u_u\mathrm{Div}F+u:_uF2u:(uF)`$
$`u(u\mathrm{Div}F)u(u:F).`$
###### Proof.
The proof is a simple computation which we leave to the interested reader, c.f. Lemma 3 in Shkoller . ∎
Set $`=\mathrm{Def}^{}\left[\left(gF\right):\mathrm{Def}\right]`$. Then $``$ is a positive unbounded self-adjoint operator on $`L^2`$ with domain $`𝒱_\mu ^2`$. Define the inner-product $`(,)`$ on $`𝒱_\mu ^2`$ by
$$(u,v)=(1\alpha ^2)u,v_{L^2}.$$
For $`s>n/2+1`$, $`(,)`$ defines an inner-product on $`T_e𝒟_{\mu ,D}^s`$, the tangent space at the identity of the subgroup $`𝒟_{\mu ,D}^s`$ consisting of those elements of $`𝒟_\mu ^s`$ which restrict to the identity on $`M`$. Right-translating $`(,)`$ to the entire group $`𝒟_{\mu ,D}^s`$ defines a $`C^{\mathrm{}}`$ weak Riemannian metric by Proposition 3 of Shkoller .
###### Proposition 4.2.
For $`r1`$ we have the following well-defined decomposition
$$𝒱^r=𝒱_\mu ^r(1)^1\mathrm{grad}H^{r1}(M).$$
(4.1)
Thus, if $`F𝒱^r`$, then there exists $`(v,p)𝒱_\mu ^r\times H^{r1}(M)/`$ such that
$$F=v+(1)^1\mathrm{grad}p$$
and the pair $`(v,p)`$ are solutions of the Stokes problem
$$\begin{array}{c}(1)v+\mathrm{grad}p=(1)F,\\ \mathrm{div}v=0,\\ v=\text{on }M.\end{array}$$
(4.2)
The summands in (4.1) are $`(,)`$-orthogonal. Now, define the Stokes projector
$$\begin{array}{c}𝒫_e:𝒱^r𝒱_\mu ^r,\\ 𝒫_e(F)=F(1)^1\mathrm{grad}p.\end{array}$$
(4.3)
Then, for $`s>(n/2)+1`$, $`\overline{𝒫}:T𝒟_D^sT𝒟_{\mu ,D}^s`$, given on each fiber by
$$\begin{array}{c}\overline{𝒫}_\eta :T_\eta 𝒟_D^sT_\eta 𝒟_{\mu ,D}^s,\\ \overline{𝒫}_\eta (X_\eta )=\left[𝒫_e(X_\eta \eta ^1)\right]\eta ,\end{array}$$
is a $`C^{\mathrm{}}`$ bundle map covering the identity.
###### Proof.
The proof is identical to the proof of Proposition 2 in Shkoller . ∎
###### Theorem 4.3.
Set $`s>(n/2)+2`$, and let $`,`$ denote the right invariant metric on $`𝒟_{\mu ,D}^s`$ given at the identity by $`(,)`$. For $`u_0T_e𝒟_{\mu ,D}^s`$ and $`F_0C^{\mathrm{}}(T^{2,0})`$, there exists an interval $`I=(T,T)`$, depending on $`|u_0|_s`$, and a unique geodesic $`\dot{\eta }`$ of $`,`$ with initial data $`\eta (0)=e`$ and $`\dot{\eta }(0)=u_0`$ such that
$$\dot{\eta }\text{ is in }C^{\mathrm{}}(I,T𝒟_{\mu ,D}^s)$$
and has $`C^{\mathrm{}}`$ dependence on the initial velocity $`u_0`$.
The geodesic $`\eta `$ is the Lagrangian flow of the time-dependent vector field $`u(t,x)`$ given by
$$_t\eta (t,x)=u(t,\eta (t,x)),$$
and, with $`F(t,x)=\left(\eta _t\right)_{}F_0(x)`$,
$$(u,F)C^0(I,𝒱_\mu ^s)C^1(I,𝒱_\mu ^{s1})\times C^0(I,H^{s1}(T^{2,0})).$$
uniquely solves the anisotropic averaged Euler equations with Dirichlet boundary conditions $`u=0`$, and depends continuously on $`(u_0,F_0)`$.
###### Proof.
The key to the proof rests in the fact that the pair $`(u,F)`$ solves the anisotropic averaged Euler equations if and only if $`\eta `$ is a solution of
$$\ddot{\eta }+𝒰^\alpha (\dot{\eta }\eta ^1)\eta =\left[(1\alpha ^2)^1\mathrm{grad}p\right]\eta ,$$
(4.4)
where
$`𝒰^\alpha (u)=`$ $`\alpha ^2(1\alpha ^2)^1\{\mathrm{Div}[_uFuuuFuFu^t]`$
$`u(_u\mathrm{Div}F)u:_uF+2u:(uF)`$
$`+u(u\mathrm{Div}F)+u(u:F)u^t𝒞u`$
$`+2F:(\mathrm{Def}u^2)4\mathrm{Div}[F\mathrm{Def}u^2]\}.`$
This expression is obtained using Lemma 4.1. Now it is clear that $`𝒰^\alpha `$ maps $`H^s`$ vector fields into $`H^s`$ vector fields since $`H^{s2}`$ forms a multiplicative algebra, and since the fluctuation tensor $`F`$ at $`t=0`$ is given by $`F_0`$ which is $`C^{\mathrm{}}`$. In particular, in the Lagrangian frame, $`F`$ is frozen, so the elliptic operator $`[(1\alpha ^2\mathrm{\Delta })(\dot{\eta }\eta ^1)]\eta `$ has $`C^{\mathrm{}}`$ coefficients.
Thus, the proof of Theorem 2 in Shkoller gives a unique curve $`\eta C^{\mathrm{}}(I,𝒟_{\mu ,D}^s)`$ solving (4.4). That $`u`$ is only $`C^0`$ in time follows from the fact that the map $`\eta \eta ^1:𝒟_{\mu ,D}^s𝒟_{\mu ,D}^s`$ is only $`C^0`$. That $`F`$ is in $`C^0(I,H^{s1}(T^{2,0}))`$ follows from the regularity of $`u`$. ∎
### 4.2 A Corrector for the Macroscopic Velocity
The solution to the anisotropic averaged Euler equations (3.7) and (3.8) yields the pair $`(u,F)`$. The macroscopic spatial velocity field $`u`$ is only the zeroth-order term in the expansion (in $`ϵ`$) for the velocity field $`u^{ϵ,\theta }`$. We have computed, in equation (2.6), the expansion of $`u^{ϵ,\theta }`$ to order $`O(ϵ^2)`$ as
$$u^{ϵ,\theta }(t,x)=u(t,x)+2ϵ\mathrm{Def}(u)(t,x)\xi ^{}(t,x,\theta )+O(ϵ^2).$$
Since $`|ϵ|`$ is bounded by $`\alpha /2`$, we have that
$$u^{\alpha ,\theta }(t,x)=u(t,x)+\alpha \mathrm{Def}(u)(t,x)\xi ^{}(t,x,\theta )+O(\alpha ^2),$$
so that we may add the $`O(\alpha )`$ term to the expansion by solving for the infinitesimal fluctuation vector $`\xi ^{}(t,x,\theta )`$. This, however, only requires the solution of the simple linear advection problem (2.5) given by
$$\dot{\xi }^{}(t,x,\theta )^{\mathrm{}}+\mathrm{\pounds }_{u(t,x)}\xi ^{}(t,x,\theta )^{\mathrm{}}=0.$$
Computationally, this means that we may solve for the macroscopic velocity field $`u`$ at spatial scales larger than $`\alpha `$ and correct for the unresolved small scales to $`O(\alpha ^2)`$.
### 4.3 Limits of Zero Viscosity
Peskin showed that by perturbing the Euler solution’s Lagrangian particle trajectory $`\eta _t(x)`$ by Brownian motions and averaging over such perturbations, the Navier-Stokes equations are obtained. In other words, letting Euler trajectories take random-walks produces the viscosity term $`\nu \mathrm{\Delta }u`$, where $`\eta _t`$ is the flow map for the velocity field $`u`$. In the setting of the averaged Euler equations, the Lagrangian trajectory $`\eta _t(x)`$ of a particle $`x`$ corresponds to the flow of the velocity $`u(t,x)`$ solving the anisotropic averaged Euler equations. Thus, Peskin’s argument can be carried over in this setting to obtain the same viscous term $`\nu \mathrm{\Delta }u`$.
We are hence motivated to define the anisotropic averaged Navier-Stokes equations by
$`_t(1\alpha ^2𝒞)u^\nu `$ $`+_{u^\nu }(1\alpha ^2𝒞)u^\nu \alpha ^2[u^\nu ]^t𝒞u^\nu +2\alpha ^2F:[(\mathrm{Def}u_{}^{\nu }{}_{}{}^{\mathrm{}})^2]^{\mathrm{}}`$
$`4\alpha ^2\mathrm{Div}\left(\left(\mathrm{Def}u^\nu \right)^2F\right)=\mathrm{grad}p+\nu \mathrm{\Delta }u^\nu ,\nu >0`$ (4.5)
together with the advection for the fluctuation tensor $`F`$ given by (3.8), the incompressibility constraint $`\mathrm{div}u=0`$, initial data $`u(0)=u_0`$ and $`F(0)=F_0`$, and the no-slip conditions $`u=0`$ on the boundary.
Let $`\eta _t^\nu `$ denote the Lagrangian flow of the solution $`u^\nu `$ of the anisotropic averaged Navier-Stokes equations (4.5), and let $`\dot{\eta }^\nu `$ denote the partial time derivative of the flow, i.e., the material velocity field.
###### Theorem 4.4.
For $`s>(n/2)+2`$ and $`(u_0,F_0)𝒱_\mu ^s\times C^{\mathrm{}}(T^{2,0})`$, there exists a $`T>0`$, depending only on $`u_0_{H^s}`$ on not on the viscosity $`\nu `$, such that for each $`\nu >0`$
$$\dot{\eta }\text{ is in }C^{\mathrm{}}([0,T),T𝒟_{\mu ,D}^s),$$
and has $`C^{\mathrm{}}`$ dependence on the initial velocity field $`u_0`$. Furthermore, $`u_t^\nu =\dot{\eta }_t^\nu \eta _{}^{\nu }{}_{t}{}^{1}`$ is in $`C^0([0,T),𝒱_\mu ^s)C^r([0,T),𝒱_\mu ^{sr})`$ and depends continuously on $`u_0`$.
The proof follows the proof of Theorem 2 in Shkoller ; we refer the interested reader there for the details. As a consequence of the time interval $`[0,T)`$ of solutions $`u^\nu `$ being independent of $`\nu `$, we immediately obtain the following.
###### Corollary 4.5.
For $`s>(n/2)+2`$, solutions $`u^\nu `$ of (4.5) converge regularly to the inviscid solutions $`u`$ of (3.7) as $`\nu 0`$. Furthermore, letting $`u^\nu =_t\eta _\nu \eta _\nu ^1`$, the viscous Lagrangian flow $`\eta _\nu `$ converges regularly in the $`H^s`$ topology to the inviscid Lagrangian flow $`\eta =\eta ^0`$.
This result states that we can generate smooth-in-time classical solutions to the anisotropic averaged Euler equations by obtaining a sequence of viscous solutions and allowing $`\nu `$ to go to zero, and what is surprizing, this holds even in the presense of boundaries. Results of this type were conjectures in Marsden et al. and Barenblatt and Chorin \[1998a\] (see also Barenblatt and Chorin \[1998b\]).
In the case of the isotropic averaged Euler equations, Foias, Holm, and Titi have added the dissipative term $`\nu \mathrm{\Delta }(1\alpha ^2\mathrm{\Delta })u`$ instead of using $`\nu \mathrm{\Delta }u`$, and this is enough to give global in time classical solutions in dimension three. It is, however, the term $`\nu \mathrm{\Delta }u`$ that arises from either the approach of Peskin noted above, or from the constitutive theory approach of Rivlin and Erickson .
### Future Directions
There are several interesting directions
1. Of course numerical implementation for specific flows will be of great interest.
2. Modeling the mean velocity profile for turbulent flows in channels and pipes.
3. Specific flows and special solutions.
4. Links with elliptical vortex blob methods of Zabusky and coworkers (see, e.g., Melander et al. ) would be of interest to establish; it is reasonable to expect that solutions of this sort would provide the anisotropic analog of the vortex blob solutions of Oliver and Shkoller .
5. Further investigation of the vorticity formulation and its relation with the coadjoint orbit structure in the semidirect product for the Hamiltonian version of this theory.
#### Acknowledgments.
We thank G.I. Barenblatt for useful comments regarding his work on averaging for damage propagation and the connection to our work, Alexandre Chorin for useful discussions on links between our work and optimal prediction methods, and Albert Fannjiang on extensions to stochastic perturbation methods. We also thank Tudor Ratiu for his helpful comments. In particular, the important idea of advecting the fluctuations as a one-form rather than as a vector field (see equation (2.5)) was done in collaboration with him. We thank Darryl Holm for keeping us informed about his work on fluctuation effects.
JEM and SS were partially supported by the NSF-KDI grant ATM-98-73133. SS was partially supported by the Alfred P. Sloan Foundation Research Fellowship.
|
warning/0005/gr-qc0005011.html
|
ar5iv
|
text
|
# Chaplygin–like gas and branes in black hole bulks
## 1 Introduction
In this letter we will study branes in black hole bulks. We will consider mainly two examples, namely BHTZ and anti-de Sitter–Schwarzschild black holes, where we study classes of branes which are naturally adapted to the black hole geometry at hand.
The result we will find is that in both cases there is no possible choice of brane tension so that the geometry under consideration solves the corresponding Einstein equations in the bulk spacetime. It is necessary to go beyond the scheme of Randall and Sundrum by introducing matter on the brane which we will do by making use of perfect fluids. From a geometrical viewpoint, the reason for this new situation is that either the geometry (in the Schwarschild-AdS case) or the foliations (in the BHTZ case) of the bulks we are going to study are less symmetric than in the well-studied cases of . This excludes the possibility of writing the metric as a genuine warped manifold . Rather, we are considering a more general setting, that of “multi-warped” manifolds, in which different directions on the brane are multiplied by different functions of the transverse coordinate.
In the Schwarzschild-AdS case there is an exceptional choice of brane location where the possibility exists to consider Einstein equations including only pure brane tension as in and no other type of matter. This particular solution has also appeared recently in . However in our treatment it will appear clearly that this fact is rather a coincidence because the required equality between energy and opposite of the pressure is numerical but not functional.
The type of matter found in our investigation are worth commenting. Both in the BHTZ and Schwarzschild-AdS cases it is possible to identify state equations for the fluids on the branes which are universal (in the geometry considered) in the sense that they are not depending on the choice of brane location. The matter populating branes located in BHTZ bulks has a state equation that is particularly simple: the pressure has to be inversely proportional to minus the (positive) energy density. A fluid satisfying such a state equation is called “Chaplygin gas” . This result is rather interesting since this type of fluid is precisely of interest in the $`d`$-brane context . In the AdS-Schwarschild case the state equations we have found are more complicate, but retain the Chaplygin gas form near the horizon.
## 2 BHTZ black holes
In this section we concentrate on BHTZ black holes. These holes are most easily constructed by taking quotients of the three-dimensional anti–de Sitter spacetime
$$AdS_3=\{x^4:x_{}^{0}{}_{}{}^{2}x_{}^{1}{}_{}{}^{2}x_{}^{2}{}_{}{}^{2}+x_{}^{3}{}_{}{}^{2}=l^2\},$$
(1)
w.r.t. discrete groups of translations along suitable Killing vector fields . For this purpose, a two-parameter class of coordinate systems for $`AdS_3`$ is constructed in the following way :
$`x^0=l\sqrt{{\displaystyle \frac{r^2r_{}^2}{r_+^2r_{}^2}}}\mathrm{cosh}\left({\displaystyle \frac{r_+\varphi }{l}}{\displaystyle \frac{r_{}t}{l^2}}\right),`$
$`x^1=l\sqrt{{\displaystyle \frac{r^2r_{}^2}{r_+^2r_{}^2}}}\mathrm{sinh}\left({\displaystyle \frac{r_+\varphi }{l}}{\displaystyle \frac{r_{}t}{l^2}}\right),`$
$`x^2=l\sqrt{{\displaystyle \frac{r^2r_+^2}{r_+^2r_{}^2}}}\mathrm{cosh}\left({\displaystyle \frac{r_+t}{l^2}}{\displaystyle \frac{r_{}\varphi }{l}}\right),`$
$`x^3=l\sqrt{{\displaystyle \frac{r^2r_+^2}{r_+^2r_{}^2}}}\mathrm{sinh}\left({\displaystyle \frac{r_+t}{l^2}}{\displaystyle \frac{r_{}\varphi }{l}}\right),`$ (2)
with $`r>r_+r_{}`$. In these coordinates the AdS metric is written as follows:
$$ds^2=\frac{(r^2r_+^2)(r^2r_{}^2)}{l^2r^2}dt^2\frac{l^2r^2}{(r^2r_+^2)(r^2r_{}^2)}dr^2r^2\left(d\varphi \frac{r_+r_{}}{r^2l}dt\right)^2.$$
(3)
BHTZ black holes are then obtained by identifying those points parametrized by $`\varphi +2\pi k`$ with integer $`k`$ .
Let us introduce a new coordinate $`\chi `$ as follows:
$$\mathrm{cosh}\frac{2\chi }{l}=\frac{(r^2r_+^2)+(r^2r_{}^2)}{r_+^2r_{}^2},$$
(4)
which gives
$$d\chi =\frac{lrdr}{\sqrt{r^2r_+^2}\sqrt{r^2r_{}^2}}.$$
(5)
The use of the coordinate $`\chi `$ recasts the metric in a form which is convenient for a brane “insertion”:
$$ds^2=\frac{B^2\mathrm{sinh}^2\frac{2\chi }{l}}{A+B\mathrm{cosh}\frac{2\chi }{l}}\frac{dt^2}{l^2}\left(A+B\mathrm{cosh}\frac{2\chi }{l}\right)\left(\frac{\sqrt{A^2B^2}}{A+B\mathrm{cosh}\frac{2\chi }{l}}\frac{dt}{l}d\varphi \right)^2d\chi ^2,$$
(6)
where
$$2A=r_+^2+r_{}^2,\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}2}B=r_+^2r_{}^2.$$
(7)
It is clear from this expression that the coordinate $`\chi `$ can play the role of a transverse space-like coordinate in the bulk as in . The difference w.r.t the situation studied in is that the metric is now multi-warped, i.e. the remaining terms of $`ds^2`$ are multiplied by different functions of $`\chi `$.
Let us consider now a brane located at $`\chi =\chi _b`$. The geometry we are going to study is made up by two slices of the BHTZ black hole geometry glued together at the brane with a symmetry (orbifold-type) condition analogous to that exploited in . This geometry is described by the following components of the metric tensor (w.r.t. the chosen coordinate system):
$`g_{tt}=Bl^2\mathrm{cosh}{\displaystyle \frac{2u}{l}}Al^2,`$
$`g_{\varphi \varphi }=B\mathrm{cosh}{\displaystyle \frac{2u}{l}}A,`$
$`g_{t\varphi }=l^1\sqrt{A^2B^2},`$
$`g_{\chi \chi }=1.`$ (8)
where we have introduced $`u=\chi _b+|\chi \chi _b|`$. The contravariant components are the following:
$`g^{tt}={\displaystyle \frac{A+B\mathrm{cosh}\frac{2u}{l}}{B^2\mathrm{sinh}^2\frac{2u}{l}}}l^2,`$
$`g^{\varphi \varphi }={\displaystyle \frac{AB\mathrm{cosh}\frac{2u}{l}}{B^2\mathrm{sinh}^2\frac{2u}{l}}},`$
$`g^{t\varphi }={\displaystyle \frac{\sqrt{A^2B^2}}{B^2\mathrm{sinh}^2\frac{2u}{l}}},`$
$`g^{\chi \chi }=1.`$ (9)
The metric induced on the brane is obtained as the restriction of the components given in Eq. (8) (first fundamental form):
$`h_{tt}=Bl^2\mathrm{cosh}{\displaystyle \frac{2\chi _b}{l}}Al^2,`$
$`h_{\varphi \varphi }=B\mathrm{cosh}{\displaystyle \frac{2\chi _b}{l}}A,`$
$`h_{t\varphi }=l^1\sqrt{A^2B^2}.`$ (10)
Since the components of the bulk metric are not continuosly differentiable, Chistoffel’s symbols have finite jumps at $`\chi =\chi _b`$. These jumps give rise to the following Einstein tensor on the brane:
$$G_{ij}^{(b)}=\frac{4}{l^2}\mathrm{coth}\frac{2\chi _b}{l}h_{ij}+\stackrel{~}{G}_{ij}^{(b)},$$
(11)
where
$$\stackrel{~}{G}_{tt}^{(b)}=\frac{2B}{l^3\mathrm{sinh}\frac{2\chi _b}{l}},\stackrel{~}{G}_{\varphi \varphi }^{(b)}=\frac{2B}{l\mathrm{sinh}\frac{2\chi _b}{l}},\stackrel{~}{G}_{t\varphi }^{(b)}=0.$$
(12)
Outside the brane the Einstein tensor obviously coincides with the anti–de Sitter one. We have therefore the following structure:
$$G_{\mu \nu }=G_{\mu \nu }^{AdS}+\delta _\mu ^i\delta _\nu ^jG_{ij}^{(b)}\delta (\chi \chi _b).$$
(13)
Let us first of all consider the non-rotating case. This corresponds to $`2A=2B=r_+^2`$ and the metric is static: $`g_{t\varphi }=0`$ and therefore $`h_{t\varphi }=0`$. However, due to the presence of the term $`\stackrel{~}{G}^{(b)}`$, the brane tension alone is not enough to solve the corresponding Einstein equations and it is necessary to introduce matter on the brane. We will concentrate on perfect fluids (but other types of matter are worth investigating). The following relations on the brane are thus obtained:
$`G_{tt}^{(b)}=(\epsilon +p)u_t^2ph_{tt},`$
$`G_{\varphi \varphi }^{(b)}=(\epsilon +p)u_\varphi ^2ph_{\varphi \varphi },`$
$`G_{t\varphi }^{(b)}=(\epsilon +p)u_tu_\varphi =0.`$ (14)
The third equation calls for a static fluid, corresponding to the condition $`u_\varphi =0`$. The first two equations in the previous array become simply $`G_{tt}^{(b)}=\epsilon h_{tt}`$ and $`G_{\varphi \varphi }^{(b)}=ph_{\varphi \varphi }`$, which are solved by
$`\epsilon =+{\displaystyle \frac{2}{l}}\mathrm{tanh}{\displaystyle \frac{\chi _b}{l}}`$
$`p={\displaystyle \frac{2}{l}}\mathrm{coth}{\displaystyle \frac{\chi _b}{l}}`$ (15)
It is possible to write an equation of state which includes all the previous solutions:
$$p\epsilon =\frac{4}{l^2}.$$
(16)
This state equation is universal in the sense that it is the only one which has the same form for any brane location. It corresponds formally to the so-called Chaplygin gas which has recently raised a certain interest . The pressure of this fluid is negative and inversely proportional to the energy density. The interesting fact is that this type of fluid can be obtained also from the Nambu-Goto action for $`d`$-branes moving in a $`(d+2)`$-dimensional spacetime in the light-cone parametrization . For a string embedded in a three-dimensional flat spacetime the argument is so simple that we may reproduce it: the relevant Hamiltonian is
$$H=\frac{1}{2}\left[P^2+(_\sigma x)^2\right]𝑑\sigma ,$$
where $`x`$ is the only transversal coordinate. Let us introduce euristically the density $`\epsilon (x)=(_\sigma x)^1`$ and the velocity $`v=P`$. One of the equations of motion is then current conservation:
$$_t\epsilon +_x(\epsilon v)=0.$$
The other gives
$$\dot{P}=_\sigma ^2x=_\sigma x_x(1/\epsilon )$$
that is
$$\epsilon \dot{P}=\epsilon (_t+v_x)v=_x(1/\epsilon )$$
which is Euler equation for a fluid such that $`p=1/\epsilon `$.
Let us pass now to a short discussion of the rotating case. Einstein equations now give the following relations on the brane:
$`G_{tt}^{(b)}=(\epsilon +p)u_t^2ph_{tt},`$
$`G_{\varphi \varphi }^{(b)}=(\epsilon +p)u_\varphi ^2ph_{\varphi \varphi },`$
$`G_{t\varphi }^{(b)}=(\epsilon +p)u_tu_\varphi ph_{t\varphi }.`$ (17)
This system should be supplemented by the normalization condition
$$h^{tt}u_t^2+2h^{t\varphi }u_tu_\varphi +h^{\varphi \varphi }u_\varphi ^2=1.$$
(18)
After some algebraic computation the following solution is obtained:
$$u_t^2=\frac{r_+^2}{l^2}\mathrm{sinh}^2\frac{\chi _b}{l},u_\varphi ^2=r_{}^2\mathrm{sinh}^2\frac{\chi _b}{l}.$$
(19)
Now we can proceed as before and solve for the fluid energy density and pressure. The solution, and consequently the equation of state, are exactly the same as in the static case, see Eqs. (15,16). The corresponding full energy-momentum tensor is now not diagonal.
The coincidence of the energy and pressure obtained in the rotating and nonrotating cases deserves a further explanation. A closer look to Eq. (4) reveals that the brane does not really depends on the choice of $`r_+`$ and $`r_{}`$ and consequently of the corresponding Killing vectors. Indeed, from the viewpoint of the space in which the anti-de Sitter universe is embedded, the brane can be decribed by the following equations:
$$\{\begin{array}{c}x_{}^{0}{}_{}{}^{2}x_{}^{1}{}_{}{}^{2}+x_{}^{2}{}_{}{}^{2}x_{}^{3}{}_{}{}^{2}=l^2\mathrm{cosh}\frac{2\chi _b}{l}\hfill \\ x_{}^{0}{}_{}{}^{2}x_{}^{1}{}_{}{}^{2}x_{}^{2}{}_{}{}^{2}+x_{}^{3}{}_{}{}^{2}=l^2\hfill \end{array}$$
(20)
This means that we could have studied such a brane in the anti-de Sitter bulk before making identifications along the Killing vectors. The simplest choice amounts to studying a bulk spacetime with interval
$$ds^2=\mathrm{sinh}^2udt^2\mathrm{cosh}^2ud\varphi ^2d\chi ^2$$
(21)
which describes the geometry of two regions of the anti de Sitter manifold glued together at a two-dimensional brane having the topology of a plane. The BHTZ procedure then changes both the bulk and the brane topology but has no consequence on the energy-momentum tensor of the matter living on the brane.
It is interesting to note that the brane is obtained in Eq. (20) as the intersection of two AdS spacetimes embedded in $`^4`$. Of course this is peculiar to two-dimensional branes. The present study could in principle be extended to other dimensionalities , but in this case the fluid would not be isotropic.
## 3 Schwarzschild-AdS black holes
We now examine the situation for the Schwarzschild-AdS metric , as recently studied also in . We do this in the usual case in which the bulk spacetime is five-dimensional and the brane four-dimensional. The Schwarzschild-AdS metric is given by
$$ds^2=\left(1+\frac{r^2}{l^2}\frac{2M}{r^2}\right)dt^2\frac{1}{1+\frac{r^2}{l^2}\frac{2M}{r^2}}dr^2r^2d^2\mathrm{\Omega }^{(3)},$$
(22)
where $`d^2\mathrm{\Omega }^{(3)}`$ is the metric of a unit three-dimensional spherical surface. As before we may introduce a radial coordinate $`\chi `$ by
$$\mathrm{cosh}\frac{2\chi }{l}=\frac{2r^2+l^2}{\sqrt{8Ml^2+l^4}},$$
(23)
and insert a “brane” at $`\chi =\chi _b(r_b)`$ by requiring the same symmetry properties of the metric w.r.t. $`\chi _b`$ as in the previous section. The brane is now substantially a copy of a Einstein static universe. The Einstein tensor includes a term proportional to $`\delta (\chi \chi _b)`$, which should be matched by matter on the brane so as to compensate it. Again, a perfect fluid with the following energy density and pressure does the job:
$`\epsilon ={\displaystyle \frac{6}{r_b}}\sqrt{1+{\displaystyle \frac{r_b^2}{l^2}}{\displaystyle \frac{2M}{r_b^2}}},`$ (24)
$`p={\displaystyle \frac{4\left(1+\frac{3r_b^2}{2l^2}\frac{M}{r_b^2}\right)}{r_b\sqrt{1+\frac{r_b^2}{l^2}\frac{2M}{r_b^2}}}}.`$ (25)
The state equation now depends on the black hole mass $`M`$:
$$p=\frac{\epsilon }{3}\frac{24}{\epsilon l^2}\frac{12}{\epsilon r_b^2(\epsilon )}$$
(26)
where $`r_b^2(\epsilon )`$ is obtained by solving Eq. (24). Two regimes are possible: if $`0<\epsilon <6/l`$ there is only a solution, given by
$`r_b^2(\epsilon )={\displaystyle \frac{18l^2}{36\epsilon ^2l^2}}+\sqrt{\left({\displaystyle \frac{18l^2}{36\epsilon ^2l^2}}\right)^2+{\displaystyle \frac{72Ml^2}{36\epsilon ^2l^2}}}`$ (27)
When $`\epsilon >6/l`$ the following two solutions are possible:
$`r_b^2(\epsilon )={\displaystyle \frac{18l^2}{\epsilon ^2l^236}}\pm \sqrt{\left({\displaystyle \frac{18l^2}{\epsilon ^2l^236}}\right)^2{\displaystyle \frac{72Ml^2}{\epsilon ^2l^236}}}`$ (28)
In this regime there is a special value $`r_b^2=4M`$ where to locate the brane. Here, the energy density $`\epsilon `$ has a maximum value
$$\epsilon _{\mathrm{max}}=\frac{6}{l}\sqrt{1+\frac{l^2}{8M}},$$
(29)
and the two solutions coalesce; beyond this value of the energy density there is no solution (see figure 1).
Since in this very case one also has that $`\epsilon =p`$ numerically, the fluid can be alternatively interpreted as tension of the brane as in the original work by Randall and Sundrum . This particular situation has been recently considered by . However, in our scheme it appears clearly that this case is rather special and in any other location one is obliged to add other matter contributions on the brane.
It is worthwhile to remark that the state equation (26) reduces to the Chaplygin equation when the brane is located near the horizon. For $`r_b^2>4M`$ there exists also the possibility to think of two distinguished types of matter on the brane, namely brane tension and a standard fluid having both energy density and pressure positive.
When $`l`$ tends to infinity we are describing a brane in a Schwarzschild bulk. Eqs. (24), (25), (26) and (28) can be easily reduced to this case; the situation is substantially similar to that we have described above in the Schwarschild-AdS case and we do not reproduce the results.
On the other side, when $`M`$ is set to zero the metric (22) reduces to the metric of an empty AdS spacetime and we are back to the Randall-Sundrum setting. However, by putting a brane at some value $`r=r_b`$ of the chosen coordinates we are exploring the different case in which the brane has the four-geometry of an Einstein static universe. The constant $`r_b`$ has therefore also the meaning of the radius of this universe. Also in this case we need to add matter beyond brane tension and this because of the brane geometry. Energy and pressure of the matter populating this universe have a comparatively simple form:
$`\epsilon ={\displaystyle \frac{6}{r_b}}\sqrt{1+{\displaystyle \frac{r_b^2}{l^2}}},`$ (30)
$`p={\displaystyle \frac{4\left(1+\frac{3r_b^2}{2l^2}\right)}{r_b\sqrt{1+\frac{r_b^2}{l^2}}}}.`$ (31)
The state equation is now as follows:
$$p=\frac{2\epsilon }{3}\frac{12}{\epsilon l^2}.$$
(32)
When the radius $`r_b`$ of the brane tends to infinity we are finally back to the original Randall-Sundrum case . Indeed in this limit we have that numerically
$$\epsilon =\frac{6}{l}=p.$$
(33)
Eq. (33) recovers the relation betweeen the brane tension and the bulk cosmological term given in (up to notations).
Finally, the state equation(32) can be easily generalized to the $`(n+2)`$-dimensional anti de Sitter bulk spacetime as follows:
$$p=\frac{\epsilon (n1)}{n}\frac{4n}{\epsilon l^2}.$$
(34)
## 4 Concluding remarks
We have shown that the study of branes in black hole bulks requires the use of multi-warped manifolds and there is the necessity of matter on the branes to set up acceptable Einstein equations.
The matter we find is rather interesting. It is a “Chaplygin gas” in BHTZ situation and a more complicate type of gas in the Schwarschild-AdS case, which however retains the “Chaplygin” form near the horizon. In both cases the matter we find is of “cosmological” type: a (positive) cosmological can be thought as a fluid having (constant) positive energy density and negative pressure. This type of matter may also be of interest in the context of “quintessential” expansion models and, from an observative viewpoint, it may have a connection with the by now accepted existence of a positive acceleration of the universe expansion .
It would be interesting to explore if there exist an interpretation for the equation of state we have found in the Schwarschild-AdS case from the viewpoint of a theory of extended objects.
## Acknowledgements
We are grateful to V. Gorini for useful discussions. A.K. is grateful to the CARIPLO Science Foundation for the financial support. His work was also partially supported by RFBR via grant No 99-02-18409.
|
warning/0005/hep-th0005244.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
D-branes are sources of RR fields . In IIA theory a D$`2p`$-brane is coupled to the RR gauge fields $`C^{(1)},C^{(3)},\mathrm{},C^{(2p+1)}`$ (they are 1-, 3-, $`\mathrm{}`$ and $`(2p+1)`$-forms respectively). Their couplings to the brane depend on the gauge invariant field $`B_D=B+F`$, where $`F=dA`$ is the tension of $`U(1)`$ gauge field living on the brane, and $`B`$ is the gauge 2-form, pertinent for string theory. The problem is that $`B`$ is not a tension and therefore the integrals $`B_D^k`$ over cycles in brane’s world sheet are not necessarily integer-valued (only $`F^k`$ are) and can not serve as charges .
In order to resolve this problem W.Taylor and J.Polchinski suggested to substitute the naive formula
$$\begin{array}{c}Q^{(1)}=_{M_2}B_D\end{array}$$
(1)
for the $`C^{(1)}`$-charge of a topologically trivial closed D2-brane by:
$$\begin{array}{c}Q^{(1)}=_{M_2=V_3}B_D_{V_3}H=_{M_2}F\end{array}$$
(2)
where $`M_2=V_3`$ is the position of D2-brane and $`V_3`$ is any 3-volume with the boundary at $`M_2`$. If the tension $`H=dB`$ is an exact 3-form, $`dH=0`$ (i.e. when $`B`$ is well defined), this expression does not depend on the choice of $`V_3`$. The argument of in favor of (2) is essentially that the bulk action mixes RR fields $`C^{(2k+1)}`$ with different $`k`$ and one needs to diagonalize the action before defining the physical charges (as measured by remote probes). The bulk action is diagonal in terms of the tensions $`G^{(2)}=dC^{(1)}`$ and $`G^{(4)}=dC^{(3)}C^{(1)}H`$ and the transformation from $`C`$’s to $`G`$’s of different degree is given by a triangular matrix. So, the diagonalization does not change the coupling to the highest RR field. In order to define effective couplings to lower RR fields, we should first integrate out all higher RR fields $`C^{(2m+1)}`$ with $`m>k`$ and then read the coupling of $`C^{(2k+1)}`$ to the brane. Such diagonalization procedure gives rise to non-local contributions, so that the resulting source terms are no longer concentrated on the brane world sheet. However, the non-locality disappears for constant RR fields which are used to probe the charges.
The purpose of the paper is to rephrase the reasoning of and to generalize it to arbitrary RR fields. In the standard $`d=10`$ IIA string theory the RR fields $`C^{(2p+1)}`$ and $`C^{(72p)}`$ are dual to each other. In this paper we ignore this restriction and consider instead a different model which preserves the RR gauge symmetries but all RR fields are independent of each other and the space-time dimension is not specified. In this case, we show that the effective coupling of the RR fields to D-branes (after integrating out higher RR fields) depends only on the tension $`F`$ of the U(1) gauge field on the brane. Since our interest is in the $`B`$-dependence of the RR charges, we ignore the curvature-dependent corrections and do not use the related $`K`$-theory formalism .
The paper is organized as follows. We begin in sect.2 with the case of the D2-brane with $`C^{(3)}`$ and $`C^{(1)}`$ fields, considered in . Then, after a brief discussion of the $`d^1`$-operation in sect.3, we proceed in sect.4 to generic consideration of $`C^{(2p+1)}`$ fields (gauge odd forms) in the presence of $`B^{(2k)}`$ fields (gauge even forms). For cohomologically non-trivial fields $`H`$ and/or topologically non-trivial branes, the gauge invariance produces constraints on the brane dynamics. They are discussed in sect.5. Sect.6 addresses the ambiguities in RR charges arising in these circumstances and their potential implications for quantum theory of branes.
## 2 The case of D2-brane
The action of (the massless sector of) the IIA theory can be obtained by dimensional reduction from the $`d=11`$ supergravity. In $`11d`$ the bosonic sector consists of the metric $`G_{MN}`$ and the 3-form $`𝒜_{MNP}`$, subject to gauge transformations $`𝒜𝒜+d\sigma `$ with any 2-form $`\sigma `$. The Lagrangian in $`11d`$ is:
$$\begin{array}{c}L_{11}=\sqrt{G}R(G)+|d𝒜|^2+𝒜d𝒜d𝒜\end{array}$$
(3)
Here $`|d𝒜|^2=\sqrt{G}G^{M\stackrel{~}{M}}G^{N\stackrel{~}{N}}G^{P\stackrel{~}{P}}G^{Q\stackrel{~}{Q}}_M𝒜_{NPQ}_{\stackrel{~}{M}}𝒜_{\stackrel{~}{N}\stackrel{~}{P}\stackrel{~}{Q}}`$, and the Chern-Simons term is independent of the metric.
After dimensional reduction to $`10d`$, the $`11d`$ bosonic fields turn into:
$$\begin{array}{c}g_{\mu \nu }=G_{\mu \nu }+C_\mu ^{(1)}C_\nu ^{(1)},C_\mu ^{(1)}=G_{\mu ,11}\end{array}$$
(4)
and
$$\begin{array}{c}C_{\mu \nu \lambda }^{(3)}=𝒜_{\mu \nu \lambda },B_{\mu \nu }=𝒜_{\mu \nu ,11}\end{array}$$
(5)
The $`G_{11,11}`$ component of the metric turns into the exponential of the dilaton field, which is irrelevant for our purposes and is omitted from all the formulas in the present paper. The gauge invariances (in addition to general coordinate transformations in $`10d`$) are inherited from the general coordinate boosts in the 11-th dimension ($`ϵ^{(0)}`$) and from the gauge invariance of $`A`$ ($`\mathrm{\Lambda }`$ and $`ϵ^{(2)}`$),
$$\begin{array}{c}BB+d\mathrm{\Lambda }(\mathrm{\Lambda }_\mu =\mathrm{\Lambda }_{\mu ,11}),\\ C^{(1)}C^{(1)}+dϵ^{(0)},\\ C^{(3)}C^{(3)}+dϵ^{(2)}dϵ^{(0)}B\end{array}$$
The Lagrangian in $`10d`$ is of the form,
$$\begin{array}{c}L_{10}=\sqrt{g}R(g)+|dC^{(1)}|^2+\left|dC^{(3)}C^{(1)}H\right|^2+BdC^{(3)}dC^{(3)}=\\ =\sqrt{g}R(g)+|G^{(2)}|^2+|G^{(4)}|^2+BG^{(4)}G^{(4)}\\ BBG^{(4)}G^{(2)}\frac{1}{3}BBBG^{(2)}G^{(2)}+\text{ total derivative}\end{array}$$
Here $`H=dB`$, $`G^{(2p+2)}=dC^{(2p+1)}C^{(2p1)}H`$, and the total derivative appears from the transformation of the Wess-Zumino term. The bulk Lagrangian $`L_{10}`$ contains terms mixing the fields $`C^{(1)}`$ and $`C^{(3)}`$. It can be diagonalized by the transformation $`CG`$. However, the inverse transformation $`GC`$ is non-local.
Introduce now even-dimensional D$`2p`$-branes located at $`M_{2p}`$ with the world sheets $`W_{2p+1}`$. They contribute the source (Wess-Zumino) terms to the action ,
$$\begin{array}{c}_{W_1}C^{(1)}+_{W_3}(C^{(3)}+C^{(1)}B_D)+\mathrm{}\end{array}$$
(6)
Here $`B_D=B+F`$ is a gauge invariant field, which can be considered as a linear combination of the gauge field $`B`$, living in the bulk, and the tension of the $`U(1)`$ gauge field $`F=dA`$, on the brane. Under the $`\mathrm{\Lambda }`$-transformation, the fields $`A`$ and $`B`$ transform as $`AA\mathrm{\Lambda },BB+d\mathrm{\Lambda }`$. The tension $`F`$ is a closed form (satisfies the Bianchi identity), $`dF=0`$, so that on the brane $`dB_D=dB=H`$.
The source (surface) terms (6) are gauge invariant up to total derivatives (i.e. for closed world-sheets $`W_{2p+1}=0`$). However, they are not expressed in terms of the $`G^{(2p+2)}`$-fields. If one still wants to get such an expression, a non-local formula occurs, no longer concentrated on the world-sheets. For the time being we loosely use the operation $`d^1`$, its more adequate substitute will be discussed in the following section 3.
So, $`C^{(1)}=d^1G^{(2)}`$ and
$$\begin{array}{c}C^{(3)}+C^{(1)}B_D=d^1G^{(4)}+C^{(1)}B_D+d^1(C^{(1)}H)\end{array}$$
(7)
In order to obtain the $`C^{(1)}`$ (i.e. $`G^{(2)}`$) charge, one needs to consider the coupling to D-brane of (the time-component of) the constant $`C^{(1)}`$ field (emitted or felt by a remote $`C^{(1)}`$-probe like a D0-brane). For constant $`C^{(1)}`$, however, $`d^1(C^{(1)}H)=C^{(1)}B`$ and we obtain for $`Q^{(1)}`$ exactly the formula (2). This is the argument of . It deserves mentioning that the Chern-Simons term in the bulk Lagrangian vanishes for constant $`C^{(1)}`$.
The same argument is easily applied to the derivation of $`Q^{(2p1)}`$, the charge distribution of $`C^{(2p1)}`$ (provided the bulk Lagrangian for $`C^{(2p+1)}`$ fields is given by $`|G^{(2p+2)}|^2`$, as implied by duality transformations $`C^{(2p+1)}C^{(72p)}`$). However, in order to iterate our procedure and reach lower $`Q^{(2k+1)}`$ with $`k<p1`$ one can not keep $`C^{(2p1)}`$ constant. This motivates a more detailed discussion of the operation $`d^1`$ in the next section.
## 3 Comment on the $`d^1`$ operation
Because of the nilpotency property $`d^2=0`$ of exterior derivative the $`d^1`$ operator is not well-defined. Its proper substitute is the $`K`$-operation (de Rham homotopy), satisfying
$$\begin{array}{c}Kd+dK=1\end{array}$$
(8)
As required for $`d^1`$, $`K`$ maps $`k`$-forms into $`(k1)`$-forms. It is defined modulo $`d`$, for exact forms $`K(dV)=Vd(KV)=V\text{mod}(d)`$. In application to integrals over topologically trivial surfaces, $`K`$ can be given, for example, by the following explicit construction (known for physicists from the fixed-point gauge formalism of the early days of gauge theories, see, e.g., ). For a 1-form $`A(x)`$, the 0-form
$$\begin{array}{c}KA(x)=_0^1A_\mu (tx)x^\mu 𝑑t\end{array}$$
(9)
and in general for a $`k`$-form $`A(x)`$
$$\begin{array}{c}\left(KA(x)\right)_{\mu _1\mu _2\mathrm{}\mu _{k1}}=_0^1A_{\mu _1\mathrm{}\mu _k}(tx)x^{\mu _k}t^{k1}𝑑t\end{array}$$
(10)
or
$$\begin{array}{c}_{S_{k1}}KA=_{C(S)_k}A\end{array}$$
(11)
where $`C(S)`$ is a cone with the base $`S`$ and the vertex somewhere outside $`S`$. Of course, the $`K`$ operation depends on the choice of this vertex, but this dependence results in a gauge transformation for gauge forms. The $`n`$-fold application of $`K`$ builds up an $`n`$-dimensional simplex $`C^n(S)`$ over the surface $`S`$. For $`k_i`$-forms $`A_i`$, $`k_i=k`$ one obtains,
$$\begin{array}{c}_{S_{kn}}K\left(A_{k_1}K\left(A_{k_2}K\left(\mathrm{}KA_{k_n}\right)\right)\right)=_{C^n(S)_k}A_{k_1}A_{k_2}\mathrm{}A_{k_n}\end{array}$$
(12)
Coming back to the surface term (7), one can note that $`d^1`$ in this expression can be safely substituted by $`K`$, because $`KdC^{(3)}C_3`$ modulo gauge transformation. At the same time, in variance with (7),
$$\begin{array}{c}_{W_3}\left(C^{(3)}+C^{(1)}B_D\right)=_{W_3}\left(KG^{(4)}+C^{(1)}B_D+K(C^{(1)}H)\right)=\\ =_{W_3}C^{(1)}B_D+_{C(W_3)}\left(G^{(4)}+C^{(1)}H\right)\end{array}$$
is a well defined expression for non-constant fields $`C^{(1)}`$.
## 4 Generic RR fields
At this point we deviate from conventional IIA theory, and assume that the RR field is an odd gauge poli-form in the bulk,
$$\begin{array}{c}C=_{k=0}C^{(2k+1)}\end{array}$$
(13)
with no interrelations on $`C^{(2k+1)}`$ imposed. The background is given by an even gauge poli-form $`B`$ in the bulk,
$$\begin{array}{c}B=_{k=1}B^{(2k)}\end{array}$$
(14)
and by an odd gauge poli-form $`A`$ on the D-brane world-sheets,
$$\begin{array}{c}A=_{k=1}A^{(2k1)},F=dA=_{k=1}dA^{(2k1)},B_DB+F\end{array}$$
(15)
The gauge symmetries are as follows,
$$\begin{array}{c}CC+e^Bdϵ,BB+d\mathrm{\Lambda },AA\mathrm{\Lambda }+d\alpha ,\end{array}$$
(16)
where
$$\begin{array}{c}ϵ=_{k=0}ϵ^{(2k)},\mathrm{\Lambda }=_{k=1}\mathrm{\Lambda }^{(2k1)}\end{array}$$
(17)
are even and odd poli-forms in the bulk respectively, and
$$\begin{array}{c}\alpha =_{k=0}\alpha ^{(2k)}\end{array}$$
(18)
is an even poli-form on the brane world sheet. The tension $`H=dB`$ of the $`B`$-field and the tension $`G=dCCH`$ of the RR field are gauge invariant bulk fields. As before, the combination $`B_D=B+F`$ is a gauge invariant field on the brane world-sheet.
In analogy with the IIA theory we assume the following Lagrangian of RR fields (in neglect of possible Chern-Simons terms) in the bulk,
$$\begin{array}{c}|G|^2=\left|dCCdB\right|^2=_i\left|dC^{(2i+1)}C^{(2i1)}dB\right|^2.\end{array}$$
(19)
Here and in what follows we often omit the sign of wedge product, meaning that contractions with the help of the metric involve either the squares $`||^2`$ or the Hodge operation $``$. In particular $`e^B=e^B`$, $`CH=CH`$ etc.
The Wess-Zumino contributions of D-branes are <sup>*</sup><sup>*</sup>* We neglect here the curvature terms, which are given by a pullback of the factor $`\sqrt{\widehat{A}(R)}`$ in the integrand , where the $`\widehat{A}`$-genus is the one, appearing in index theorems, like $`ind(D)=\widehat{A}(R)\text{tr}e^F`$.:
$$\begin{array}{c}_p_{W_{2p+1}}Ce^{B_D}=_p\left(_{i=0}\frac{1}{i!}_{W_{2p+1}}C^{2p+12i}B_D^i\right)=_{k,i}\frac{1}{i!}_{W_{2k+2i+1}}C^{(2k+1)}B_D^i.\end{array}$$
(20)
According to our general strategy we should rewrite the source term (20) in terms of $`G`$, which diagonalizes the bulk action. The resulting expression will essentially contain the $`K`$ operation, but formulas for the RR charges are obtained at constant RR fields when the expression becomes local. Formally, in order to express $`C`$ through $`G`$, one needs to solve the equation
$$\begin{array}{c}G=dCCH=(d+H)C=\left(e^Bde^B\right)\end{array}$$
(21)
For our purposes, we need a formula with $`K`$-operation acting directly on $`C`$. Since up to a gauge transformation $`CK(dC)`$, we write $`CKGKHC`$, and
$$\begin{array}{c}C\frac{1}{1+KH}KG=KGK(HKG)+K(HK(HKG))+\mathrm{}\end{array}$$
(22)
Since $`K`$ lowers the rank of the form by one and multiplication by $`H`$ raises it by three or more (note that there is no terms with $`k=0`$ in (14)), this formula contains only finitely many terms for $`C`$ of a given finite rank. Finally, the source term turns into:
$$\begin{array}{c}_p_{W_{2p+1}}Ce^{B_D}=_p_{W_{2p+1}}e^{B_D}\frac{1}{1+KH}KG\end{array}$$
(23)
Now the entire Lagrangian is diagonalized in terms of the fields $`G`$ and we can proceed to the definition of effective couplings.
### Examples.
For a D2-brane we have
$$\begin{array}{c}_{W_3}\left(C^{(3)}+B_D^{(2)}C^{(1)}\right)=_{W_3}(1+B_D+\mathrm{})(1KH+\mathrm{})(KG^{(2)}+KG^{(4)}+\mathrm{})=\\ =_{W_3}\left(KG^{(4)}+B_DKG^{(2)}K(H(KG^{(2)}))\right)=_{W_3}KG^{(4)}+_{W_3}\left(B_DC^{(1)}+K(C^{(1)}H)\right)\end{array}$$
(we used the fact that $`KG^{(2)}C^{(1)}`$). This reproduces eq.(7).
Similarly, for a D4-brane:
$$\begin{array}{c}_{W_5}(1+B_D^{(2)}+B_D^{(4)}+\frac{1}{2}(B_D^{(2)})^2+\mathrm{})\\ (1KH^{(3)}KH^{(5)}+KH^{(3)}KH^{(3)}+\mathrm{})(KG^{(2)}+KG^{(4)}+KG^{(6)}+\mathrm{})=\\ =_{W_5}KG^{(6)}+_{W_5}\left(B_D^{(2)}KG^{(4)}K(H^{(3)}(KG^{(4)}))\right)+\\ +_{W_5}\left(\left(B_D^{(4)}+\frac{1}{2}(B_D^{(2)})^2\right)KG^{(2)}K(H^{(5)}(KG^{(2)}))+K\left(H^{(3)}K(H^{(3)}KG^{(2)})\right)\right)\end{array}$$
In order to read off the coupling of $`C^{(5)}`$ from (4) one notes that this field enters only through $`G^{(6)}`$ and the $`K`$-operation can be explicitly applied to this term: $`_{W_5}K(dC^{(5)})=_{W_5}C^{(5)}`$.
The procedure to define the $`C^{(3)}`$-“charge” literally repeats the one for the $`C^{(1)}`$-charge of D2-brane. First, we assume (following ) that $`G^{(6)}`$ is already integrated out, thus one does not look at $`C^{(3)}`$ in $`G^{(6)}=dC^{(5)}C^{(3)}H^{(3)}C^{(1)}H^{(5)}`$. Then $`C^{(3)}`$ enters (4) only through $`G^{(4)}`$. Second, one should put $`C^{(3)}=const`$, then the $`K`$ operation can be applied explicitly and give:
$$\begin{array}{c}_{W_5}\left(B_D^{(2)}C^{(3)}K(H^{(3)}C^{(3)})\right)\stackrel{C^{(3)}=const}{=}_{W_5}(B_DB)C^{(3)}=_{W_5}C^{(3)}F\end{array}$$
(24)
Applying the same procedure to the case of the $`C^{(1)}`$-charge, i.e. omitting the terms with $`G^{(6)}`$ and $`G^{(4)}`$, putting $`KG^{(2)}=C^{(1)}=const`$ and applying $`K`$ explicitly) we obtain from (4):
$$\begin{array}{c}KG^{(2)}=_{W_5}\left(\left(B_D^{(4)}+\frac{1}{2}(B_D^{(2)})^2\right)C^{(1)}K(H^{(5)}C^{(1)})+K\left(H^{(3)}K(H^{(3)}C^{(1)})\right)\right)=\\ \stackrel{C^{(1)}=const}{=}_{W_5}\left(\frac{1}{2}(B_D^{(2)}B^{(2)})^2+(B_D^{(4)}B^{(4)})\right)C^{(1)}=_{W_5}\left(\frac{1}{2}(F^{(2)})^2+F^{(4)}\right)C^{(1)}\end{array}$$
so that for a D4-brane
$$\begin{array}{c}Q^{(1)}=_{M_4}\left(\frac{1}{2}(F^{(2)})^2+F^{(4)}\right)\end{array}$$
(25)
### $`B`$-independence of RR charges.
These examples illustrate the general prescription to define the effective D-brane coupling to the constant RR field $`C^{(2k+1)}`$: in (23) neglect all the contributions of $`G^{(2m+2)}`$ with $`mk`$ and in the term with $`KG^{(2k+2)}`$ put $`C^{(2k+1)}=const`$. Then, it appears that the non-localities remaining in the $`K`$-operation can be explicitly eliminated, the contributions of $`B`$ to $`B_D`$ are completely canceled, and the final answer depends only on $`F`$. It deserves noting that while $`F`$ itself is not gauge invariant, the integrals $`F^k`$ over compact $`2k`$-cycles are.
The way to prove these claims is actually clear from above examples. It is enough to note that whenever $`KG`$ in (23) is substituted by a constant $`C`$ to obtain,
$$\begin{array}{c}\frac{1}{1+KH}C\stackrel{C=const}{=}_{j=0}()^j(KH)^jC=_{j=0}\frac{()^j}{j!}B^jC=e^BC.\end{array}$$
(26)
Then, since $`e^{B_DB}=e^F`$, the source term (23) becomes $`B`$-independent,
$$\begin{array}{c}_p_{W_{2p+1}}e^FC.\end{array}$$
(27)
This formula for the effective coupling of a D-brane to the constant RR gauge fields is the main result of the paper. It shows that only the integral 2-form $`F`$ contributes to the properly defined RR charges, in analogy to the claim of Taylor-Polchinski for D$`2`$-branes.
## 5 Implications of gauge non-invariance of the source terms
Though both the bulk and the world-sheet actions are expressed through the gauge-invariant $`G`$-fields, the non-locality of the operation $`K`$ can – and does – (slightly) diminish the gauge invariance of the source (world-sheet) terms. This leads to additional constraints imposed on the brane configurations contributing to the functional integral over RR fields.
For example, the bulk action is invariant under
$$\begin{array}{c}CC\epsilon de^B\end{array}$$
(28)
with $`\epsilon =const`$. (Indeed, $`Gd\epsilon de^B\epsilon de^BdB`$, and the first term vanishes for constant $`\epsilon `$, while the second one vanishes because $`dBdB=dBdB=0`$.) However, the boundary (source) term is not invariant:
$$\begin{array}{c}\delta Ce^{B_D}=\epsilon 𝑑Be^{B_DB}=\epsilon 𝑑Be^F\end{array}$$
(29)
Integration over the zero-mode $`\epsilon `$ of the full action provides the constraints,
$$\begin{array}{c}k0_{V_{3+2k}}F^k𝑑B=0\end{array}$$
(30)
for any cycle in the D$`2p`$-brane’s world-volume. In particular, for a D2-brane one gets $`_{W_3}H=0`$, i.e. the D2-brane can not wrap around a source of $`H`$-field (like an NS brane) in its time-evolution (such trajectories do not contribute to the functional integral).
Generalizing (28) to
$$CC+\epsilon e^BB^mdB$$
one obtains extra constraints:
$$F^kB^m𝑑B=0$$
or their linear combinations $`F^kB_D𝑑B_D`$. At the same time, the $`B`$-independent gauge transformation
$$CC+\epsilon $$
with constant $`\epsilon `$ can be easily excluded from the gauge group from the very beginning, and therefore the integrals $`F^k`$ without $`B`$-fields need not vanish, i.e. dynamics of RR fields eliminates wrappings around “magnetic” sources of $`B`$ fields, but not of $`F`$ fields.
## 6 RR charge in cohomologically non-trivial $`H`$
When $`H`$ is cohomologically nontrivial, expressions like (2), (7) and (4), in variance with the naive (1) are ill-defined. Still, it seems that they are adequate for description of the actual situation, and the ambiguities may appear to have physical significance. An ambiguity problem arises if there are non-contractable 3-cycles in the space-time with non-vanishing $`H`$ (e.g., in Calabi-Yau compactifications) or sources of “magnetic” $`H`$-fields, where $`dH0`$ (e.g., the NS5-branes). Similar problems occur for topologically-nontrivial D-branes, i.e. when $`M_2V_3`$: then (7) and (4), in variance with (2) are still applicable, but ambiguous.
In order to understand what happens in such situations, one can analyze a 1-dimensional toy-example. For a puzzle, involving realistic D2-branes, see ref.. Consider the 1-dimensional Gaussian theory on a circle $`S_1`$ whose partition function is
$$\begin{array}{c}𝒟C(x)\mathrm{exp}\left(\frac{i}{2\pi }_{S_1}\left|\frac{C}{x}\sigma \frac{b}{x}\right|^2+C(x_1)C(x_2)\right)\end{array}$$
(31)
Here $`C(x)`$ models the $`C^{(3)}`$ field in the bulk, $`b(x)`$ and $`h=b/x`$ – the $`B`$ and $`H`$ fields respectively, $`\sigma `$ plays the role of the constant $`C^{(1)}`$ in the bulk, and the source term $`C^{(3)}`$ (dipole charge) on the brane world-sheet is imitated by $`C(x_1)C(x_2)`$. Cohomologically non-trivial $`h`$, such that $`_{S_1}h(x)𝑑x0`$, arises when $`b(x)`$ is non-periodic, $`\mathrm{\Delta }b=b(x+1)b(x)0`$, with periodic $`h(x)`$.
The answer for the Gaussian integral (31) (with eliminated gauge zero-mode $`C(x)=const`$) is
$$\begin{array}{c}\mathrm{exp}\left[2\pi i\sigma \left(b(x_1)b(x_2)\right)\right]\end{array}$$
(32)
and for non-periodic $`b(x)`$ it is an ambiguous function of $`x_1`$ (with $`x_2`$ fixed). If there were no $`\sigma `$ (no $`C^{(1)}`$ field) coupled to $`h(x)`$ in the bulk, one could have handled this problem by imposing a Dirac quantization condition on $`b`$: $`\mathrm{\Delta }b=integer`$. However, in our model $`\sigma `$ is a dynamical variable and, even if constant, can vary continuously. In other words, coupling to RR fields makes the phase ambiguity physically relevant. As usual in such situations, the physical state is not defined solely by the current position of the brane (by the points $`x_1`$ and $`x_2`$ on $`S_1`$), it also depends on its pre-history: given original state at some moment, one can obtain at another moment the physically distinct states with the same $`x_1`$ and $`x_2`$, differing by the number of times the $`x_1`$ and $`x_2`$ wrapped around the circle. The situation may be reminiscent of the theory of anyons. The state of such a system is not fully determined by current positions of the anyons, it is also labeled by an attached (invisible) braid. Moreover, the ambiguity is not automatically resolved by simply summing over all the pre-histories: dynamical constraints, discussed in the previous section lead to a superselection rule, preventing dynamical interference of topologically different pre-histories.
## 7 Acknowledgements
We are indebted to the ESI, Vienna, for the hospitality and support which made this cooperation possible. Our work is partly supported by grants: NFR F 674/1999, INTAS 96-0196, INTAS 99-01705 (A.A.), RFBR 98-01-00328, INTAS 97-0103 (A.Mir), RFBR 98-02-16575 (A.Mor.), CRDF #6531 (A.M.’s) and the Russian President’s grant 00-15-99296 (A.Mor.).
|
warning/0005/astro-ph0005491.html
|
ar5iv
|
text
|
# Multiwavelength Examination of the COS–B Field 2CG 075+00 Yields a Blazar Identification for 3EG J2016+3657
## 1 Introduction
Since the first surveys of the $`\gamma `$-ray sky with the COS–B satellite the nature of most $`\gamma `$-ray sources remain a mystery, as few of these sources have firmly established counterparts at any other waveband (Swanenburg et al. 1981). With the launch of the Compton Gamma Ray Observatory (CGRO) in 1991, improved surveys at relatively higher angular resolution in the $`\gamma `$-ray band was made possible. In particular, a systematic survey of the $`\gamma `$-ray sky, including the COS–B source regions was carried out with the on-board EGRET (Energetic Gamma-ray Experiment Telescope) instrument at energies above 100 MeV. EGRET has so far detected 271 sources of high energy $`\gamma `$-rays in the third EGRET (3EG) catalog (Hartman et al. 1999), of which 169 remain unidentified (74 of these are at $`|b|<10^{}`$), with no convincing counterparts at other wavelengths. Surprisingly, only two of the unidentified COS–B sources have been subsequently associated with EGRET sources, and both are pulsars, namely Geminga (Bertsch et al. 1992), and 2CG 342-02 (PSR B1706-44) (Thompson et al. 1992). A third source, PSR B1046-58, could possibly be the candidate for identification of 2CG 288-00 (Kaspi et al. 2000).
To date, the only sources of high energy $`\gamma `$-rays convincingly identified are the blazars at high galactic latitudes and the pulsars at low latitudes. Studies of individual unidentified EGRET $`\gamma `$-ray source fields have been inconclusive in locating the origin of their emission. Comprehensive surveys of the field associated with these sources have met with limited success (see Mukherjee, Thompson & Grenier 1997 for a review). Several researchers have noted that the unidentified EGRET sources in the Galactic plane lie in proximity to star formation sites and supernova remnants (Yadigaroglu & Romani 1997, Sturner & Dermer 1995, Esposito et al. 1996), while others report a correlation with OB associations and massive stars (Kaaret & Cottam 1996; Kaul & Mitra 1997; Romero et al. 1999). Efforts to identify the $`\gamma `$-ray sources at other wavelengths include systematic multifrequency radio observations (e.g. Özel et al. 1988) and X-ray imaging studies (e.g., Brazier et al. 1996; 1998; Roberts & Romani 1998; Mirabal et al. 2000).
In this article we re-visit the region containing the unidentified COS–B source 2CG 075+00, located in the Cygnus region, for which a significant amount of archival $`\gamma `$-ray (EGRET) and X-ray (ASCA & ROSAT) data have accumulated. Previous attempts to locate the origin of the high energy emission were fraught with frustration, as the position associated with 2CG 075+00 in the second EGRET catalog (Thompson et al. 1995), 2EG J2019+3719, has shifted in the 3EG catalog, and has split into two discrete sources, 3EG J2016+3657 and 3EG J2021+3716. Fortunately, both revised EGRET error boxes have overlapping archival ASCA and ROSAT observations. We now re-analyze the EGRET data along with the corresponding X-ray fields using the refined $`\gamma `$-ray positions from the later catalog.
## 2 The $`\gamma `$-ray Observations
2CG 075+00, first observed by COS–B, is located in the Galactic plane. The revised COS–B position for the source is $`l=76.1^{}`$, $`b=0.5^{},`$ with an error radius of $`1.0^{}`$ (Pollock et al. 1985). In the Second COS–B Catalog, Swanenburg et al. (1981) note that the source structure could possibly be interpreted as containing extended features. The integrated $`\gamma `$-ray flux from the source at energies greater than 100 MeV was found to be $`1.3\times 10^6`$ photons cm<sup>-2</sup> s<sup>-1</sup>, although no spectral information was available (Swanenburg, et al. 1981).
Since its launch in 1991, EGRET has observed the error circle of 2CG 075+00 several times. EGRET is sensitive to $`\gamma `$-rays in the energy range from 30 MeV to 30 GeV, with a point source sensitivity of $`3\times 10^7`$ ph cm<sup>-2</sup> s<sup>-1</sup> ($`>100`$ MeV). It has an effective area of $`1.5\times 10^3`$ cm<sup>2</sup> in the energy range 0.2 – 1 GeV, decreasing to about one-half the on-axis value at $`18^{}`$ off-axis and to one-sixth at $`30^{}`$. The nominal angular resolution is $`0.6^{}`$ at 500 MeV, improving to $`0.4^{}`$ at 3 GeV. Details of the instrument preflight and in-flight calibration are described elsewhere (Thompson et al. 1993; Esposito et al. 1999).
In the second EGRET (2EG) catalog, 2CG 075+00 was weakly detected as 2EG J2019+3719. Reanalysis of the region using a larger data set for the 3EG revealed two sources, 3EG J2016+3657 and 3EG J2021+3716, with error radii of $`33^{}`$ and $`18^{}`$, respectively, at the 95 % contour (Hartman et al. 1999). In the summed Phase 1 through Cycle 4 EGRET observations, 3EG J2016+3657 and 3EG J2021+3716 were detected at the highest significances of $`6.4\sigma `$ and $`10.3\sigma `$, respectively, and these observations were used for the position determination. Figure 1 shows the EGRET source positions superimposed on the ROSAT X-ray image that is described in §3. 3EG J2021+3716 is completely within the error circle of the COS–B source 2CG 075+00, also shown in Figure 1.
An analysis of 4.5 years of EGRET data based on photons with energies greater than 1 GeV gives a slightly different result for source positions in the Cygnus region (GeV catalog: Lamb & Macomb 1997). The GeV catalog lists one source with an error circle overlapping that of 2CG 075+00, namely, GEV J2020+3658, at $`l=75.29^{}`$, $`b=0.24^{}`$ (see Fig. 1).
We examined the flux history of 3EG J2016+3657 and 3EG J2021+3716 to search for variability, such as is characteristic of high latitude EGRET blazars. The light curve for these sources are displayed in Fig. 2 using data from the 3EG catalog (Hartman et al. 1999) for the Phase 1 through Cycle 4 observations (1991-1995). Two additional observations were made in Cycle 6 in viewing period (VP) 601.1 (1996 October) and VP 623.5 (1997 July). These data were analyzed using the standard EGRET data processing technique, as described in Mattox et al. (1996) and Hartman et al. (1999), and are included in Fig. 2. The horizontal bars on the individual data points denote the extent of the VP for that observation. Fluxes have been plotted for all detections greater than $`2\sigma `$. For detections below $`2\sigma `$, upper limits at the 95% confidence level are shown. A $`\chi ^2`$ analysis can be used to calculate a variability index according to that defined by McLaughlin et al. (1996). Although somewhat arbitrary, the quantity $`V`$ can be used to judge the strength of the evidence of flux variability. Following the classification used in McLaughlin et al. (1996), we use $`V<0.5`$ to indicate non-variability, and $`V1`$ to indicate variability. In this case, we obtain $`V=1.1`$ for 3EG J2016+3657 and $`V=1.57`$ for 3EG J2021+3716. We note, however, that 3EG J2016+3657 and 3EG J2021+3716 are in a confused region (as indicated in the 3EG Catalog), and the variability numbers could be an artifact of the analysis, rather than being intrinsic to the sources. In comparison to the EGRET blazars, 3EG J2016+3657 and 3EG J2021+3716 have variability indices similar to a large fraction of the blazars detected by EGRET (Mukherjee et al. 1997).
The background-subtracted $`\gamma `$-ray spectra of 3EG J2016+3657 and 3EG J2021+3716 were determined by dividing the EGRET energy band of 30 MeV – 10 GeV into 4 bins for 3EG 2016+3657 and 10 bins for 3EG 2021+3716, and estimating the number of source photons in each interval, following the EGRET spectral analysis technique of Nolan et al. (1993). The data were fitted to a single power law of the form $`F(E)(E)^\mathrm{\Gamma }`$ photon cm<sup>-2</sup> s<sup>-1</sup> MeV<sup>-1</sup>, where $`F(E)`$ is the flux, $`E`$ is the energy, $`\mathrm{\Gamma }`$ is the photon spectral index. The photon spectral indices of 3EG J2016+3657 and 3EG J2021+3716 were found to be, respectively, $`1.99\pm 0.20`$ and $`1.86\pm 0.10`$.
## 3 The X-ray Observations
The error boxes of both 3EG sources are covered by archival X-ray imaging observations acquired with the ROSAT and ASCA observatories. Two adjacent observations with each observatory fall nicely on the two 3EG error boxes. Historically, these fields have been studied in X-rays both because of the existence of the 2CG source, as well as due to several known X-ray sources in the region. The ASCA observations herein were a part of a program to study unidentified sources in the Galactic plane (Tavani et al. 2000).
Archival data for the region were available for the ROSAT Position Sensitive Proportional Counter (PSPC), the ROSAT High Resolution Imager (HRI) and the ASCA Gas Imaging Spectrometer (GIS) which allow complementary broad-band X-ray data in the 0.2 – 2.0 keV (PSPC) and 1 – 10 keV (GIS) range with arcmin spatial resolution and moderate energy resolution. The PSPC $`1^{}`$ radius field-of-view is about twice that of the GIS. The ROSAT HRI observations took place on 1994 November 12 – 13, with a total exposure time of 43 ks. The ROSAT PSPC observations were on 1991 November 22 – 30, 1992 April 28 – 30, 1993 October 24 – 25 and 1994 June 5, with a total exposure time of $`12`$ ks. The ASCA observations took place on 1995 May 29 – 31, 1995 October 14 – 15, and 1996 February 2. All data were obtained from the HEASARC archive at Goddard Space Flight Center and edited using the latest standard processing for each mission.
We created ROSAT (Fig. 1) and ASCA (Fig. 3) images of the region containing 3EG J2016+3657 and 3EG J2021+3716 by co-adding exposure corrected sky maps from each mission. These images are centered on the position of the earlier Second Catalog source, 2EG 2019+3719. However, the PSPC image size is large enough to include the 95 % error contours of both the 3EG sources. Note that the ASCA images are not centered on the EGRET positions, and only part of the error contour of 3EG 2016+3657 is covered by the ASCA observation. In both Fig. 1 and Fig. 3 we show the error circles of 2CG 075+00 with dashed lines. Also shown is the error circle of the GeV source GeV J2020+2658 (Lamb & Macomb 1997).
To search for a possible X-ray counterpart to the $`\gamma `$-ray sources, we examined the ROSAT point sources enclosed by the 95 % contours of the two 3EG sources. The detected positions are numbered in the image (Fig. 1) and are tabulated in Table 1. There are 9 bright sources in the ROSAT image of 3EG J2016+3657, but none that are significant within the error circle of 3EG J2021+3716. Note that Source 3 (indicated with an arrow) is very faint, and barely resolved in the ROSAT PSPC image. Its position is determined from the ROSAT HRI image shown in §4. There are at least 5 X-ray sources in the error circle of 2CG 075+00, which reached a minimum detectable intrinsic flux of $`6.5\times 10^{13}`$ erg cm<sup>-2</sup> s<sup>-1</sup> in the 0.1 – 2.4 keV band, assuming a power-law photon spectral index of 2.0, and a Galactic column absorption of $`1\times 10^{22}`$ cm<sup>-2</sup>.
The ASCA images were, similarly, searched for corresponding X-ray counterparts. In deriving the ASCA positions, we were able to use the ROSAT point sources seen in the ASCA images to improve the astrometry for the ASCA sources to $`10^{\prime \prime }`$ by registering the ASCA images using the overlapping ROSAT sources. There are no significant point sources in the ASCA image within the 95 % contour of 3EG J2021+3716. The ASCA image of 3EG J2016+3657 reveals 5 point sources which are indicated with numbers in Fig. 3. Source numbers 1, 2, 3, 4 and 5 correspond to ROSAT sources of the same numbers in Fig. 1.
To measure the ASCA and ROSAT source count rates we extracted photons using a $`2^{}`$ radius aperture and estimated the background contribution using a large annulus away from the other source following the method described in Gotthelf & Kaspi (1998). For ASCA, we define hardness ratio as
$$\frac{S(<2\mathrm{keV})S(>2\mathrm{keV})}{S(<2\mathrm{keV})+S(>2\mathrm{keV})}.$$
The ASCA and ROSAT sources are listed in Table 1 along with their background subtracted count rates, detection significances, and hardness ratios. We found no other point sources in the ASCA image at the level of $`5\sigma `$ or higher, other than those listed in Table 1.
We have searched for counterparts of the X-ray sources in the ROSAT and ASCA images. Several of the sources have optical identifications and are listed in Table 1. Source 1 is a known supernova remnant, CTB 87. Two of the sources can be identified with radio sources. Notes on the individual sources in Table 1 are given in the following section.
## 4 Notes on Individual Sources
CTB 87: Source number 1 in Table 1 is coincident with SNR CTB 87 (G74.9+1.2), an extended source with a flat radio spectrum. G74.9+1.2 is a filled-center SNR in the radio with high polarization and a high frequency turnover. Its HI absorption indicates a distance of 12 kpc. It has a relatively flat spectrum in the radio with a spectral index of $`0.26\pm 0.2`$ below 11 GHz, beyond which a transition to a steeper spectral index occurs with the spectrum steepening to an index $`>1`$. (Morsi & Reich 1987; Salter et al. 1989). It is generally believed that filled-center SNRs are remnants in which a central object is responsible for the relativistic electrons, whose synchrotron emission is detected at radio frequencies (Koyama et al. 1997). The flat radio spectrum is believed to be due to the central pulsar which injects particles into the nebula (Reynolds & Chevalier 1984).
B2013+370: The hard X-ray source marked ‘3’ in the ASCA image (Fig. 3), that is barely resolved in the ROSAT PSPC image (Fig. 1), is consistent with the radio source B2013+370 (G74.87+1.22). B2013+370 is a well-studied compact, flat-spectrum radio source that was first noticed during the study of the SNR CTB 87 (Duin et al. 1975). It is located approximately $`7^{}`$ west of the brightness peak of CTB 87. Wilson (1980) first noted the association of the radio source with an extended or possibly double X-ray source observed in the 0.15 – 3.0 keV band with the Imaging Proportional Counter (IPC) on the Einstein Observatory. The IPC source, 1E 2013.7+3701, was situated roughly between the locations of our sources 2 and 3. The ROSAT HRI image (Fig. 4), however, shows that the Einstein source is clearly resolved into two point sources, corresponding to sources 2 and 3.
Due to its proximity to the SNR, the possibility that B2013+370 and CTB 87 are related cannot be excluded. However, such an association is unlikely based on the requirements of an unprecedented velocity for the radio source if it were to be associated with the nearby SNR (see for example the arguments presented in Shaffer et al. 1978).
In fact, B2013+370 has all the characteristics of a compact, extragalactic, non-thermal radio source and is typical of the many extragalctic sources seen by EGRET. Duin et al. (1975) report a radio spectral index of $`\alpha =0.2`$ at high frequencies (above $`7500`$ MHz), and a low frequency spectral index of about $`\alpha =+0.4\pm 0.06`$. The source exhibits a lack of recombination line emission, the presence of linear polarization and a spectrum consistent with a non-thermal source showing synchrotron self absorption below 8 GHz, as expected for a magnetic field of $`<0.1`$ Gauss (Duin et al. 1975). VLBI measurements indicate an angular extent of $`<0.001^{\prime \prime }`$ at 8 GHz and $`>0.009^{\prime \prime }`$ at 0.8 GHz (Weiler & Shaver 1978). From its radio properties, B2013+370 is very likely to be a flat-spectrum radio quasar or a BL Lac object. It has a 5 GHz flux of about 2 Jy (Duin et al. 1975), typical of many blazars seen by EGRET. In addition, B2013+370 is detected at 90 GHz and 142 GHz with the IRAM 30 m telescope and exhibits variability at these wavelengths, a specific property of EGRET blazars (Bloom et al. 1999, Mattox et al. 1997). This is demonstrated in Fig. 5 which shows the light curves of B2013+370 at 90 GHz and 142 GHz, measured between 1993 and 1995 (Reuter et al. 1997).
HD 228766: This is a Wolf-Rayet star, also known as SAO 69765 (Hog et al. 1998), and is the counterpart to source number 6 in Table 1. It has a $`B`$ magnitude of 9.72 and a $`V`$ magnitude of 9.22 Its spectral type is O5.5f.
HD 193077: Source number 7 in Table 1 is a bright ($`B=8.34`$, $`V=8.06`$) Wolf-Rayet star, HD 193077 (Perryman et al. 1997), also known as SAO 69755. The star corresponds to WR 138 in the Wolf-Rayet catalog (van der Hucht et al. 1981). Its a star of the WN sequence (subtype WN5+OB), with its spectra dominated by broad emission lines of helium and nitrogen (Lepine & Moffat 1999).
CCDM J20215+3758A: Together with CCDM J20215+3758B, this corresponds to a double star system and is the counterpart to Source 12 in Table 1. The USNO-A2.0 catalog gives the magnitudes of the two stars as follows: CCDM J20215+3758A: $`R=11.6`$, $`B=12.1`$, and CCDM J20215+3758B: $`R=11.9`$, $`B=13.5`$.
HD 229153: This is the counterpart to source 13 in Table 1 (Hog et al. 1998). It is a bright star ($`B=10.09`$, $`V=9.14`$) of spectral type BOIab.
We also find three of the other X-ray sources to be coincident with bright stars in the USNO-A2.0 catalog. Source 8 is possibly a star ($`B=12.4`$, $`R=11.3`$) with coordinates (J2000) 20 16 37.55, +37 05 55.0, and Source 9 is most likely a star of approximately 12th magnitude at (J2000) 20 17 35.86, +36 38 02.3. Source 14 probably corresponds to a star ($`B=11.7`$, $`R=10.8`$) with coordinates (J2000) 20 19 44.16, +37 35 26.7. In addition, we find two point sources in the ROSAT HRI image (Fig. 4), not listed in Table 1, to be coincident with bright stars. These are marked in the figure as: HD 228600 at (J2000) 20 15 30.8, +37 20 03.1 ($`B=10.5`$, $`V=10.1`$), and a star ($`B=15.7`$, $`R=14.2`$) at (J2000) 20 16 49.0, +36 57 48.
We do not find any likely counterparts in the literature to the other remaining sources in Table 1.
### 4.1 Optical Observations
We obtained CCD images in the $`R`$ band using the 2.4m telescope of the MDM Observatory on 2000 April 24, and in the $`I`$ band on March 18 using the MDM 1.3m. The regions covered were $`2^{}\times 2^{}`$ on the 2.4m, and $`8^{}\times 8^{}`$ on the 1.3m. In seeing of $`0.^{\prime \prime }75`$, an optical object of $`R=21.6\pm 0.2`$ is detected at (J2000) 20 15 28.76, +37 10 59.9 in the USNO–A2.0 reference system (Monet et al. 1996), with an uncertainty of $`0.^{\prime \prime }3`$. This coincides with the NVSS position (Table 2) of the blazar B2013+370, identified with X-ray source 3, as shown in Figure 6. This object was also detected optically by Geldzahler et al. (1984), who found $`I=19.5\pm 0.5`$, while also noting that it appeared extended. Our $`R`$-band image shows that the object closest to the radio position is unresolved, while a fainter object $`1.^{\prime \prime }7`$ to the northwest was almost certainly responsible for the extended appearance in the Geldzahler et al. (1984) image. Assuming standard conversions of $`N_\mathrm{H}`$ to extinction, the absorption in the $`R`$ band is $`5.1`$ mag to an extragalactic object, making its intrinsic magnitude $`R=16.5`$. Our $`I`$-band images have inferior seeing, but they also detect the blended pair at the radio position, as well as optical objects at the positions of X-ray sources 2 and 4. However, in the absence of further information such as radio detection or optical spectroscopy, the severe crowding and Galactic extinction makes the identification of these additional X-ray sources by positional coincidence alone dangerous, even from HRI positions.
### 4.2 Other Radio Sources
We have searched the NRAO/VLA Sky Survey (NVSS) catalog (Condon et al. 1998) of 1.4 GHz radio sources for other possible counterparts to the X-ray and $`\gamma `$-ray sources. A search within the error box of 3EG J2016+3657 revealed 126 radio sources. Of these, only 5 had integrated radio fluxes $`>0.3`$ Jy and are shown in Fig. 7 and Table 2. The two brightest of these are positionally coincident with the ROSAT and ASCA sources 1 (CTB 87) and 3 (B2013+370). The two sources clustered near CTB 87 are actually extended regions within the supernova remnant. None of the other NVSS sources match the positions of the X-ray sources in the ASCA and ROSAT images.
There are two bright $`(>0.3`$ Jy) radio sources within the error circle of 3EG J2021+3716 (see Fig. 7, Table 2). Both of these were also detected by IRAS, and were shown to be H II regions from their radio recombination lines (Lockman 1989).
A search of the Westerbork Northern Sky Survey (WENSS) catalog (Rengelink et al. 1997) of 92 cm (325 MHz) radio sources yielded similar results. This survey has a limiting flux density of about 18 mJy. A search within the error box of 3EG J2016+3657 yielded 21 radio sources, of which only 2 matched the positions of the X-ray sources. These are (a) WNB2014.1+3703 positionally coincident with ROSAT and ASCA source numbers 1, or CTB 87 and (b) WNB2013.6+3701 matching the ASCA source 3, or B2013+370. A search within the error circle of 3EG J2021+3716 yielded a pair of very bright radio source, WNB2019.7+3718B and WNB2019.7+3718C, which correspond to the HII regions also seen in the NVSS data (see Table 2).
## 5 Discussion
Our study of archival X-ray (ASCA and ROSAT) data yields several faint sources within the error boxes of the two 3EG sources. The region contains bright stars, the SNR CTB 87, a compact radio source, and HII regions.
The presence of the SNR CTB 87 (G74.9+1.2) in the field of 3EG J2016+3657 is potentially quite important in light of the $`\gamma `$-ray source/SNR associations noticed in previous investigations (Sturner & Dermer 1995; Esposito et al. 1996). Gamma-ray production from SNRs, Wolf-Rayet stars and OB associations is expected in several theoretical models. This subject was extensively investigated in the past for COS–B sources (Montmerle 1979; Völk & Forman 1982), and recently for EGRET (Romero et al. 1999; Esposito et al. 1996; Kaaret & Cottam 1996). If the $`\gamma `$-ray emission were from a young pulsar associated with the SNR, a conclusive way to prove this would be to find pulsations. This is unfortunately not possible for such a weak X-ray source.
However, the energetics of the CTB 87 remnant are far from adequate to produce a source detectable by EGRET at the inferred distance of 12 kpc. Wilson (1980) argued convincingly that the X-ray luminosity of CTB 87, which is 100 times less than that of the Crab, implies that the spin-down power of the embedded pulsar must be correspondingly less, $`I\mathrm{\Omega }\dot{\mathrm{\Omega }}1\times 10^{36}(d/12\mathrm{kpc})`$ ergs s<sup>-1</sup>. Even assuming that 100% of this power is emitted in the EGRET energy band, the resulting flux of $`6\times 10^{11}`$ ergs cm<sup>-2</sup> s<sup>-1</sup> is 20 times less that the EGRET measured average flux of 3EG J2016+3657. This argument is insensitive to distance as long as the X-ray synchrotron nebula is considered a calorimeter of the present pulsar power. If, instead, CTB 87 hosts a Geminga-like pulsar whose energy is no longer trapped by the nebula, and is maximally efficient in the production of $`\gamma `$-rays, then we would expect a spin-down power of only $`3\times 10^{34}`$ ergs s<sup>-1</sup>. Such a pulsar is inadequate to explain the flux of 3EG J2016+3657 unless it were at $`d<500`$ pc, which is certainly ruled out by the H I and X-ray measured column density to the SNR. We believe that the SNR CTB 87 is an interesting Crab-like remnant, but is too weak and too far away to be a good candidate for the EGRET source 3EG J2016+3657.
Similarly, neither the bright stars in the X-ray images nor the H II regions are likely to be responsible for the EGRET source.
We believe that the most likely candidate for 3EG J2016+3657 is the radio source B2013+370, described in §4. Based on its radio properties B2013+370 is very likely to be a blazar, similar to the others seen by EGRET. Mattox et al. (1997) find that only the brightest radio-flat AGN can be identified with EGRET sources with any level of confidence, and demonstrate that there is a high degree of correlation between $`\gamma `$-ray and radio fluxes of EGRET blazars. EGRET has detected more than 65 active galactic nuclei (AGN) (Hartman et al. 1999), almost all of which can be classified as blazars. The blazars seen by EGRET all share the common characteristic that they are radio-loud, flat spectrum sources, with radio spectral indices $`0.6>\alpha >0.6`$ (von Montigny et al. 1995). Several of these blazars exhibit superluminal motion of components resolved with VLBI (e. g., 3C 279, 3C 273, PKS 0528+134). The blazar class of AGN includes highly polarized quasars, BL Lac objects, or optically violent variable (OVV) quasars. The sources are characterized by one or more properties of this source class, namely, a flat radio spectrum, a continuum spectrum that is non-thermal, optical polarization and strong variability. For most EGRET blazars, the $`\gamma `$-ray luminosity dominates that other wavebands.
The probability that the blazar B2013+370 is the correct identification for the EGRET source 3EG J2016+3657 can be estimated by following the calculations presented by Mattox et al. (1997). The a priori probability that EGRET will detect a random flat-spectrum blazar with a 5 GHz flux of 2 Jy is 5.8% (Mattox et al. 1997). However, since in this case there is an EGRET source at this location, the conditional (a posteriori) probability must be used, which takes into account the fact that a gamma-ray source has already been detected. The a posteriori probability that a blazar of the type that we see in the error circle of an EGRET source is in fact the correct identification is about 98.8%. The factors that enter into this calculation are the radio flux (2 Jy) and spectral index (+0.3), the 95% error radius of the EGRET source (0.55), the distance of the radio source from the center of the EGRET circle (0.27), and the mean distance between radio sources which are at least as strong and at least as flat as this one ($`16.7^{}`$.)
Fig. 8 shows the spectral energy distribution of B2013+370, assuming that it is the EGRET source 3EG J2016+3657. The figure shows the relative amounts of energy detected in equal logarithmic frequency ranges. The radio fluxes were obtained from the NRAO/VLA Sky Survey, the Westerbork Northern Sky Survey (see §4.2 ), and from a compilation of radio fluxes in Weiler and Shaver (1978). The fluxes at mm wavelengths were obtained from Reynolds et al. (1997), taken with the IRAM 30 m telescope. The optical point was measured by us at the MDM Observatory and has been corrected for extinction (§4.1). The estimated unabsorbed X-ray flux for source 3 (=B2013+370) in the Asca band of 1 – 10 keV is $`(6\pm 1)\times 10^{13}`$ erg cm<sup>-2</sup> s<sup>-1</sup>, and in the ROSAT band of 0.1 – 2.4 keV is $`(1.9\pm 0.2)\times 10^{12}`$ erg cm<sup>-2</sup> s<sup>-1</sup>. For the flux estimations we have assumed $`N_H=10^{22}`$ cm<sup>-2</sup>, and a power-law photon index of 2. The data in Fig. 8 are not contemporaneous. The EGRET spectrum was derived as explained in §2.
The broad band spectrum shown in Fig. 8 is that of a typical EGRET blazar, dominated by the power output in $`\gamma `$-rays. The spectrum shows the characteristic features of a blazar, with a synchrotron peak at lower energies, and an inverse Compton peak at higher energies. The high $`\gamma `$-ray luminosity of a blazar suggests that the emission is likely to be beamed, and therefore Doppler-boosted, along the line of sight. In this scenario, synchrotron radiation from high-energy electrons in a relativistically outflowing jet are responsible for the radio to UV continuum. The high-energy photons come from inverse Compton scattering of low-energy photons by the same relativistic electrons in the jet. Details of this model remain unresolved (e.g. see Hartman et al. 1997, for a review). The relative power output in $`\gamma `$-rays for B2013+370 is less than that of a typical flat-spectrum radio quasar seen by EGRET, and is more similar to that observed in BL Lac objects. The location of the broad synchrotron peak in the optical-IR band rather than in the X-ray band is an indication that B2013+370 could be a low-energy peaked BL Lac object (LBL), according to the classification suggested by Giommi & Padovani (1994).
We believe that the association of 3EG J2016+3657 with B2013+370 is real and that the source of both is most likely a blazar.
In conclusion, we have made a comprehensive study of the X-ray and $`\gamma `$-ray sources in the error circle of the COS–B source 2CG 075+00. We have identified most of the X-ray sources in the error boxes of the EGRET sources. One reasonable hypothesis is that 2CG 075+00 is largely the same source as 3EG J2021+3716 and that 3EG J2016+3657 is a new source which can be identified with the radio blazar 2013+3710 = X-ray source 3. Herein 2CG 075+00 = 3EG J2021+3716 remains unidentified. Clearly, for the 3EG sources considered here, we need more refined $`\gamma `$-ray positions and extensive monitoring (possibly by AGILE and GLAST) to establish their ultimate nature.
We thank Jonathan Kemp for obtaining optical images at MDM Observatory. We acknowledge support by NASA Grant NAG5-3696 (R. M.), NASA Grant NAG5–7935 (E. V. G.), and NASA Grant NAG5-3229 (J. P. H). This research has made use of data obtained from HEASARC at Goddard Space Flight Center and the SIMBAD astronomical database.
Figure Captions
Fig. 1. — ROSAT soft X-ray image of 3EG J2016+3657 and 3EG J2021+3716. The circles for the two 3EG sources correspond to the $`95`$ % confidence contours. The dashed circle corresponds to the COS–B source 2CG 075+00. The GeV Catalog source is also shown.
Fig. 2. — EGRET $`\gamma `$-ray light curves for (a) 3EG J2016+3657 and (b) 3EG J2021+3716 from 1991 to 1997. $`2\sigma `$ upper limits are shown as downward arrows. The horizontal error bars correspond to the extent of an individual observation.
Fig. 3. — ASCA image of 3EG J2016+3657 and 3EG J2021+3716. The circles for the two 3EG sources correspond to the $`95`$ % confidence contours. The dashed circle corresponds to the COS–B source 2CG 075+00. The GeV Catalog source is also shown.
Fig. 4. — ROSAT HRI X-ray image of the field around 3EG J2016+3657. The image shows the sources 2 and 3 (B2013+370) as clearly resolved point sources. The image is scaled to highlight point features. The source numbers correspond to those listed in Table 1.
Fig. 5. — Flux history of B2013+370 at 90 GHz and 140 GHz between 1993 and 1995. (Data compiled from Reuter et al. 1997).
Fig. 6. — A section of the combined $`R`$-band CCD images from the MDM 2.4m telescope. Total exposure time was 30 min. The field shown is $`70^{\prime \prime }\times 70^{\prime \prime }`$. Tick marks are centered on the NVSS position of the blazar B2013+370. The optical object between the tick marks has $`R=21.6\pm 0.2`$, and is located at (J2000) 20 15 28.76, +37 10 59.9.
Fig. 7. — ROSAT PSPC Image of 3EG J2016+3657 and 3EG J2021+3716 showing the bright 1.4 GHz radio sources from the NRAO/VLA Sky Survey within the error circles of the EGRET sources. Only those sources with radio fluxes $`>0.3`$ Jy are shown. Refer to Table 2 for fluxes and source positions. The error circles of 2CG 075+00, the two EGRET catalog sources, and the GeV Catalog source are indicated.
Fig. 8. — Broad band spectrum of B2013+370, assuming that it is the $`\gamma `$-ray source 3EG J2016+3657.
|
warning/0005/astro-ph0005281.html
|
ar5iv
|
text
|
# A DYNAMICAL STUDY OF GALAXIES IN THE HICKSON COMPACT GROUPS
## 1 INTRODUCTION
Dynamical properties of a galaxy are basically governed by both the mass and angular-momentum distributions in the galaxy. Since such dynamical properties are deeply related to the formation and evolution of galaxies, many dynamical studies have been done for various kinds of galaxies (e.g., Rubin et al. 1985; see for recent papers, Rubin, Waterman, & Kenney 1999; Sofue et al. 1999). Dynamical properties also provide important information on the interaction between galaxies (Keel 1993, 1996; Chengalur et al. 1994; Márquez & Moles 1996; Barton, Bromley, & Gellar 1999) and on the galaxy environment such as clusters of galaxies (Rubin et al. 1988, 1999; Whitmore et al. 1988).
In addition to the central region of clusters of galaxies, compact groups (CGs) of galaxies are also useful laboratories to investigate violent interactions between/among galaxies because they are small and isolated systems whose galaxy number densities are comparable to those of the center of cluster of galaxies; e.g., $`10^{46}`$ galaxies Mpc<sup>-3</sup> (Shakhbazyan 1973; Rose 1977; Hickson 1982). Among such CGs, the Hickson compact groups (HCGs) of galaxies have been studied extensively (Hickson 1982, 1993). Many galaxies in the HCGs show peculiar morphologies such as tidal tails, tidal bridges, distorted isophotes, shell structures, and so on (Mendes de Oliveira & Hickson 1994). It is also known that a number of early-type galaxies in the HCGs have unusually blue colors (Zepf et al. 1991; Moles et al. 1994). These observational results suggest that galaxies in the HCGs have experienced frequent dynamical interactions. Indeed, Rubin et al. (1991) showed that rotation curves of many spiral galaxies in HCGs appear abnormal.
In order to investigate the effect of dynamical interactions in the CG environment, we newly conducted an optical spectroscopy program of HCG galaxies (Shimada et al. 2000: Paper I). This paper presents results of statistical studies with rotation curve properties of HCG spirals. We describe our observation and the data reduction in section 2. Making rotation curve, estimation of rotation curve asymmetry, classifying rotation curve shapes are described in section 3. In section 4 we compare the rotation curve properties of the HCG spiral galaxies with the optical morphologies, nuclear activities, group properties (the group size, the velocity dispersion, the galaxy number density, and the crossing time). We compare the rotation curve properties between the HCG spirals and field ones in section 5 and clusters ones in section 6. In section 7 we discuss our results.
We adopt a Hubble constant $`H_0=100`$ km s<sup>-1</sup> Mpc<sup>-1</sup> and a deceleration parameter $`q_0=0`$ throughout this paper.
## 2 OBSERVATIONS
We have obtained optical long-slit spectra along major-axis of 30 galaxies (mostly disk galaxies) in 20 HCGs. The sample galaxies were selected randomly from the HCG catalog (Hickson 1993). The optical spectroscopy was made using the new Cassegrain spectrograph with an SITe 512 $`\times `$ 512 CCD camera attached to the 188 cm telescope at the Okayama Astrophysical Observatory (OAO) during a period between 1996 February and 1997 January. A journal of the observations is given in Table 1. Basic data of the observed galaxies from Hickson (1993) are summarized in Table 2. As for the morphology type, we preferentially adopted the Hubble Type taken from de Vaucouleurs et al. (1991, hereafter RC3). For galaxies whose Hubble type are uncertain in RC3, we adopted the Hubble type taken from Hickson (1993).
A long (5 arcmin) slit with a width of 1.8 arcsec was used and put on each target galaxy with a position angle of the major axis. The 600 grooves mm<sup>-1</sup> grating was used to cover 6300 – 7050 Å region with the spectral resolution of 3.4 Å ($``$ 157 km s<sup>-1</sup> in velocity at 6500 Å). Two-pixel binning was made of the CCD along the slit and thus the spatial resolution was 1.75 arcsec per element. The typical seeing during the runs was 2 arcsec.
The data were analyzed using IRAF<sup>1</sup><sup>1</sup>1Image Reduction and Analysis Facility (IRAF) is distributed by the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation.. We also used a special data reduction package, SNGRED (Kosugi et al. 1995), developed for OAO new Cassegrain spectrograph data. The reduction was made with a standard procedure; bias subtraction, flat fielding with the data of the dome flats, and cosmic ray removal. Flux calibration was obtained using standard stars available in IRAF.
## 3 RESULTS
### 3.1 Major-axis Velocity Curves
We use the H$`\alpha `$ emission line to construct a heliocentric velocity curve as a function of the radial distance ($`r`$) from the nucleus for each galaxy; i.e., $`V_{\mathrm{obs}}(r)=c[\lambda _{\mathrm{obs}}(r)/\lambda _01]`$ where $`\lambda _{\mathrm{obs}}(r)`$ and $`\lambda _0`$ are the measured and the rest-frame wavelengths of H$`\alpha `$, respectively. For each galaxy, we derive the rotation velocity curve correcting for the inclination effect; i.e., $`V(r)=[V_{\mathrm{obs}}(r)V_{}]/[(1+V_{}/c)\mathrm{sin}i]`$ where $`V_{}`$ is the heliocentric velocity of the galaxy center, $`c`$ is the light velocity, and $`i`$ is the inclination angle of the galaxy ($`i=0^{}`$ corresponds to the face-on view). Following Rubin et al. (1982), we estimate the inclination angle using a relation of $`\mathrm{sin}i=1.042^{0.5}(110^{2x})^{0.5}`$ where $`x=\mathrm{log}(R_{\mathrm{major}}/R_{\mathrm{minor}})`$; $`R_{\mathrm{major}}`$ and $`R_{\mathrm{minor}}`$ are the semimajor and semiminor axis of the isophote at 25 mag arcsec<sup>-2</sup> in the B band, respectively (Hickson 1993). The adopted values of $`\mathrm{sin}i`$ are given in Table 2. Distances from galactic nucleus $`r`$, rotation velocities $`V(r)`$ and their 1 $`\sigma `$ fitting error $`dV(r)`$ are given in Table 15 (Appendix). In Figure 1, we show the rotation velocity curves as a function of distance from the galactic nucleus in units of arcsec for the observed galaxies.
### 3.2 Asymmetry of the Rotation Curve
It is known that galaxy collisions disturb the rotation curves of galaxies. The kinematical effect due to the tidal disturbance is often different between the side facing to the colliding partner and the opposite side and thus the rotation curve tends to show an asymmetrical property (Barton et al. 1999 and references therein). Therefore, it is interesting to investigate the asymmetry of the rotation curve. In order to quantify the asymmetry of the rotation curve, we define an asymmetry parameter $`A`$,
$$A\left[\frac{1}{N}\underset{j}{\overset{N}{}}\left(\frac{\left[V(r_j)V(r_j)\right]}{\left[V(r_j)+V(r_j)\right]}\right)^2\right]^{1/2},$$
(1)
where $`j`$ is the bin number along the major axes and $`N`$ is the total bin number. Here we use the data between $`r=0.2R_{25}`$ and $`r=0.5R_{25}`$, where $`R_{25}`$ is length of the radius of the isophote at 25 mag arcsec<sup>-2</sup> in the $`B`$ band. The reason for this is as follows. The minimum radius, $`0.2R_{25}`$, is adopted to exclude the data in the central region of galaxies where the tidal disturbance is expected to be negligibly small; i.e., if we include the data with $`r<0.2R_{25}`$, the difference of the asymmetry parameter among the galaxies could be less pronounced. On the other hand, the maximum radius, $`0.5R_{25}`$, is adopted to cover the observed rotation curves for most of the galaxies studied here. According to the definition of $`A`$, galaxies with higher asymmetric rotation curves tend to have larger values of $`A`$. The results are given in Table 3. We cannot estimate $`A`$ for six spirals (HCG 37b, 47a, 61c, 87a, 92c, and 96a) because of the small data points in their rotation curves.
### 3.3 Shape of the Rotation Curves
As shown in Figure 1, the observed velocity curves show various shapes. However, we simply adopt following three types of shapes; 1) Type “f”: the rotation velocity monotonically rises near the center and tend to be almost flat at $`r/R_{25}1`$; note that most ordinary spiral galaxies have this type of rotation curves (e.g., Rubin et al. 1985), 2) Type “fp”: the rotation curve appears almost flat but some dips and/or bumps are seen, 3) Type “p”: the rotation curve shows a significantly peculiar shape. Rotation curves with a sinusoidal shape or a linearly-rising shape are included in this type. The results of our classification are given in Table 3. Note that rotation curves of two galaxies (HCG 37b, and 92c) cannot be classified because there are only few data points.
## 4 THE ROTATION CURVE PROPERTIES OF HCG GALAXIES
### 4.1 Enlarged HCG Sample
Among the 30 galaxies observed by us, two galaxies are redshift-discordant galaxies in the HCGs (HCG 73a and 92a). One of remaining 28 galaxies is classified as an S0 in RC3 (HCG 87a). Therefore, our HCG sample contains 27 spiral galaxies. In order to enlarge the sample, we add HCG galaxies observed by Rubin et al. (1991) (see Table 4). Their sample contains 34 spiral galaxies. Excluding two no-available galaxies (HCG 10a and 37c) of which rotation curve data were not shown, two redshift-discordant galaxies (HCG 31d and 78a), and five S0 galaxies classified in RC3 (HCG 16c, 16d, 23a, 34b, and 57e), we obtained the rotation curve data of the 25 HCG spirals from Rubin et al. (1991). For these 25 HCG spiral galaxies, we estimate the asymmetry parameter and classify the rotation curve shape with the same method for our HCG data. The results are also given in Table 4. Twelve of these 25 spiral galaxies are commonly observed (HCG 31a, 31b, 31c, 37b, 40c, 44a, 44b, 44d, 79d, 88a, 88c, and 89a). The $`A`$ values of these twelve spirals are adopted the averaged $`A`$. Finally, we obtain an enlarged HCG sample which contains 40 spiral galaxies. Hereafter, we discuss the rotation curves of HCG spirals with the enlarged HCG spiral sample containing 40 spirals.
### 4.2 Rotation Curve Properties versus Morphologies of Host Galaxies
Mendes de Oliveira & Hickson (1994) showed that about a half of galaxies in HCGs have peculiar morphologies such as tidal tails, tidal debris, distorted isophotes, and so on, indicating evidence for galaxy collisions. On the other hand, Rubin et al. (1991) found that there is a loose correlation between peculiarity of rotation curve and peculiarity of morphology. In Table 5, we classified our HCGs spiral galaxies into two categories based on the optical morphology from Mendes de Oliveira & Hickson (1994) and RC3. The spiral galaxies with or without optical peculiar morphologies are listed in Table 5. We compare the shape of rotation curves with the optical morphologies of host galaxies. The result is shown in Table 6. Although it is generally expected that spiral galaxies with normal rotation curve have normal morphology and those with peculiar rotation curve have peculiar morphology, there are normal morphology galaxies with peculiar kinematics, and peculiar morphology galaxies with normal kinematics. We adopted the null hypothesis that distribution for the HCG spiral galaxies with peculiar morphology is the same as that for with normal morphology and apply the $`\chi ^2`$ test. The result is shown in Table 6. We obtained a probability, $`P(\chi ^2)=0.818`$. We find that there is no significantly statistical difference between the HCGs spiral galaxies with peculiar morphology and that with normal morphology.
### 4.3 Rotation Curve Properties versus Nuclear Activities
It has often been considered that frequent galaxy collisions trigger some nuclear activities (active galactic nuclei or intense star formation) in HCG member galaxies. Therefore, it is interesting to compare the kinematical properties in the member galaxies with the nuclear activity.
Using observed flux ratio of $`[\mathrm{N}\mathrm{ii}]`$ to H$`\alpha `$, we classify the activities of galaxies into active galactic nuclei (AGN) and Hii nucleus galaxies (Hii); galaxies with $`f([\mathrm{N}\mathrm{ii}])/f(\mathrm{H}\alpha )0.6`$ are classified into AGN, and those with $`f([\mathrm{N}\mathrm{ii}])/f(\mathrm{H}\alpha )<0.6`$ are classified into Hii (Ho et al. 1997). The $`f([\mathrm{N}\mathrm{ii}])/f(\mathrm{H}\alpha `$) ratio and the activity type for each galaxy are listed in Table 5. In Figure 2, we compare the distributions of the asymmetry parameter $`A`$ between AGNs and Hii nuclei. We apply the Kolmogorov-Smirnov (KS) test (e.g. Press et al. 1988). The null hypothesis is that the distributions of $`A`$ of both the AGN and the Hii nuclei come from the same underlying population. We obtain a KS probability, 0.947. This indicates that there is no difference in the distribution of the asymmetry parameter between the HCG AGN and the HCG Hii nuclei. In Figure 3, we show diagrams of $`[\mathrm{N}\mathrm{ii}]/\mathrm{H}\alpha `$ ratio against the asymmetry parameter $`A`$. The left panel is the diagram for the case of early-type (S0/a-Sbc) and late-type (Sc and later) spirals and the right panel is the one for the case of “f” rotation curves, “fp” ones, and “p” ones. We adopt the null hypothesis that the asymmetry parameter $`A`$ is not correlated with $`[\mathrm{N}\mathrm{ii}]/\mathrm{H}\alpha `$ ratio and apply the Spearman-rank statistical test for all the case shown in Figure 3. A summary of the statistical tests is given in Table 7. There is no correlation between the $`A`$ and $`[\mathrm{N}\mathrm{ii}]/\mathrm{H}\alpha `$ ratio in the HCG spirals.
### 4.4 Rotation Curve Properties versus Group Properties
Rubin et al. (1991) mentioned that there appears no correlation between normal or abnormal rotation curves and the velocity dispersion of the group. In order to confirm this, we investigate correlations between dynamics and other properties of group. From previous studies by Hickson et al. (1988, 1992), the group size, the velocity dispersion, the galaxy number density, and the crossing time are listed in Table 5 for each HCG galaxy. Figure 4(a)-(d) show relations between the asymmetry parameter $`A`$ and the group size (Hickson et a. 1992), the velocity dispersion (Hickson et al. 1992), the galaxy number density (Hickson et al. 1988), and the crossing time (Hickson et al. 1992) for case of early-type (S0/a-Sbc), late-type (Sc and later), “f” rotation curves, “fp” ones, and “p” ones. We adopt the null hypothesis that the asymmetry parameter $`A`$ is not correlated with each group property and apply the Spearman-rank statistical test for all the correlations shown in Figure 4(a)-(d). A summary of the statistical tests is given in Table 8. Although the probability between $`A`$ and group velocity dispersion for “f” rotation curves is $`4.81\times 10^3`$, the three $`\sigma `$ confidence level is $`1.3\times 10^3`$. Therefore, for any cases, it is found that there is no correlation between the asymmetry parameter $`A`$ and the group properties.
## 5 COMPARISON OF ROTATION CURVES BETWEEN HCG GALAXIES AND FIELD GALAXIES
### 5.1 The Field Sample
In order to investigate how the dynamical properties of HCG galaxies are different from those of isolated galaxies, we need a reference sample of field galaxies. Rubin et al. (1980, 1982, 1985) published rotation curves for 60 spiral galaxies (16 Sa, 23 Sb, and 21 Sc galaxies). Excluding two Virgo spiral galaxies (NGC 4321 and NGC 4419) and two galaxies in groups (NGC 1353 and NGC 4448), we adopt 56 spiral galaxies (15 Sa, 21 Sb, and 20 Sc) as a field galaxy sample. The basic data of these field galaxies are summarized in Table 9.
Unfortunately, Rubin et al. (1980, 1982, 1985) show only average rotation velocity $`V(|r|)`$ which is the mean of velocities within the radial bins of both $`r`$ and $`r`$, and 1 $`\sigma `$ error of the mean, $`dV(|r|)`$. It is hard to know how many data points either $`r`$ or $`r`$ contains and to estimate $`V(r)`$ and $`V(r)`$. Therefore, we simply assume that there is only one velocity data point in each bin of $`r`$ and $`r`$. Making $`V(r)`$ and $`V(r)`$ that here we reconstruct with $`V(|r|)`$ and $`dV(|r|)`$ of Rubin et al. (1980, 1982, 1985) under such assumption, one finds that the rotation velocity at distance $`r`$ is $`V(r)=V(|r|)+dV(|r|)/\sqrt{2}`$ and that the rotation velocity at distance $`r`$ is $`V(r)=V(|r|)dV(|r|)/\sqrt{2}`$. In this case we can define the asymmetry parameter $`A`$ as,
$$A\frac{1}{\sqrt{2}}\left[\frac{1}{N}\underset{j}{\overset{N}{}}\left(\frac{dV(|r_j|)}{V(|r_j|)}\right)^2\right]^{1/2}.$$
(2)
The meaning of $`j`$ and $`N`$ are the same as those for the equation (1). Note that the asymmetry parameters $`A`$ for field spirals are all upper limits according to the above definition.
We also classify the shape of rotation curves for field spiral galaxies. The asymmetry parameter $`A`$ and the shape of the rotation curves for the field spiral galaxies are listed in Table 9.
### 5.2 Comparison of the Rotation Velocity between the HCG Spiral Galaxies and the Field Spiral Galaxies
In Figure 5, we show the rotation curves of our HCGs spiral galaxies as a function of distance from galactic nucleus in units of $`R_{25}`$. We also show expected rotation curves of the field galaxies calculated from synthetic rotation curves in Rubin et al. (1985). The rotation curves of the HCG spiral galaxies are different from those of the field spiral galaxies. Many spiral galaxies in HCGs have peculiar rotation curves.
By numerical simulations, Barton et al. (1999) found that galaxy collision caused rising rotation curve. Some HCG spirals have such a rising rotation curve (e.g. HCG 61c and 88a). There are also spiral galaxies with higher rotation velocities (e.g. HCG 40c) and those with lower rotation velocities (e.g. HCG 88b and 88d) with respect to the field spirals with both the same Hubble type and the same luminosity.
Rubin et al. (1991) show that the HCG spirals tend to have lower rotation velocity with respect to the field ones. For some HCG spirals, we found the same tendency. However, it is difficult to determine intrinsic inclination angles i of HCG spirals showing peculiar morphologies. Although we estimated the inclination angles i of the HCG spirals, there is an uncertainty of i. Most of the rotation curves of HCG spirals are too peculiar to discuss the mass distribution in the galaxies.
### 5.3 Comparison of the Asymmetry Parameter between the HCG Spiral Galaxies and the Field Spiral Galaxies
We compare the frequency distributions of $`A`$ between the HCGs and the field spiral galaxies in Figure 6. The comparison is carried out for the case of all Hubble type, S0/a $``$ Sbc type, and Sc and later type galaxies. We include two field spiral galaxies with $`A=0`$ into the smallest bin (NGC 2608 and IC 724). We find that $`A`$ values of the HCG spirals are generally larger than those of the field spirals for all the morphological types. In particular, we mention that the maximum value of $`A`$ of the field spirals is 0.06 (NGC 3054 and NGC 7171) while 18 HCGs spirals (53%) have $`A>1.0`$.
We apply the KS test. The null hypothesis is that the observed distributions of $`A`$ of both the HCG galaxies and the field ones come from the same underlying population. We obtain KS possibilities (see Table 10); $`1.15\times 10^{11}`$ for all the galaxies, $`1.76\times 10^6`$ for S0/a – Sbc galaxies, and $`7.40\times 10^7`$ for Sc and later-type spiral galaxies. Therefore, we conclude that the HCG galaxies tend to have asymmetrical rotation curves with respect to the field galaxies. This is consistent with the result of Rubin et al. (1991).
For the HCGs spirals, we find that the late-type (Sc and later) spirals have larger $`A`$ than the early-type (S0/a $``$ Sbc) ones. Applying the KS test, we obtain a probability of $`2.32\times 10^3`$ between early-type HCG spirals and late-type HCG spirals while 0.792 for a comparison for that between early-type spirals in the field and late-type spirals in the field (see Table 10). Therefore, it is suggested that the HCG late-type spirals tend to have more asymmetrical rotation curves with respect to the HCG early-type ones.
### 5.4 Comparison of the Rotation Curve Shape between the HCG Spiral Galaxies and the Field Spiral Galaxies
We compare the rotation curve shapes between the HCG galaxies and the field ones. In Figure 7 we compare the frequency distributions of rotation curve shapes between the HCGs spirals and the field ones for all galaxies, early-type spirals, and late-type spirals. We adopt the null hypothesis that the distribution for the HCGs spirals is the same as that for the field ones and apply the $`\chi ^2`$ test. The results are given in Table 11. We find that there are significant statistical differences between the HCGs spirals and the field ones.
Among the 39 HCG galaxies, 33 ($``$85%) galaxies have either “fp” or “p” rotation curves. It is remarkable that almost all the HCG late-type spirals ($``$93%) have “p” rotation curves. On the contrary, only 42% of the HCG early-type spirals have “p” rotation curves and 21% of the HCG early-type spirals have “f” rotation curves.
The probability of the $`\chi ^2`$ test between the early-type field spirals and late-type field spirals is 0.673, while that between the early-type HCG spirals and the late-type HCG spirals is $`4.44\times 10^3`$. Although the latter probability suggests a marginally significant difference, a larger sample will be necessary to confirm it.
## 6 COMPARISON OF ROTATION CURVES BETWEEN HCG GALAXIES AND CLUSTER GALAXIES
### 6.1 The Cluster Sample
As we previously mentioned, the local galaxy number density of HCGs is comparable to that of cluster centers. Therefore it is interesting to compare the dynamical properties of the HCG spiral galaxies with those of cluster spiral galaxies. We use a sample of cluster spiral galaxies taken from Amram et al. (1992, 1994, and 1995) who published rotation curves of 41 spiral galaxies in eight clusters. The basic data of these cluster galaxies are summarized in Table 12. Their morphological types are taken from RC3 if available (32 galaxies). For the remaining nine spiral galaxies without RC3, we adopt the morphological types taken from Bell & Whitmore (1989), Amram et al. (1995), and Gavazzi & Boselli (1996).
Using the equation (1), we calculated the asymmetry parameter $`A`$ of the cluster spiral galaxies. We also classified the shape of the rotation curves for them. The results are also listed in Table 12.
### 6.2 Comparison of the Asymmetry Parameter between the HCG Spiral Galaxies and the Cluster Spiral Galaxies
In Figure 6, we compare the frequency distribution of the asymmetry parameter $`A`$ between the HCG spiral galaxies and the cluster ones. The comparison is carried out for the case of all Hubble types, S0/a – Sbc type, and Sc and later type galaxies. The null hypothesis is that the observed distributions of $`A`$ of both HCG spirals and the cluster ones come from the same underlying population. Applying the KS test, we obtain the following KS probabilities (see Table 13); $`6.64\times 10^4`$ for all the galaxies, $`4.50\times 10^2`$ for the S0/a – Sbc spiral galaxies, and $`8.48\times 10^4`$ for the Sc and later-type spiral galaxies. We find that the $`A`$ values of HCG spiral galaxies tend to be larger than those of the cluster ones for all the case.
### 6.3 Comparison of the Rotation Curve Shape between the HCG Spiral Galaxies and the Cluster Spiral Galaxies
We show the frequency distribution of the rotation curve shape for the HCG, the field, and the cluster spirals in Figure 7. We compare the both distributions for the case of all Hubble type, S0/a-Sbc type, and Sc and later-type spirals. The null hypothesis is that the observed distributions of the rotation curve shape of both the HCG spirals and the cluster ones come from the same underlying population. Applying the $`\chi ^2`$ test, we obtain the following probabilities (see Table 14); $`9.28\times 10^7`$ for all the galaxies, $`3.84\times 10^3`$ for the S0/a-Sbc spirals, and $`3.23\times 10^5`$ for the Sc and later-type spirals. These results suggest that the HCG spirals tend to show more peculiar shapes of rotation curve.
## 7 DISCUSSION
In this paper, we have investigated the asymmetry and the shape of the rotation curves of spiral galaxies in HCGs.
First, we investigated the relation between the morphological peculiarity and dynamical peculiarity of HCG spiral galaxies. We found that there is no statistical difference between them (Table 6), being consistent with the finding by Rubin et al. (1991). It is likely that the HCG spirals with both the morphological peculiarity and the dynamical peculiarity have experienced recent galaxy collisions. However, there are normal-morphology galaxies with peculiar dynamical properties and peculiar-morphology ones with normal dynamical properties. No correlation between the dynamical peculiarity and the morphological peculiarity suggests that the dynamical properties of the HCG spirals may be governed by galaxy collision parameters such as the difference of masses, and the orbital parameters. While the morphological peculiarities such as tidal tails and tidal bridges, and asymmetry come to be more clearly seen in outer regions of galaxies, the dynamical peculiarities probed by the H$`\alpha `$ line emission are observed in inner regions ($`r<0.5R_{25}`$) of galaxies (see Figure 5). The morphological peculiarity can be more easily induced by galaxy collisions than the dynamical peculiarity. Thus, weak galaxy collisions could not perturb the galaxy rotation curves, although it is experienced that morphological deformation could be induced in outer parts of the galaxies. On the other hand, minor mergers could perturb the rotation curve in inner regions of galaxies without causing global morphological peculiarities.
Second, we investigated the relation between the properties of the rotation curves and the properties of the nuclear activity. Since it has often been suggested that the galaxy collisions trigger nuclear activities such as AGN and nuclear starburst phenomena (Kennicutt & Keel 1984; Keel 1996), it is expected that such nuclear activities are more often observed in HCG galaxies. However, we found that there is no significantly statistical difference in the distribution of the asymmetry parameter $`A`$ between the nuclear activities (Figure 2). We also found that there is no correlation between the asymmetry parameter $`A`$ and $`[`$Nii$`]`$/H$`\alpha `$ ratio (Figure 3). All these indicate that galaxy collisions do not always trigger the nuclear activity
Third, we investigated the relation between the dynamical properties of HCG spiral galaxies and the properties of the compact groups such as the group size, the velocity dispersion of member galaxies, the galaxy number density, and the crossing time (see Figure 4a-4d and Table 8). In compact groups with higher number densities and with smaller sizes, more frequent galaxy collision would occur and thus spiral galaxies in such groups would show more asymmetric rotation curves. However, we found that there are no significant correlations between the asymmetry parameter $`A`$ and any group properties. There may be two alternative interpretations for this finding. 1) Many compact groups containing spiral galaxies are false groups; e.g., a pair of galaxies with a few field galaxies (Mamon 1986, 1992, 1994, 1995; Hernquist et al. 1995). 2) The projected size and the observed velocity dispersion of the compact groups are significantly different from the real (i.e., three-dimensional) ones. As mentioned in section 5.4, 85% of HCG spirals are peculiar and that 93 % of the late type HCG spirals are peculiar. This high rate of peculiarity seems inconsistent with the false group hypothesis. However, these two possibilities will be taken into account in future studies.
Finally, we compared the dynamical properties of spiral galaxies among the field, the clusters, and the HCGs. We found that the HCG spirals tend to have more asymmetric and more peculiar rotation curves with respect to the field and the cluster spirals (see Figure 6 and 7, Table 10, 11, 13, and 14). It is interesting to note that these are the significant differences in the distributions of the $`A`$ and the rotation curve morphologies between the HCG early-type spirals and the HCG late-type ones, although we found no such difference between the field early-type spirals and the field late-type ones. This may be attributed to the stabilization of a disk by a large bulge of early-type spirals (Mihos & Hernquist 1994; Velázquez & White 1999). Since a small bulge cannot stabilize the disk, late-type spirals are more sensitive to galaxy interactions than the early-type spirals and are expected to show the peculiar kinematical properties for a longer duration.
We are grateful to the staff of OAO for kind help of the observations. We would like to thank Daisuke Kawata and Tohru Nagao for useful comments and suggestions. TM is thankful for support from a Research Fellowship from the Japan Society for the Promotion of Science for Young Scientists. This work was partly supported by the Ministry of Education, Science, Culture, and Sports (Nos. 07044054, 10044052, and 10304013).
## Appendix A Velocities for spirals in Hickson compact groups
Table 15 lists the distances from galactic nucleus $`r`$ in units of arcsec and of $`R_{25}`$, the rotation velocities $`V(r)`$ at $`r`$, and their 1 $`\sigma `$ errors $`dV(r)`$ for each observed HCG spirals.
|
warning/0005/nlin0005066.html
|
ar5iv
|
text
|
# Spatial period-multiplying instabilities of hexagonal Faraday waves
## 1 Introduction
The classical hydrodynamic problem of parametrically driven surface waves – or Faraday waves – concerns the spontaneous generation of standing waves at the free surface of a horizontal layer of fluid when subjected to vertical oscillations whose amplitude exceeds a critical value. Its usefulness as a tool to study nonlinear pattern-forming dynamics in non-equilibrium systems is reflected in the considerable amount of interest shown in the subject by experimentalists and theoreticians alike. A review of earlier works, mostly conducted with low-viscosity fluids in small vessels and a single forcing frequency, can be found in . More recently, Edwards & Fauve have performed experiments in the small-depth, high-viscosity and large-aspect ratio regime using a forcing function with two commensurate frequency components that modulates gravity periodically. In this regime, where it can be shown that the wavenumber of the selected pattern is less sensitive to the size and shape of the container, and that long-wavelength modes are heavily damped, observations of spatially periodic patterns (stripes, squares, hexagons), circular patterns (targets and spirals) and quasi-patterns have been reported . A survey of more recent results has been carried out by Müller et al. . Over the past two years, a new class of ‘superlattice patterns’ has been independently observed by Kudrolli et al. and Arbell & Fineberg in experiments employing two-frequency forcing functions, and by Wagner et al. in experiments using non-Newtonian fluids. These superlattices are so termed because of their distinctive feature of having spatial structures on two different length scales when viewed at any instant in time . Steady patterns that display similar characteristics have also been observed in convection experiments on fluids with temperature-dependent viscosity and have been investigated in a model of long wave convection and in reaction-diffusion systems near a Turing bifurcation .
Two types of superlattice patterns have been reported in for different parameter values. Both of them, despite their different spatial and spatio-temporal symmetry properties, are found to be possible transitions from harmonic standing hexagons as the forcing amplitude is increased. (In this context, harmonic indicates an oscillation with the same period as that of the external forcing, denoted by $`T`$, while subharmonic indicates an oscillation with twice that period.) The first of these patterns (called ‘superlattice-one’ by Kudrolli et al. ) is a harmonic response with triangular symmetry on a small scale and hexagonal lattice periodicity on a larger scale. This pattern has been studied by Silber & Proctor , who showed that it (along with standing hexagons) can arise in a bifurcation from the flat, undisturbed state when a hexagonal lattice with spatial periodicity larger than that dictated by the critical wavelength is considered . Silber & Proctor also suggested that stability might be transfered from standing hexagons to superlattice-one through an intermediate branch.
The second type of superlattice pattern (‘superlattice-two’ ), in contrast to the first, arises in a period-doubling (or subharmonic) instability of the standing hexagons. If we let $`u(𝐱,t)`$ measure the deformation of the free fluid surface at time $`t`$, it satisfies
$$u(𝐱,t+2T)=u(𝐱,t),u(𝐱,t+T)u(𝐱,t).$$
(1)
Further, this pattern exhibits a complicated mixture of spatial symmetry and time-dependent behaviour. When averaged over two periods of the driving function, its image displays hexagonal symmetry with two well-defined spatial scales in the ratio $`1:\sqrt{3}`$ (figure 1(a)). Remarkably, at any instant, a wavy, stripe-like spatial modulation destroys the average hexagonal symmetry, resulting in a pattern that appears vastly different from its time-averaged image (see figures 1(b) and 1(c)). Arbell & Fineberg (unpublished) have also found the superlattice-two pattern for similar experimental parameters.
The superlattice-two pattern presents a number of theoretical challenges that motivate this paper: the disappearance of the stripes from the time-averaged pattern; the reduced spatial period in the time-average; and the apparent $`60^{}`$ rotation symmetry of the time-average. We present a symmetry-based approach to the study of this pattern by taking the view that it arises as a symmetry-breaking instability from the underlying standing hexagons in a spatial period-multiplying bifurcation. Our aim is to classify qualitatively the range of possible bifurcating solutions and to understand how their symmetry properties can be related to the experimental observations described above. We emphasize that we are examining instabilities of fully nonlinear states, so our approach differs from weakly nonlinear studies of the primary Faraday instability.
There are three stages in our approach. First, by using the experimentally observed instantaneous spatial symmetry information of the superlattice-two instability and by making the assumption that all solutions are periodic in the plane, we can restrict all patterns to a suitably chosen spatially periodic lattice. This lattice in turn defines a compact symmetry group, which we denote by $`\mathrm{\Gamma }^s`$, with the key properties that its action leaves standing hexagons invariant and that it has a subgroup, which we denote by $`\mathrm{\Sigma }^s`$, that describes the instantaneous spatial symmetry of the bifurcating superlattice pattern. Due to the compactness and special structure of $`\mathrm{\Gamma }^s`$, we can compute explicitly all its irreducible representations. Second, we observe that since $`\mathrm{\Sigma }^s`$ is by definition the isotropy subgroup of the bifurcating solution under the action of $`\mathrm{\Gamma }^s`$, it must have a non-trivial fixed-point set. This restriction allows us to identify the one relevant irreducible representation of $`\mathrm{\Gamma }^s`$ that describes the spatial symmetry properties of the marginal eigenfunctions at the superlattice bifurcation point. Finally, by considering the action of the time-shift symmetry $`\tau _t:tt+T`$ on the period-doubling marginal modes, we obtain the irreducible representation of the full symmetry group (denoted by $`\mathrm{\Gamma }`$) and hence the normal form of the bifurcation problem. We can then invoke the equivariant branching lemma to show that there are at least six primary branches of solutions bifurcating from standing hexagons.
With one proviso, the superlattice-two pattern observed by Kudrolli et al. can be identified as one of these branches, which we show can bifurcate as a stable branch from standing hexagons. By applying techniques for studying the averaged symmetry of periodic orbits (cf ), we show that the time-average of this branch of solutions has the hexagonal lattice periodicity observed in the experiment (as in figure 1(a)); this change in the spatial length scale on time-averaging is a consequence of the branch of solutions possessing a spatio-temporal symmetry. This symmetry was not reported in . The proviso mentioned above is that our time-average pattern does not possess $`60^{}`$ rotation symmetry – we will return to this discrepancy in the final section.
A further stability analysis predicts that other patterns, displaying different spatial and spatio-temporal symmetry properties, can bifurcate as stable branches of solutions from standing hexagons in different regions of parameter space. More generally, our analysis indicates that patterns that display superlattice structures can arise in two-dimensional spatial period-multiplying bifurcations from an underlying non-trivial solution, and our approach could also be used to investigate other superlattice patterns of Arbell & Fineberg and Wagner et al. . In particular, it may be possible to analyse some of those experimental results in terms of other irreducible representations of the same group $`\mathrm{\Gamma }^s`$.
The issue of spatial period-multiplying instabilities is an interesting one that has arisen in a variety of experimental and theoretical contexts. Period-multiplying bifurcations in one lateral direction have arisen in convection problems , magnetoconvection , Taylor–Couette experiments and in numerical solutions of the Kuramoto–Sivashinsky equations . Much less is known about spatial period-multiplying bifurcations in two directions. There are now several experimental observations of this phenomenon in the Faraday wave problem as well as in convection experiments and magnetoconvection calculations .
In the next section, we introduce some fundamental definitions and results from equivariant bifurcation theory to help us describe how this problem can be cast into a theoretical framework. In section 3, we fully describe the symmetry group of the bifurcation problem that will give rise to the observed symmetry-breaking behaviour. We also show that, under suitable phenomenological assumptions, we can identify and hence explicitly compute the irreducible representation that is relevant to the action of the symmetry group on the observed bifurcating modes. The normal form of the bifurcation problem and a stability analysis are presented in section 4. Discussions of our approach and a comparison with the experiments follow in section 5.
## 2 Group theoretic ideas
In order to study the superlattice-two pattern as a symmetry-breaking instability from standing hexagons, it is necessary to identify all the symmetries that are initially present. Due to the apparent absence of side-wall effects in the observed patterns, we consider the mathematical idealisation that all physical fields are defined in a laterally unbounded domain. Standing hexagons are then easily seen to be invariant under the action generated by a reflection, a $`60^{}`$ rotation, and two linearly independent translations. The group generated by these symmetry actions is isomorphic to $`\mathrm{}^2\mathrm{D}_6`$, which is non-compact. (Here $`\mathrm{}`$ denotes the group of integers under addition, and $`\mathrm{D}_6`$ is the twelve-element symmetry group of a regular hexagon.) Consequently a bifurcation problem that is equivariant under the action of this group can have an infinite number of modes related by symmetry becoming marginally stable simultaneously. This difficulty can be resolved if we restrict possible solutions to doubly-periodic functions defined on a suitably chosen lattice, an assumption justified by the distinct spatial periodicity of the observed patterns. A suitable lattice, which can be viewed as a finite cell with periodic boundary conditions, is one that captures the spatial periodicity of both the bifurcating modes as well as the standing hexagons. With respect to such a periodic cell, the symmetries that leave standing hexagons invariant now form a finite, and hence compact group that can be studied via representation theory. Our task therefore, is to make use of the available symmetry information taken from experimental observations to choose a lattice on which we can define a suitable spatial symmetry group $`\mathrm{\Gamma }^s`$ with the properties outlined in section 1. The idea of suitability can be made precise after we have introduced some basic group theoretic results.
Since we are considering bifurcations from a time-periodic solution, we formulate the bifurcation problem of the superlattice-two pattern, a period-doubling instability, by expanding about standing hexagons using a stroboscopic map $`𝒢`$ in the manner described by Crawford & Knobloch and Silber & Proctor . Specifically, we are assuming that standing hexagons, a fixed point of the map $`𝒢`$, lose stability to subharmonic waves with period $`2T`$ as a bifurcation parameter $`\mu `$ is varied past zero. This implies that the linearised map $`D𝒢`$ evaluated at the fixed point has a real eigenvalue passing through the value $`1`$ as the bifurcating waves become unstable. With the assumption that all fields are defined in a periodic cell such that symmetries of the standing hexagons are described by a compact group $`\mathrm{\Gamma }^s`$, the linearised map $`D𝒢`$ has a finite number ($`p`$) of marginal eigenfunctions associated with the eigenvalue $`1`$ as $`\mu `$ crosses zero. We denote the amplitudes of these $`p`$ marginal modes at time $`t=qT,q\mathrm{}`$ by $`𝐳_q=[z_1(qT),\mathrm{},z_p(qT)]\mathrm{}^p`$. In addition, the pattern has two neutrally stable modes (eigenvalues equal to $`1`$) associated with translations of the standing hexagons (see ); the amplitudes of these two modes, which correspond to the translation of the pattern in the plane, are denoted by $`𝐝_q`$. Close to the onset of the period-doubling instability, $`𝒢`$ can be reduced to a finite-dimensional map $`𝐠`$ defined on the centre manifold spanned by these amplitudes:
$$𝐳_{q+1}=𝐠(𝐳_q;\mu ),𝐠:\mathrm{}^p\times \mathrm{}\mathrm{}^p,$$
(2)
coupled with a map $`𝐡:\mathrm{}^p\times \mathrm{}\mathrm{}^2`$ describing how the perturbation drives translations of the pattern:
$$𝐝_{q+1}=𝐝_q+𝐡(𝐳_q;\mu ).$$
(3)
The map $`𝐠`$ is forced by symmetry to be $`\mathrm{\Gamma }^s`$-equivariant:
$$\gamma 𝐠(𝐳_q;\mu )=𝐠(\gamma 𝐳_q;\mu )\text{for all}\gamma \mathrm{\Gamma }^s,$$
(4)
while the map $`𝐡`$ obeys
$$N_\gamma 𝐡(𝐳_q;\mu )=𝐡(\gamma 𝐳_q;\mu )\text{for all}\gamma \mathrm{\Gamma }^s,$$
(5)
where $`N_\gamma `$ is the $`2\times 2`$ matrix that represents how the symmetry $`\gamma `$ acts on a horizontal displacement vector . In terms of the marginal modes, standing hexagons correspond to the trivial state, $`𝐳=\mathrm{𝟎}`$. When considered as a symmetry-breaking bifurcation from the underlying standing hexagons, the superlattice pattern corresponds to a non-trivial, period-two solution to the map $`𝐠`$, denoted by $`𝐳_q^{}`$, whose instantaneous spatial symmetry is specified by its isotropy subgroup $`\mathrm{\Sigma }_{𝐳_q^{}}^s`$:
$$\mathrm{\Sigma }_{𝐳_q^{}}^s=\{\sigma \mathrm{\Gamma }^s:\sigma 𝐳_q^{}=𝐳_q^{}\}\mathrm{\Gamma }^s.$$
(6)
In fact, this solution must lie in the fixed-point subspace of $`\mathrm{\Sigma }_{𝐳_q^{}}^s`$:
$$\text{Fix}(\mathrm{\Sigma }_{𝐳_q^{}}^s)=\{𝐳\mathrm{}^p:\sigma 𝐳=𝐳,\text{for all}\sigma \mathrm{\Sigma }_{𝐳_q^{}}^s\},$$
(7)
which is a linear subspace of $`\mathrm{}^p`$ and invariant under $`𝐠`$ .
Since standing hexagons do not possess spatio-temporal symmetries , $`\mathrm{\Gamma }^s`$-equivariance is sufficient to determine the normal form of the map $`𝐠`$ in the case of (temporal) period-preserving bifurcations. However, for period-doubling bifurcations there is an extra symmetry pertaining to the normal form, related to the time-shift action $`\tau _t:tt+T`$ on the bifurcating modes. In this case, the normal form of the map $`𝐠`$ is $`\left(\mathrm{\Gamma }^s\times \mathrm{Z}_2\right)`$-equivariant . Once the full symmetry group of the normal form of $`𝐠`$ is determined, we can apply the equivariant branching lemma, which, with suitable interpretation, states that if certain non-degeneracy conditions are satisfied, there is a unique branch of bifurcating solutions for each isotropy subgroup of $`\mathrm{\Gamma }\mathrm{\Gamma }^s\times \mathrm{Z}_2`$ with a one-dimensional fixed-point subspace. So instead of solving for solutions of the nonlinear vector field $`𝐠`$, we can simply look for isotropy subgroups of $`\mathrm{\Gamma }`$ with this property. To apply the equivariant branching lemma, we need to know explicitly how symmetry acts on all the marginal modes, but experimental observations only provide information about the instantaneous spatial symmetry of one of these modes. We cannot infer directly from the observations the total number of marginal modes that are related by symmetry at the bifurcation point, nor the set of matrices that represent the action of the symmetry group $`\mathrm{\Gamma }`$ on the marginal modes and the map $`𝐠`$. However, this difficulty can be resolved if we make the (generic) assumption that the bifurcation is associated with an irreducible representation of the group $`\mathrm{\Gamma }`$, and this is where the need to invoke representation theory arises.
In order to introduce the key properties of irreducible representations (irreps) and describe how they can be computed for a finite group $`\mathrm{\Gamma }`$, we recall the following definitions .
1. A representation of the group $`\mathrm{\Gamma }`$ is a homomorphism $`\psi `$ that maps $`\mathrm{\Gamma }`$ into a set of invertible $`n\times n`$ matrices $`\mathrm{M}_\mathrm{\Gamma }`$ acting on $`\mathrm{}^n`$ or $`\mathrm{}^n`$, in other words
$$\psi \left(\gamma \right)=M_\gamma ,\gamma \mathrm{\Gamma },M_\gamma \mathrm{M}_\mathrm{\Gamma }$$
such that $`\psi \left(\gamma _1\gamma _2\right)=\psi (\gamma _1)\psi (\gamma _2)\text{for all}\gamma _1,\gamma _2\mathrm{\Gamma }`$. The integer $`n`$ is the dimension of the representation.
2. Two $`n`$-dimensional representations $`\mathrm{M}_\mathrm{\Gamma }`$ and $`\mathrm{N}_\mathrm{\Gamma }`$ of $`\mathrm{\Gamma }`$ are called equivalent if there is an invertible $`n\times n`$ matrix $`Q`$ such that for each $`\gamma \mathrm{\Gamma }`$,
$$N_\gamma =Q^1M_\gamma Q,M_\gamma \mathrm{M}_\mathrm{\Gamma },N_\gamma \mathrm{N}_\mathrm{\Gamma }.$$
3. A conjugacy class of $`\mathrm{\Gamma }`$ is a subset $`C`$ of $`\mathrm{\Gamma }`$ such that $`\gamma ^1c\gamma C,\text{for all}cC\text{and}\gamma \mathrm{\Gamma }`$.
4. The character of an element $`\gamma \mathrm{\Gamma }`$ in a representation $`\mathrm{M}_\mathrm{\Gamma }`$ is defined to be the trace of the matrix $`M_\gamma `$, and we denote this value by $`\chi _{M_\gamma }`$.
5. A representation of $`\mathrm{\Gamma }`$ on $`\mathrm{}^n`$ ($`\mathrm{}^n`$) is said to be irreducible if it does not leave invariant any proper subspace of $`\mathrm{}^n`$ ($`\mathrm{}^n`$).
Simplistically we can consider a representation as a set of $`n\times n`$ nonsingular matrices that specifies the action of $`\mathrm{\Gamma }`$ on the vector space $`\mathrm{}^n`$ or $`\mathrm{}^n`$ and at the same time preserves the group structure. It is possible to show that every representation of a finite group is equivalent to a unitary representation – one in which all matrices are unitary . A simple result of definition (iv) is that the character of the identity element in a representation is always equal to the dimension $`n`$ of that representation, and definitions (i), (iii) and (iv) imply that elements in the same conjugacy class have the same character. The characters of the irreps of $`\mathrm{\Gamma }`$ obey a set of rules inherited from the orthogonality theorem governing the underlying irreps and for simple groups such as $`\mathrm{Z}_2`$, $`\mathrm{Z}_6`$ and $`\mathrm{D}_6`$, the character tables can easily be constructed by appealing to those rules. The orthogonality theorem also implies that the number of irreps of a group is equal to the number of conjugacy classes. For finite groups with a semi-direct product structure of the form $`\mathrm{\Gamma }=𝒜`$ such that $`𝒜`$ is a normal (or invariant) subgroup of $`\mathrm{\Gamma }`$ (that is, $`\gamma ^1a\gamma 𝒜`$ for every $`a𝒜`$ and $`\gamma \mathrm{\Gamma }`$), the characters of the irreps of $`𝒜`$ and $``$ form the building blocks in determining all the characters and constructing unitary irreps of the group $`\mathrm{\Gamma }`$ via a special algorithm .
In summary, analysis of the superlattice pattern using these group theoretic tools depends on our being able to find a spatial lattice or periodic cell on which standing hexagons and the marginal modes exhibiting the observed symmetries fit. The arrangement of the standing hexagons in the periodic cell then gives us a suitable symmetry group $`\mathrm{\Gamma }^s`$, which has a subgroup $`\mathrm{\Sigma }_{𝐳_q^{}}^s`$, defined in (6), whose elements are determined from experimental observations. Once we have calculated all the characters of $`\mathrm{\Gamma }^s`$, the restriction provided by the requirement that $`\mathrm{\Sigma }_{𝐳_q^{}}^s`$ be the isotropy subgroup of the observed pattern enables us to isolate the one irrep that describes the action of $`\mathrm{\Gamma }^s`$ on all the marginal modes related to the observed bifurcating mode by symmetry. Indeterminacy in the choice of irreps can be avoided if we choose a unit cell that captures exactly one spatial period of the observed pattern. The details of this procedure are the subject of the next section.
## 3 Finding the symmetry group $`\mathrm{\Gamma }`$ and the irrep for the superlattice-two bifurcation problem
A closer examination of images obtained from the experiment reveals that it is possible to impose a hexagonal lattice on the observed patterns, whose instantaneous spatial symmetries are depicted in figure 2. The choice of lattice is not unique as it can be shown that there are many possible candidates (for example a $`\sqrt{3}:1`$ rectangular lattice), but a hexagonal lattice is a natural choice due to the symmetry of the standing hexagons. Let us denote the two generating vectors of the hexagonal lattice $``$ by $`𝐞_1,𝐞_2\mathrm{}^2`$ such that
$$|𝐞_1|=|𝐞_2|=c,$$
(8)
where $`c`$ is a scaling factor (figure 2(c)). Functions in the plane that are doubly-periodic with respect to $``$ satisfy
$$u(𝐱,t)=u(𝐱+𝐥,t),𝐱=(x,y)\mathrm{}^2,𝐥,$$
(9)
where the lattice is defined as
$$=\{n_1𝐞_1+n_2𝐞_2:(n_1,n_2)\mathrm{}^2\}.$$
First let us consider the spatial symmetries of the standing hexagons shown schematically in figure 2(a). They are invariant under the action of $`\mathrm{D}_6`$ as well as two translations, which we define as follows:
$$\tau _1:𝐱𝐱+\stackrel{~}{𝐞}_1,\tau _2:𝐱𝐱+\stackrel{~}{𝐞}_2,$$
(10)
and let $`|\stackrel{~}{𝐞}_1|=|\stackrel{~}{𝐞}_2|=c_0`$ be the observed size of the periodic cell in which the basic standing hexagons fit. Our aim is to pick a value for the scaling factor $`c`$ in (8) in terms of $`c_0`$. As indicated at the end of section 2, a suitable choice of the value $`c`$ is one that gives a hexagonal cell whose size captures precisely one spatial period of the bifurcating modes, as shown schematically in figure 2(c). In fact, the observed ratio of the two lengths $`|𝐞_i|`$ and $`|\stackrel{~}{𝐞}_i|`$ is $`c/c_0=2\sqrt{3}`$, and for this value of $`c`$, the symmetry group $`\mathrm{\Gamma }^s`$ of the standing hexagons includes non-trivial translations generated by $`\tau _1`$ and $`\tau _2`$. The structure of the group $`\tau _1,\tau _2`$ can be determined if we express each of the translations in (10) in terms of $`𝐞_1`$ and $`𝐞_2`$. We can then look for the lowest powers $`n_1,n_2,n_3,n_4\mathrm{}^+`$ such that $`\tau _1^{n_1}`$, $`\tau _2^{n_2}`$, $`\tau _1^{n_3}\tau _2^{n_4}`$ map the lattice $``$ to itself, and thus determine the order of the group $`\tau _1,\tau _2`$.
Guided by the experimental observations, we choose $`\stackrel{~}{𝐞}_1=c_0(\frac{\sqrt{3}}{2},\frac{1}{2})`$, $`\stackrel{~}{𝐞}_2=c_0(0,1)`$ such that $`𝐞_1=4\stackrel{~}{𝐞}_12\stackrel{~}{𝐞}_2=c_0(2\sqrt{3},0)`$, $`𝐞_2=2\stackrel{~}{𝐞}_1+2\stackrel{~}{𝐞}_2=c_0(\sqrt{3},3)`$ (see figure 2). The translations can now be written as
$$\tau _1:𝐱𝐱+\frac{1}{6}𝐞_1+\frac{1}{6}𝐞_2,\tau _2:𝐱𝐱\frac{1}{6}𝐞_1+\frac{1}{3}𝐞_2,$$
(11)
and we can easily show that they satisfy $`\tau _1^6=\tau _2^6=\tau _1^2\tau _2^2=\text{identity}`$, as vectors of the form $`𝐱+m_1𝐞_1+m_2𝐞_2`$ for any integers $`m_1`$ and $`m_2`$ lie in $``$ and are therefore identified. Since $`\tau _1`$ and $`\tau _2`$ commute we can also see that every element generated by $`\tau _1`$ and $`\tau _2`$ can be written as $`\tau _1^n\tau _2`$ or $`\tau _1^n`$ for $`n=0,\mathrm{},5`$. In total there are twelve different translations, forming a group that is isomorphic to $`\mathrm{Z}_6\times \mathrm{Z}_2`$. The order of this group being twelve corresponds to the fact that each of the large hexagonal cells in figure 2(c) contains exactly twelve of the smaller hexagons.
So in terms of the lattice $``$, the full spatial symmetry of the standing hexagons is given by the group $`\mathrm{\Gamma }^s=\left(\mathrm{Z}_6\times \mathrm{Z}_2\right)\mathrm{D}_6`$, where $`\mathrm{D}_6`$ is generated by a reflection $`\kappa _x`$ and a $`60^{}`$ rotation $`\rho `$ and its standard action on $`\mathrm{}^2`$ is given by
$$\kappa _x:(x,y)(x,y),\rho :(x,y)\frac{1}{2}(x\sqrt{3}y,\sqrt{3}x+y),$$
(12)
and $`\mathrm{Z}_6\times \mathrm{Z}_2`$, an invariant subgroup of $`\mathrm{\Gamma }^s`$, is generated by the two translations
$$\tau _1:(x,y)(x+\frac{\sqrt{3}}{2}c_0,y+\frac{1}{2}c_0),\tau _2:(x,y)(x,y+c_0).$$
(13)
The group $`\mathrm{\Gamma }^s`$ has the semi-direct product structure mentioned in section 2. As a result we can apply the algorithm taken from to calculate all its characters and irreps, and we present the characters of its irreps in table 1.
Any elements that have the same character as the identity in a unitary irrep of $`\mathrm{\Gamma }^s`$ must also act like the identity . Using this simple idea and the information taken from experimental observations about the spatial symmetries of the unstable mode, we can single out the irrep of $`\mathrm{\Gamma }^s`$ that describes the instantaneous symmetry-breaking behaviour. Careful study of snapshots of the superlattice-two pattern shows that it is invariant under the action of $`\tau _2^3`$, $`\kappa _x`$ and $`\tau _1^3\rho ^3`$ (see figures 1(b) and 1(c), where the pattern is shown at a slightly tilted angle, and figure 2(b)). The group generated by these elements is by definition the isotropy subgroup of the bifurcating mode under the action of $`\mathrm{\Gamma }^s`$ (cf (6)), therefore
$$\mathrm{\Sigma }_{𝐳_q^{}}^s\tau _2^3,\kappa _x,\tau _1^3\rho ^3\mathrm{\Gamma }^s.$$
(14)
We can now go through the list of characters of $`\mathrm{\Gamma }^s`$ given in table 1 and determine which irrep satisfies the criteria of permitting $`\mathrm{\Sigma }_{𝐳_q^{}}^s`$ defined in (14) to be an isotropy subgroup of the bifurcating solution. First, any irreps that satisfy
$$\chi _{M_{\gamma _s}}=\chi _{M_{id}}\text{for some}\gamma _s\mathrm{\Gamma }^s\text{and}\gamma _s\mathrm{\Sigma }_{𝐳_q^{}}^s$$
(15)
must be rejected, because in these cases the isotropy subgroup of $`𝐳_q^{}`$ must contain spatial symmetry elements apart from those that are observed. This eliminates representations 1–12. In representation 13, the class containing $`\tau _2^3`$ is represented by the identity, but this class also contains $`\tau _1^3`$ and $`\tau _1\tau _2`$, which are not in $`\mathrm{\Sigma }_{𝐳_q^{}}^s`$, so eliminating this irrep and leaving only 14 and 15. Then we can use the trace formula to calculate the dimension of the fixed-point subspace of $`\mathrm{\Sigma }_{𝐳_q^{}}^s`$:
$$\text{dim}\text{Fix}\left(\mathrm{\Sigma }\right)=\frac{1}{|\mathrm{\Sigma }|}\underset{\sigma \mathrm{\Sigma }}{}\chi _{M_\sigma },$$
which gives 0 for representation 14 and 1 for representation 15. Clearly we require $`\text{dim}\text{Fix}\left(\mathrm{\Sigma }_{𝐳_q^{}}^s\right)0`$, since $`\text{Fix}\left(\mathrm{\Sigma }_{𝐳_q^{}}^s\right)`$ is non-trivial. Thus the six-dimensional irrep $`\mathrm{M}_{\mathrm{\Gamma }^s}^{15}`$ is the only one in which $`\mathrm{\Sigma }_{𝐳_q^{}}^s`$ satisfies the conditions of being an isotropy subgroup of the observed mode.
In addition to being equivariant under the action of spatial symmetries as specified by this irrep, the normal form of the period-doubling bifurcation problem has an extra symmetry corresponding to a translation in time by one period of the external forcing:
$$\tau _t:tt+T.$$
(16)
This element can be viewed as a spatio-temporal symmetry with a trivial spatial action, and it acts independently from elements in $`\mathrm{\Gamma }^s`$ with respect to the standing hexagons. So the full symmetry group $`\mathrm{\Gamma }`$ of the normal form for the superlattice bifurcation problem is a direct product between $`\mathrm{\Gamma }^s`$ and the group $`\tau _t`$, which, as can be seen from (1) and (16), is isomorphic to $`\mathrm{Z}_2`$, hence $`\mathrm{\Gamma }=\mathrm{\Gamma }^s\times \mathrm{Z}_2`$ as we pointed out in section 2. We can write each element $`\gamma \mathrm{\Gamma }`$ as
$$\gamma =(\gamma _s,\sigma _t),\gamma _s\mathrm{\Gamma }^s,\sigma _t\tau _t,$$
(17)
such that for $`\gamma _1=(\gamma _{s_1},\sigma _{t_1})`$, $`\gamma _2=(\gamma _{s_2},\sigma _{t_2})`$, $`\gamma _1\gamma _2=(\gamma _{s_1}\gamma _{s_2},\sigma _{t_1}\sigma _{t_2})`$. Because of the direct product structure of $`\mathrm{\Gamma }`$ and the period-doubling nature of the bifurcating solution, $`\tau _t`$ must act like $`1`$ on the amplitudes of the marginal modes. Therefore the irrep of $`\mathrm{\Gamma }`$ that specifies the action of spatial and spatio-temporal symmetry elements on the marginal modes and the normal form of $`𝐠`$ can be constructed from the set of matrices $`M_{\gamma _s}\mathrm{M}_{\mathrm{\Gamma }^s}^{15}`$ as follows:
$$M_\gamma =\{\begin{array}{cc}M_{\gamma _s}\hfill & \text{if}\sigma _t=\text{identity}\hfill \\ M_{\gamma _s}\hfill & \text{if}\sigma _t=\tau _t\hfill \end{array}$$
for all $`\gamma =(\gamma _s,\sigma _t)\mathrm{\Gamma }`$. This irrep, which we denote by $`\mathrm{M}_\mathrm{\Gamma }`$, is of the same dimension as $`\mathrm{M}_{\mathrm{\Gamma }^s}^{15}`$, which implies that we have a six-dimensional centre manifold at the bifurcation point. So all bifurcating solutions can be written as $`u(𝐱,t)=u_0(𝐱,t)+\zeta (𝐱,t)`$ such that
$$\zeta (𝐱,qT)=A_qf_1(𝐱)+B_qf_2(𝐱)+C_qf_3(𝐱)+\text{c.c.}+\text{h.o.t.},q\mathrm{}$$
(18)
where $`u_0(𝐱,t)`$ represents standing hexagons, c.c. denotes complex conjugate, h.o.t. denotes the higher-order terms, and $`A_q`$, $`B_q`$, $`C_q\mathrm{}`$ are the small amplitudes of $`f_1`$, $`f_2`$ and $`f_3`$, the three complex marginal eigenfunctions that form a basis for the neutral eigenspace (excluding the two zero eigenvalues corresponding to translating the underlying pattern). Note that by including the higher-order terms, $`\zeta `$ represents the nonlinear perturbation from the standing hexagons.
Applying the method described in , we can construct all the $`6\times 6`$ matrices $`M_\gamma `$ that specify the action of $`\mathrm{\Gamma }`$ on $`\mathrm{}^6`$ (or $`\mathrm{}^3`$) for the irrep $`\mathrm{M}_\mathrm{\Gamma }`$. Rather than describe this procedure, we find it convenient to specify the group action by choosing a small number of Fourier modes to represent the marginal eigenfunctions, and working out how the amplitudes of these modes $`A_q`$, $`B_q`$ and $`C_q`$ transform under the generating elements of $`\mathrm{\Gamma }`$. Since representations are defined only up to a similarity transformation, the choice of Fourier modes we make will not matter, as long as we are careful not to introduce any accidental symmetries (which would become apparent on checking the characters).
Any function $`u(𝐱,t)`$ defined on the lattice $``$ can be written as a double Fourier series of the form
$$u(𝐱,t)=\underset{j_1\mathrm{}}{}\underset{j_2\mathrm{}}{}u_{j_1,j_2}(t)e^{2\pi i\left(j_1𝐤_1+j_2𝐤_2\right)𝐱},$$
(19)
where $`𝐤_1`$ and $`𝐤_2`$ are the generating wavevectors of the dual lattice $`^{}`$ related to $`𝐞_1`$ and $`𝐞_2`$ by $`𝐤_i𝐞_j=\delta _{ij}`$ such that (9) holds. Our choice of the vectors $`𝐞_1`$ and $`𝐞_2`$ requires $`𝐤_1=k(\frac{\sqrt{3}}{2},\frac{1}{2})`$, and $`𝐤_2=k(0,1)`$, where $`k=\frac{1}{3c_0}`$.
We use the observed instantaneous symmetry of the pattern (see figure 2(c)) to select a representative function from the full set of Fourier modes, starting with a single Fourier mode $`e^{2\pi i(j_1𝐤_1+j_2𝐤_2)𝐱}`$ (and its complex conjugate) for some choice of integers $`j_1`$ and $`j_2`$. If the pattern is to be invariant under $`\tau _2^3`$, $`j_1`$ must be even, so set $`j_1=2m`$, and, for later convenience, set $`j_2=m+n`$, where $`m`$ and $`n`$ are integers. With this choice, the Fourier mode is $`e^{2\pi ik(\sqrt{3}mx+ny)}`$. The pattern is also invariant under $`\tau _1^3\rho ^3`$. Now $`\rho ^3`$ replaces the chosen mode by its complex conjugate, and $`\tau _1^3`$ multiplies the mode by a complex number with unit modulus. Since $`\tau _1^3`$ is of order two but not equal to the identity, it must act by multiplying the mode by $`1`$. This forces $`m+n`$ to be odd (and so for the observed pattern, the amplitude of the Fourier mode must be pure imaginary). The translation $`\tau _1`$ must act with order 6 (otherwise the pattern would be invariant under a lesser translation in that direction), so $`3m+n1mod6`$ or $`3m+n5mod6`$. The second of these is essentially the complex conjugate of the first, so we choose $`3m+n1mod6`$; $`(m,n)`$ could be $`(0,1)`$, $`(2,1)`$ or $`(1,4)`$, for example. Finally, the reflection $`\kappa _x`$ generates a new function $`e^{2\pi ik(\sqrt{3}mx+ny)}`$, so the superlattice-two pattern can be exemplified by a mode of the form $`f_1=e^{2\pi ik(\sqrt{3}mx+ny)}+e^{2\pi ik(\sqrt{3}mx+ny)}`$. Sixty degree rotations of this function generate $`f_2`$ and $`f_3`$, so we have:
$$f_1=e^{2\pi i𝐊_1𝐱}+e^{2\pi i𝐊_2𝐱},f_2=e^{2\pi i𝐊_3𝐱}+e^{2\pi i𝐊_4𝐱},f_3=e^{2\pi i𝐊_5𝐱}+e^{2\pi i𝐊_6𝐱}$$
(20)
(the true eigenfunctions will be made up of linear combinations of such functions), where
| $`𝐊_1=k(\sqrt{3}m,n)`$, | $`𝐊_2=k(\sqrt{3}m,n)`$, |
| --- | --- |
| $`𝐊_3=\frac{k}{2}(\sqrt{3}(m+n),(3m+n))`$, | $`𝐊_4=\frac{k}{2}(\sqrt{3}(m+n),(3m+n))`$, |
| $`𝐊_5=\frac{k}{2}(\sqrt{3}(m+n),(3mn))`$, | $`𝐊_6=\frac{k}{2}(\sqrt{3}(m+n),(3m+n))`$. |
These wavevectors have the same wavenumber $`K(m,n)=k\sqrt{3m^2+n^2}`$, with $`m`$ and $`n`$ satisfying $`3m+n1mod6`$. With this choice of basis functions, the relevant irrep of $`\mathrm{\Gamma }`$ can be specified by the action of the generating elements of $`\mathrm{\Gamma }`$ on the amplitudes $`(A_q,B_q,C_q)`$:
$`\kappa _x:(A_q,B_q,C_q)`$ $``$ $`(A_q,\overline{C_q},\overline{B_q}),`$ (21)
$`\rho :(A_q,B_q,C_q)`$ $``$ $`(B_q,C_q,\overline{A_q}),`$ (22)
$`\tau _1:(A_q,B_q,C_q)`$ $``$ $`(e^{\frac{i\pi }{3}}A_q,e^{\frac{i2\pi }{3}}B_q,e^{\frac{i\pi }{3}}C_q),`$ (23)
$`\tau _2:(A_q,B_q,C_q)`$ $``$ $`(e^{\frac{i2\pi }{3}}A_q,e^{\frac{i\pi }{3}}B_q,e^{\frac{i\pi }{3}}C_q),`$ (24)
$`\tau _t:(A_q,B_q,C_q)`$ $``$ $`(A_q,B_q,C_q).`$ (25)
We include the subharmonic action of $`\tau _t`$ here for completeness. The same representation could be constructed using the method described in .
## 4 Normal form of the bifurcation problem
We now have sufficient information to invoke the equivariant branching lemma and describe the different patterns that must be formed in the instability that created the superlattice-two from standing hexagons. Before doing this, we will compute the normal form for the bifurcation since we need it to work out the stability of the various patterns. The irrep (2125) we identified in section 3 implies that the reduced map $`𝐠`$ introduced in (2) is six-dimensional, and we let $`𝐳_q=(A_q,B_q,C_q)`$, $`q\mathrm{}`$, $`A_q,B_q,C_q\mathrm{}`$. As indicated earlier, the action of $`\tau _t`$ defined in (16) is due to the subharmonic nature of the bifurcating modes with respect to the overall driving period $`T`$ given in (1). If each iteration in $`𝐳_q`$ corresponds to advancing in time by $`T`$, then $`𝐳_{q+2}=𝐳_{q+1}=𝐳_q`$. Consequently, $`𝐠(𝐳_q)=𝐳_{q+1}=𝐳_{q+2}=𝐠(𝐳_{q+1})=𝐠(𝐳_q)`$ . So the map $`𝐠`$ will be an odd function of the amplitudes $`A_q`$, $`B_q`$ and $`C_q`$, as well as being $`\mathrm{\Gamma }`$-equivariant. This information enables us to write down the form of $`𝐠`$ including up to fifth order terms:
$`A_{q+1}`$ $`=`$ $`\left(1+\mu \right)A_q+\alpha _1|A_q|^2A_q+\alpha _2\left(|B_q|^2+|C_q|^2\right)A_q+\beta _1|A_q|^4A_q`$ (26)
$`+\text{ }\beta _2\left(|B_q|^4+|C_q|^4\right)A_q+\beta _3|A_q|^2\left(|B_q|^2+|C_q|^2\right)A_q+\beta _4|B_q|^2|C_q|^2A_q`$
$`+\text{ }\beta _5B_q^2\overline{C}_q^2\overline{A_q}+\nu \overline{A}_q^5,`$
$`B_{q+1}`$ $`=`$ $`\left(1+\mu \right)B_q+\alpha _1|B_q|^2B_q+\alpha _2\left(|A_q|^2+|C_q|^2\right)B_q+\beta _1|B_q|^4B_q`$ (27)
$`+\text{ }\beta _2\left(|A_q|^4+|C_q|^4\right)B_q+\beta _3|B_q|^2\left(|A_q|^2+|C_q|^2\right)B_q+\beta _4|A_q|^2|C_q|^2B_q`$
$`+\text{ }\beta _5A_q^2C_q^2\overline{B_q}+\nu \overline{B}_q^5,`$
$`C_{q+1}`$ $`=`$ $`\left(1+\mu \right)C_q+\alpha _1|C_q|^2C_q+\alpha _2\left(|A_q|^2+|B_q|^2\right)C_q+\beta _1|C_q|^4C_q`$ (28)
$`+\text{ }\beta _2\left(|A_q|^4+|B_q|^4\right)C_q+\beta _3|C_q|^2\left(|A_q|^2+|B_q|^2\right)C_q+\beta _4|A_q|^2|B_q|^2C_q`$
$`+\text{ }\beta _5\overline{A}_q^2B_q^2\overline{C_q}+\nu \overline{C}_q^5,`$
where all coefficients are forced by symmetry to be real.
Apart from the $`\nu `$ terms, the equations above are equivalent to the $`\mathrm{T}^2\mathrm{D}_6\times \mathrm{Z}_2`$-equivariant amplitude equations (truncated to the same order) that arise in the context of Boussinesq convection on a hexagonal lattice , once they are re-interpreted as amplitude equations rather than a map. The $`\nu `$ terms have the effect of breaking the full $`\mathrm{T}^2`$ (two-torus) symmetry group of translations in a periodic domain to the discrete translations allowed by the underlying pattern. A natural question to ask is why we needed to work out the details of the representation before writing down these amplitude equations. The main reason is that we did not know in advance how many linearly independent marginal eigenfunctions are involved in the instability. Even if we had assumed that there were six, it has turned out that there are two six-dimensional irreps, only one of which is involved in the bifurcation. The other six-dimensional irrep is generated by taking $`f_1=e^{2\pi ik(\sqrt{3}mx+ny)}e^{2\pi ik(\sqrt{3}mx+ny)}`$ and (following a similar analysis) results in the same amplitude equation. Without realising this, one might conclude incorrectly that patterns that are odd under $`\kappa _x`$ reflection might also be found in this instability. All the other irreps in table 1 have dimension less than six (that is, there are fewer than six independent marginal eigenfunctions), so the order of the relevant normal forms would be correspondingly less.
We also use (5) to write down the dynamics of the position $`𝐝_q`$ of the underlying standing hexagons, truncated to quartic order:
$$𝐝_{q+1}=𝐝_q+\xi \text{Im}\left[\begin{array}{c}A_q^2(\overline{C}_q^2B_q^2)2B_q^2C_q^2\\ \sqrt{3}A_q^2(B_q^2+\overline{C_q}^2)\end{array}\right],$$
(29)
where $`\xi `$ is a constant.
We can show that there are six isotropy subgroups whose fixed-point subspaces are one-dimensional, so the equivariant branching lemma tells us that there are at least six primary bifurcating branches of solutions from standing hexagons, and we summarise these solutions in table 2. Elements accented by a tilde represent spatio-temporal symmetries, which, using the notation introduced in (17), can be written as $`\stackrel{~}{\tau }_{1}^{}{}_{}{}^{3}=(\tau _1^3,\tau _t)`$, and similarly for $`\stackrel{~}{\tau }_{2}^{}{}_{}{}^{3}`$ and $`\stackrel{~}{\rho }`$. The superlattice-two pattern corresponds to branch 2 of type I.
For the choice of wave integer pair $`(m,n)=(2,1)`$, the instantaneous planforms of these six solution branches are illustrated schematically in figure 3. We can compare figures 3(e) and 3(f) with 1(b) and 1(c) and notice that the appearance of stripes at regular intervals in the grey-scale plots of solution branch 2 closely resembles the essential features of the experimentally observed superlattice-two pattern.
None of these primary branches leads to a net drift of the underlying hexagonal pattern; this can be seen in two ways: first, because the rate of drift (from (29), truncated to quartic order) is zero on all six primary branches; second (and more convincing) since $`\rho ^3`$ is in the symmetry group of all the time averaged patterns (see below). In other words, the patterns are all pinned by the $`180^{}`$ rotation symmetry on average.
### 4.1 Stability results
We summarise in table 3 the branching equations and the Floquet multipliers of the period $`2T`$ patterns, for each of the six primary solutions guaranteed to exist by the equivariant branching lemma. Floquet multipliers greater than one in magnitude indicate instability. We group the six branches into three types and denote them by I, II and III as shown in tables 2 and 3. It is evident that branches within each type are degenerate up to third-order terms, thus necessitating the inclusion of quintic terms. In particular, only one solution branch within each of types I and II can be stable depending on the signs of $`\nu `$ and $`\beta _5+\nu `$, and both branches in type III are always unstable. Only one branch can bifurcate stably, and all branches must be supercritical for one of them to be stable. One of the requirements for the observed superlattice-two pattern (i.e., branch 2 of type I) to be stable is that the quintic coefficient $`\nu >0`$. If we also assume the non-degeneracy conditions $`\alpha _10`$, $`\alpha _1+2\alpha _20`$, $`\alpha _1\pm \alpha _20`$, $`\nu 0`$ and $`\beta _5+\nu 0`$, close to the bifurcation point the relative stability of branches of the three types is illustrated by the bifurcation diagrams shown in figure 4.
The experimental results suggest that the bifurcation may have subcritical branches as there is a parameter regime in which standing hexagons and the superlattice-two pattern may coexist. On the other hand, the experimentalists report no hysteresis between standing hexagons and superlattice-two, while they do report hysteresis between hexagons and other patterns at other parameter values, so it is not clear whether or not there is a direct bifurcation from standing hexagons to the superlattice-two pattern in the experiments. With our parameters, we require $`\alpha _1>0`$, $`\alpha _2>\alpha _1`$ and $`\nu >0`$ for the superlattice-two pattern (branch 2 of type I) to bifurcate stably, but the branch could also be stable in the region $`\alpha _1<0`$, $`\alpha _2>\alpha _1`$ and $`\nu >0`$ if there were a saddle-node bifurcation on branch I.
### 4.2 Time-averaged behaviour
We can study the symmetry properties of the time-averaged image of the observed solution by integrating over a full period of the newly created periodic orbit (cf ). Specifically, we let $`u_0(𝐱,t)`$ be the standing hexagons solution and $`\zeta (𝐱,t)`$ the nonlinear perturbation to this solution such that $`u(𝐱,t)=u_0(𝐱,t)+\zeta (𝐱,t)`$ represents the observed pattern. We know that $`u_0(𝐱,t)`$ is $`\mathrm{\Gamma }`$-invariant, i.e., $`\gamma u_0(𝐱,t)=u_0(𝐱,t)`$ for all $`\gamma \mathrm{\Gamma }`$, and have also found that the spatial and spatio-temporal symmetry of $`\zeta (𝐱,t)`$ is given by its isotropy subgroup $`\tau _2^3,\kappa _x,\tau _1^3\rho ^3,\stackrel{~}{\tau }_{1}^{}{}_{}{}^{3}\mathrm{\Sigma }_\zeta \mathrm{\Gamma }`$. Let $`\gamma _s`$ and $`\gamma _t`$ denote respectively the purely spatial symmetry elements and the spatial part of spatio-temporal symmetry elements in $`\mathrm{\Sigma }_\zeta `$ that act on $`\zeta (𝐱,t)`$ as follows:
$$\gamma _s\zeta (𝐱,t)=\zeta (𝐱,t),\gamma _t\zeta (𝐱,t)=\zeta (𝐱,t+T).$$
The time-averaged value of $`u(𝐱,t)`$ can be obtained by integrating over the full period of the bifurcating solution:
$`\overline{u}(𝐱)`$ $`=`$ $`{\displaystyle \frac{1}{2T}}{\displaystyle _0^{2T}}u_0(𝐱,t)+\zeta (𝐱,t)\mathrm{d}t`$ (30)
$`=`$ $`\overline{u}_0(𝐱)+{\displaystyle \frac{1}{2T}}{\displaystyle _0^T}\zeta (𝐱,t)+\zeta (𝐱,t+T)\mathrm{d}t,`$
where we have used the fact that $`u_0(𝐱,t)=u_0(𝐱,t+T)`$. Clearly, $`\overline{u}`$ shares the same spatial symmetry with $`\zeta `$ because both $`\overline{u}_0`$ and the individual entries of the integrand in (30) are invariant under the action of $`\gamma _s\mathrm{\Sigma }_\zeta `$. It is also invariant under $`\gamma _t`$ because the integrand in (30) as a whole is invariant under $`\gamma _t`$:
$`\gamma _t\overline{u}(𝐱)`$ $`=`$ $`\gamma _t\overline{u}_0(𝐱)+{\displaystyle \frac{1}{2T}}{\displaystyle _0^T}\gamma _t\zeta (𝐱,t)+\gamma _t\zeta (𝐱,t+T)\mathrm{d}t`$
$`=`$ $`\overline{u}_0(𝐱)+{\displaystyle \frac{1}{2T}}{\displaystyle _0^T}\zeta (𝐱,t+T)+\zeta (𝐱,t)\mathrm{d}t`$
$`=`$ $`\overline{u}(𝐱).`$
This result in fact follows readily from more general results on the symmetries of chaotic attractors .
In the case of the observed superlattice-two pattern with isotropy subgroup $`\mathrm{\Sigma }_\zeta `$, the spatial component of the spatio-temporal symmetry element, namely $`\tau _1^3`$, will show up alongside $`\tau _2^3`$, $`\kappa _x`$ and $`\tau _1^3\rho ^3`$ in the time-averaged image to generate an augmented spatial symmetry group $`\mathrm{\Sigma }_{\overline{u}}=\tau _1^3,\tau _2^3,\kappa _x,\rho ^3`$. This prediction is in agreement with experimental results (figure 1(a)) and can be understood in the following way. The action of the translations $`\tau _1`$ and $`\tau _2`$ on $`\overline{u}(𝐱)`$ is of order three since $`\tau _1^3\overline{u}(𝐱)=\tau _2^3\overline{u}(𝐱)=\overline{u}(𝐱)`$, whereas the order of the same action on $`u(𝐱)`$ is six. So the averaged pattern will appear to be periodic on a lattice $`_{\mathrm{av}}`$ spanned by basis vectors $`𝐞_{\mathrm{av}}`$ such that $`|𝐞_{\mathrm{av}}|=\frac{1}{2}|𝐞_i|=\frac{1}{2}c`$. We have shown in section 3 that $`c_0=\frac{c}{2\sqrt{3}}`$, it follows that $`|𝐞_{\mathrm{av}}|=\sqrt{3}|\stackrel{~}{𝐞}_i|`$. Therefore the ratio of spatial period of the averaged pattern to that of the basic standing hexagons is $`1:\sqrt{3}`$, which is consistent with the observation reported by as shown in figure 1(a). Using the same reasoning and information from the isotropy subgroups of the primary solutions given in table 2, we therefore predict both branches in each type of solutions to have the same time-averaged symmetries.
## 5 Discussion
Starting from the observed instantaneous symmetry of the superlattice-two pattern reported in , we have been able to show (a) that a pattern with the same instantaneous spatial symmetry as the superlattice-two pattern can bifurcate stably from standing hexagons in a spatial period-multiplying instability; (b) that the pattern has the spatio-temporal symmetry (not reported in ) of advancing one driving period in time combined with a translation by three units in space (figure 3(e) and 3(f)); and (c) that this spatio-temporal symmetry accounts for the intermediate spatial scale and periodicity on a hexagonal lattice of the time-averaged pattern (figure 1(a)). We should emphasise that the intermediate spatial periodicity of the time-averaged pattern is not the spatial periodicity of the larger hexagonal lattice that we have assumed.
Arbell & Fineberg (unpublished) have found the superlattice-two state in their experiments and have confirmed that it does have the spatio-temporal symmetry that we predict. Our results also suggest that $`60^{}`$ rotations are not in fact symmetries of the time-averaged pattern, but should be weakly broken. The breaking of $`60^{}`$ rotational symmetry, if it is present, is evidently a small effect since the hexagons in figure 1(a) do appear to be invariant under $`60^{}`$ rotations (this has been confirmed by Gollub, private communication). For other parameter values, the symmetry breaking effect may be more pronounced: Fineberg (private communication) reports that his experimental time-averaged pattern is not invariant under $`60^{}`$ rotations. Clearly this would be an interesting issue to investigate in more detail, but the measurements are delicate and are liable to be prone to systematic errors or imperfections, so confirming our prediction could be difficult.
The spatio-temporal symmetry of superlattice-two arises because the instability of standing hexagons is subharmonic. Other patterns, with different combinations of spatial and spatio-temporal symmetries, are possible stable branches in the same bifurcation problem. Not all branches of solutions have spatio-temporal symmetries, and some of the patterns share the same time-averaged symmetries even though they have different instantaneous planforms. The method we have presented is based entirely on symmetry arguments and is able to deal with instabilities of a fully nonlinear time-periodic solution.
Spatial period-multiplying instabilities have arisen in a variety of contexts, in both one and two lateral directions . Most of these situations involved relatively simple groups; part of the difficulty and interest here has been the size of the symmetry group, enlarged because of the number of translations broken by the new pattern. Only one of the 15 representations is involved in the superlattice-two bifurcation; other representations may be relevant to other experiments (particularly ) in which standing hexagons lose stability to patterns that fit into the larger hexagonal cells we have used here.
As can be seen in section 3, a heuristic step in our method involves the choice of a suitable periodic cell that accommodates the observed patterns and whose size coincides with exactly one spatial period of the bifurcating modes. The arrangement of the underlying basic state in this cell then defines a spatial symmetry group $`\mathrm{\Gamma }^s`$ of the bifurcation problem and the instantaneous symmetries of the superlattice instability form its isotropy subgroup $`\mathrm{\Sigma }^s`$. If a larger hexagonal periodic cell that captures more than one spatial period of the bifurcating modes had been chosen, the translations $`\tau _1`$ and $`\tau _2`$ given in (10) that leave standing hexagons invariant would have had higher order, resulting in a larger spatial symmetry group. In this case there would have been more than one irrep of $`\mathrm{\Gamma }^s`$ in which $`\mathrm{\Sigma }^s`$ satisfied the conditions of being an isotropy subgroup. By choosing the smallest possible periodic cell, we have found that such indeterminacy can be avoided.
The method we have described in this paper for analysing certain types of symmetry-breaking instabilities bifurcating from a non-trivial basic state is based entirely on the observed spatial symmetries of these patterns. However, information on spatial symmetries of the new pattern alone may not be sufficient for our approach to be applicable in some problems. For example, consider a bifurcation problem defined on a spherical domain. Suppose a basic state with $`\mathrm{O}(3)`$ symmetry loses stability and the observed bifurcating solutions are axisymmetric, then the isotropy subgroup of the bifurcating modes is given by $`\mathrm{O}(2)`$. If the eigenfunctions are expanded in spherical harmonics, it is known that $`\mathrm{O}(2)`$ is a maximal isotropy subgroup of $`\mathrm{O}(3)`$ for all even values of the spherical harmonic index $`l`$ and so an infinite number of irreps is relevant to the observed bifurcation. This example illustrates the fact that our method breaks down if the observed symmetries of the bifurcating modes form an isotropy subgroup for more than one irrep of the symmetry group of the basic state.
We are currently involved in applying a similar method to the study of the ‘superlattice-one’ pattern reported in as a bifurcating instability from standing hexagons. Preliminary analysis of the experimental data reveals that a suitable periodic box in this case will give rise to an arrangement of standing hexagons with a ‘hidden’ reflection symmetry , which leads to extra complications in determining the spatial symmetry group. It is an interesting problem that deserves further investigation.
Unlike some time-periodic solutions (for example, standing rolls), which can also be defined on a hexagonal lattice, standing hexagons possess only trivial spatio-temporal symmetries . So our treatment of the superlattice patterns as symmetry-breaking instabilities from standing hexagons is relatively simple because only instantaneous spatial symmetries are needed to define the isotropy subgroup of these solutions. In general, our approach can be applied to the study of spatial period-multiplying bifurcations from solutions with spatio-temporal symmetries and used to investigate some of the possible symmetry-breaking behaviour, if techniques discussed by Rucklidge & Silber and Lamb & Melbourne are also included.
We would like to thank Jerry Gollub for inspiring this work, Jonathan Dawes, Marty Golubitsky, Paul Matthews and Michael Proctor for sharing their insights with us, and Jay Fineberg for showing us his unpublished experimental results. DPT is grateful to the Croucher Foundation for financial support. The research of AMR is supported by EPSRC. The research of RBH was supported by King’s College, Cambridge. The research of MS is supported by NSF grant DMS-9972059, NSF CAREER award DMS-9502266, and by NASA grant NAG3-2364.
|
warning/0005/math0005291.html
|
ar5iv
|
text
|
# Untitled Document
Homotopy field theory in dimension 3 and crossed group-categories
Vladimir Turaev
Abstract
A $`3`$-dimensional homotopy quantum field theory (HQFT) can be described as a TQFT for surfaces and $`3`$-cobordisms endowed with homotopy classes of maps into a given space. For a group $`\pi `$, we introduce a notion of a modular crossed $`\pi `$-category and show that such a category gives rise to a 3-dimensional HQFT with target space $`K(\pi ,1)`$. This includes numerical invariants of 3-dimensional $`\pi `$-manifolds and a 2-dimensional homotopy modular functor. We also introduce and discuss a parallel notion of a quasitriangular crossed Hopf $`\pi `$-coalgebra.
Contents
Introduction
1. Group-categories
2. Crossed, braided, and ribbon $`\pi `$-categories
3. Colored $`\pi `$-tangles and their invariants
4. Colored $`\pi `$-graphs and their invariants
5. Trace, dimension, and algebra of colors
6. Modular crossed $`\pi `$-categories
7. Invariants of 3-dimensional $`\pi `$-manifolds
8. A 2-dimensional homotopy modular functor
9. A 2-dimensional HQFT
10. A 3-dimensional HQFT
11. Hopf group-coalgebras
12. Canonical extensions
13. Transfer of categories
Appendix 1. Quasi-abelian cohomology of groups
Appendix 2. State sum invariants of 3-dimensional $`\pi `$-manifolds
Appendix 3. Open problems
Introduction
A homotopy quantum field theory (HQFT) is a version of a topological quantum field theory (TQFT) for manifolds endowed with maps into a fixed topological space. The HQFT’s were introduced in \[Tu3\] where the author gave an algebraic characterization of 2-dimensional HQFT’s whose target space is the Eilenberg-MacLane space $`K(\pi ,1)`$ determined by a group $`\pi `$. In this paper we focus on 3-dimensional HQFT’s with target space $`K(\pi ,1)`$. A manifold $`M`$ endowed with a homotopy class of maps $`MK(\pi ,1)`$ is called a $`\pi `$-manifold. The homotopy classes of maps $`MK(\pi ,1)`$ classify principal $`\pi `$-bundles over $`M`$ and (for connected $`M`$) bijectively correspond to the homomorphisms $`\pi _1(M)\pi `$. A 3-dimensional HQFT with target space $`K(\pi ,1)`$ comprises two ingredients: a homotopy modular functor assigning $`K`$-modules to $`\pi `$-surfaces and an invariant of 3-dimensional $`\pi `$-manifolds taking value in the module associated with the boundary. In particular, the HQFT provides numerical invariants of closed 3-dimensional $`\pi `$-manifolds. Our main aim is to introduce an algebraic technique allowing to construct 3-dimensional HQFT’s.
Our approach is based on a deep connection between the theory of braided categories and invariants of knots, links and 3-manifolds. This connection has been essential in the construction of “quantum” invariants of knots and 3-manifolds from quantum groups, see \[RT\], \[Tu2\], \[KRT\]. Here we extend these ideas to links $`\mathrm{}S^3`$ endowed with homomorphisms $`\pi _1(S^3\backslash \mathrm{})\pi `$ and to 3-dimensional $`\pi `$-manifolds. To this end we introduce a notion of a crossed $`\pi `$-category and study braidings and twists in such categories. This leads us to the notion of a modular crossed $`\pi `$-category. We show that each modular crossed $`\pi `$-category gives rise to a 3-dimensional HQFT with target $`K(\pi ,1)`$. In the case $`\pi =1`$ we recover the usual construction of 3-dimensional TQFT’s from modular categories, see \[Tu2\].
The crossed $`\pi `$-categories are quite delicate algebraic objects. We discuss a few general methods producing such categories. In particular we introduce quasitriangular Hopf $`\pi `$-coalgebras and show that they give rise to crossed $`\pi `$-categories. Other methods are based on a study of self-equivalences of a braided category, a study of quasi-abelian cohomology of $`\pi `$, and a transfer-type construction. This gives several examples of modular crossed $`\pi `$-categories. However, the problem of systematic finding of modular crossed $`\pi `$-categories is largely open. It would be most interesting to extend the quantum groups associated with semisimple finite dimensional complex Lie algebras to quasitriangular Hopf $`\pi `$-coalgebras.
The content of the paper is as follows. In Sections 1 and 2 we introduce crossed $`\pi `$-categories and various additional structures on them (braiding, twist etc.). In Sections 3 and 4 we introduce $`\pi `$-links, $`\pi `$-tangles and $`\pi `$-graphs in $`^3`$ colored over a ribbon crossed $`\pi `$-category $`𝒞`$. We also define their canonical functorial invariant taking values in $`𝒞`$. In Section 5 we study traces of morphisms in $`𝒞`$. In Section 6 we introduce modular crossed $`\pi `$-categories. They are used in Section 7 to define invariants of 3-dimensional $`\pi `$-manifolds. In Sections 8-10 we introduce the 2-dimensional and 3-dimensional HQFT’s derived from a modular crossed $`\pi `$-category. In Sections 11-13 we discuss algebraic constructions of crossed $`\pi `$-categories.
The paper is ended with three appendices. In Appendix 1 we briefly discuss quasi-abelian 3-cohomology of groups and their relations to crossed categories. In Appendix 2 we outline a state sum approach to invariants of 3-dimensional $`\pi `$-manifolds. In Appendix 3 we discuss a few open problems.
Throughout the paper, the symbol $`K`$ denotes a commutative ring with unit. The symbol $`\pi `$ denotes a group. By a category we mean a small category.
We shall work in the smooth set up although all our definitions have topological and piecewise-linear versions. Thus, by manifolds we mean smooth manifolds and by homeomorphisms we mean smooth homeomorphisms.
1. Group-categories
1.1. Generalities on monoidal categories. Let $`𝒞`$ be a monoidal category with unit object 1 I. Recall (see for instance \[Ma\]) that we have invertible associativity morphisms
$$\{a_{U,V,W}:(UV)WU(VW)\}_{U,V,W𝒞}$$
$`(1.1.a)`$
and invertible morphisms
$$\{l_U:UU\text{1}\text{ I},r_U:U\text{1}\text{ I}U\}_{U𝒞}$$
$`(1.1.b)`$
satisfying the pentagon identity
$$(\text{id}_Ua_{V,W,X})a_{U,VW,X}(a_{U,V,W}\text{id}_X)=a_{U,V,WX}a_{UV,W,X}$$
$`(1.1.c)`$
and the triangle identities
$$a_{U,\text{1}\text{ I},V}(l_U\text{id}_V)=\text{id}_Ur_V$$
$`(1.1.d)`$
for any $`U,V,W,X𝒞`$, where the inclusion $`U𝒞`$ means that $`U`$ is an object of $`𝒞`$. The morphisms $`l,r`$ should satisfy $`l_{\text{1}\text{ I}}=r_{\text{1}\text{ I}}`$ and be natural in the sense that for any morphism $`f:UV`$ we have $`l_Vf=(f\text{id}_{\text{1}\text{ I}})l_U,r_Vf=(\text{id}_{\text{1}\text{ I}}f)r_U`$. The associativity morphisms (1.1.a) should be natural in a similar sense.
A left duality in $`𝒞`$ associates to any object $`U𝒞`$ an object $`U^{}𝒞`$ and two morphisms
$$b_U:\text{1}\text{ I}UU^{},d_U:U^{}U\text{1}\text{ I}$$
$`(1.1.e)`$
such that
$$(l_U)^1(\text{id}_Ud_U)a_{U,U^{},U}(b_U\text{id}_U)r_U=\text{id}_U,$$
$`(1.1.f)`$
$$(r_U^{})^1(d_U\text{id}_U^{})(a_{U^{},U,U^{}})^1(\text{id}_U^{}b_U)l_U^{}=\text{id}_U^{}.$$
$`(1.1.g)`$
We call the morphisms (1.1.a), (1.1.b), (1.1.e) the structural morphisms of $`𝒞`$.
A monoidal category $`𝒞`$ is strict if $`(UV)W=U(VW)`$, $`U=U\text{1}\text{ I}=\text{1}\text{ I}U`$ for any $`U,V,W𝒞`$ and the morphisms $`\{a_{U,V,W}\}_{U,V,W𝒞}`$ and $`\{l_U,r_U\}_{U𝒞}`$ are the identity morphisms. It is well-known that each monoidal category is equivalent to a strict monoidal category in a canonical way.
1.2. $`\pi `$-categories. A monoidal category $`𝒞`$ is said to be $`K`$-additive if the $`\text{Hom}^{}s`$ in $`𝒞`$ are $`K`$-modules and both the composition and the tensor product of morphisms are bilinear over $`K`$. We say that a $`K`$-additive category $`𝒞`$ splits as a disjoint union of subcategories $`\{𝒞_\alpha \}`$ numerated by certain $`\alpha `$ if:
\- each $`𝒞_\alpha `$ is a full subcategory of $`𝒞`$;
\- each object of $`𝒞`$ belongs to $`𝒞_\alpha `$ for a unique $`\alpha `$;
\- if $`U𝒞_\alpha `$ and $`V𝒞_\beta `$ with $`\alpha \beta `$ then $`\text{Hom}_𝒞(U,V)=0`$.
For a group $`\pi `$, a $`\pi `$-category over $`K`$ is a $`K`$-additive monoidal category with left duality $`𝒞`$ which splits as a disjoint union of subcategories $`\{𝒞_\alpha \}`$ numerated by $`\alpha \pi `$ such that
(i) if $`U𝒞_\alpha `$ and $`V𝒞_\beta `$ then $`UV𝒞_{\alpha \beta }`$;
(ii) if $`U𝒞_\alpha `$ then $`U^{}𝒞_{\alpha ^1}`$.
We shall write $`𝒞=_\alpha 𝒞_\alpha `$ and call the subcategories $`\{𝒞_\alpha \}`$ of $`𝒞`$ the components of $`𝒞`$. The category $`𝒞_1`$ corresponding to the neutral element $`1\pi `$ is called the neutral component of $`𝒞`$. Conditions (i) and (ii) show that $`𝒞_1`$ is closed under tensor multiplication and taking the dual object. Condition (i) implies the inclusion $`\text{1}\text{ I}𝒞_1`$. Thus, $`𝒞_1`$ is a $`K`$-additive monoidal category with left duality.
1.3. Example: $`\pi `$-categories from 3-cocycles. It is well-known that 3-cocycles give rise to associativity morphisms in categories. Here we elaborate this construction in our context. Let $`a=\{a_{\alpha ,\beta ,\gamma }K^{}\}_{\alpha ,\beta ,\gamma \pi }`$ be a 3-cocycle of the group $`\pi `$ with values in the multiplicative group $`K^{}`$ consisting of invertible elements of $`K`$. Thus
$$a_{\alpha \beta ,\gamma ,\delta }a_{\alpha ,\beta ,\gamma \delta }=a_{\alpha ,\beta ,\gamma }a_{\alpha ,\beta \gamma ,\delta }a_{\beta ,\gamma ,\delta }$$
$`(1.3.a)`$
for any $`\alpha ,\beta ,\gamma ,\delta \pi `$. Let $`b=\{b_\alpha K^{}\}_{\alpha \pi }`$ be a family of elements of $`K^{}`$ numerated by $`\pi `$. With the pair $`(a,b)`$ we associate a $`\pi `$-category $`𝒞`$ as follows.
For $`\alpha \pi `$, we define $`𝒞_\alpha `$ to be a category with one object $`V_\alpha `$. For $`\alpha ,\beta \pi `$, set
$$\text{Hom}(V_\alpha ,V_\beta )=\{\begin{array}{cc}K,if\alpha =\beta ,\hfill & \\ 0,if\alpha \beta .\hfill & \end{array}$$
The tensor product is given by $`V_\alpha V_\beta =V_{\alpha \beta }`$. The composition and tensor product of morphisms is given by multiplication in $`K`$. Clearly, $`\text{1}\text{ I}=V_1`$ is the unit object of $`𝒞`$. Now we define the structural morphisms in $`𝒞`$. The associativity morphism (1.1.a) for $`U=V_\alpha ,V=V_\beta ,W=V_\gamma `$ is by definition $`a_{\alpha ,\beta ,\gamma }K^{}K=\text{End}(V_{\alpha \beta \gamma })`$. The pentagon equation follows from (1.3.a). The morphisms $`l_\alpha :V_\alpha V_\alpha \text{1}\text{ I}=V_\alpha `$ and $`r_\alpha :V_\alpha \text{1}\text{ I}V_\alpha =V_\alpha `$ are defined by
$$l_\alpha =(a_{\alpha ,1,1})^1K^{}K=\text{End}(V_\alpha ),r_\alpha =a_{1,1,\alpha }K^{}K=\text{End}(V_\alpha ).$$
The triangle identity follows from the equality $`a_{\alpha ,1,\beta }=a_{\alpha ,1,1}a_{1,1,\beta }`$ obtained from (1.3.a) by the substitution $`\beta =1,\gamma =1,\delta =\beta `$. The dual of $`V_\alpha `$ is by definition $`V_{\alpha ^1}`$. The duality morphism $`\text{1}\text{ I}V_\alpha V_{\alpha ^1}=\text{1}\text{ I}`$ is defined to be $`b_\alpha K^{}K=\text{End}(\text{1}\text{ I})`$. The morphism $`d_\alpha :V_{\alpha ^1}V_\alpha \text{1}\text{ I}`$ is determined uniquely from (1.1.f) and (1.1.g). In fact these equations give two different expressions for $`d_\alpha `$:
$$d_\alpha =(b_\alpha )^1(a_{\alpha ,\alpha ^1,\alpha }a_{\alpha ,1,1}a_{1,1,\alpha })^1,$$
$`(1.3.b)`$
$$d_\alpha =(b_\alpha )^1a_{\alpha ^1,\alpha ,\alpha ^1}a_{\alpha ^1,1,1}a_{1,1,\alpha ^1}.$$
$`(1.3.c)`$
To show that the right-hand sides are equal we substitute $`\beta =\delta =\alpha ^1,\gamma =\alpha `$ in (1.3.a). This gives
$$a_{1,\alpha ,\alpha ^1}a_{\alpha ,\alpha ^1,1}=a_{\alpha ,\alpha ^1,\alpha }a_{\alpha ,1,\alpha ^1}a_{\alpha ^1,\alpha ,\alpha ^1}$$
$$=a_{\alpha ,\alpha ^1,\alpha }a_{\alpha ,1,1}a_{1,1,\alpha ^1}a_{\alpha ^1,\alpha ,\alpha ^1}.$$
Substituting $`\alpha =\gamma =1`$ in (1.3.a) we obtain that $`a_{1,\alpha ,1}=1`$ for all $`\alpha `$. Substituting $`\gamma =\delta =1,\beta =\alpha ^1`$ (resp. $`\alpha =1,\delta =\beta =\gamma ^1`$) in (1.3.a) we obtain that $`a_{\alpha ,\alpha ^1,1}=(a_{\alpha ^1,1,1})^1`$ (resp. $`a_{1,\alpha ,\alpha ^1}=(a_{1,1,\alpha })^1`$). Combining these equalities we obtain that the right-hand sides of (1.3.b) and (1.3.c) coincide. It is clear that $`𝒞`$ is a $`\pi `$-category.
1.4. Operations on group-categories. We define here a few elementary operations on group-categories. Note first that the group-categories can be pulled back along group homomorphisms. Having a group homomorphism $`q:\pi ^{}\pi `$ we can transform any $`\pi `$-category $`𝒞`$ into a $`\pi ^{}`$-category $`𝒞^{}=q^{}(𝒞)`$ defined by $`𝒞_\alpha ^{}=𝒞_{q(\alpha )}`$ for any $`\alpha \pi ^{}`$. Composition, tensor multiplication, and the structural morphisms in $`𝒞`$ induce the corresponding operations in $`𝒞^{}`$ in the obvious way.
The group-categories can be pushed forward along group homomorphisms. Having a group homomorphism $`q:\pi ^{}\pi `$ we can transform any $`\pi ^{}`$-category $`𝒞^{}`$ into a $`\pi `$-category $`𝒞=q_{}(𝒞^{})`$ defined by $`𝒞_\alpha =_{\beta q^1(\alpha )}𝒞_\beta ^{}`$ for any $`\alpha \pi `$. Composition, tensor multiplication, and the structural morphisms in $`𝒞^{}`$ induce the corresponding operations in $`𝒞`$ in the obvious way.
For any family of $`\pi `$-categories $`\{𝒞^i\}_{iI}`$, we define a direct product $`𝒞=_i𝒞^i`$. The category $`𝒞`$ is a disjoint union of categories $`\{𝒞_\alpha \}_{\alpha \pi }`$. The objects of $`𝒞_\alpha `$ are families $`\{U_i𝒞_\alpha ^i\}_{iI}`$. The operations on the objects and the unit object are defined by
$$\{U_i\}_i\{U_i^{}\}_i=\{U_iU_i^{}\}_i,(\{U_i\}_i)^{}=\{U_i^{}\}_i,\text{1}\text{ I}=\{\text{1}\text{ I}_{𝒞^i}\}_i,$$
$`(1.4.a)`$
where $`i`$ runs over $`I`$. A morphism $`\{U_i𝒞^i\}_i\{U_i^{}𝒞^i\}_i`$ in $`𝒞`$ is a family $`\{f_i:U_iU_i^{}\}_i`$ where each $`f_i`$ is a morphism in $`𝒞^i`$. The $`K`$-additive structure in $`𝒞`$ is defined coordinate-wisely so that
$$\text{Hom}_𝒞(\{U_i\}_i,\{U_i^{}\}_i)=\underset{iI}{}\text{Hom}_{𝒞^i}(U_i,U_i^{}).$$
The composition of morphisms is coordinate-wise, i.e., $`\{f_i^{}:U_i^{}U_i^{\prime \prime }\}\{f_i:U_iU_i^{}\}=\{f_i^{}f_i:U_iU_i^{\prime \prime }\}`$. The tensor product for morphisms and the structural morphisms $`a,l,r,d,b`$ are defined coordinate-wisely. All the axioms of a $`\pi `$-category follow from the fact that they are satisfied coordinate-wisely.
For a finite family of $`\pi `$-categories $`\{𝒞^i\}_{iI}`$, we define a tensor product $`𝒞^{}=_i𝒞^i`$. The category $`𝒞^{}`$ is a disjoint union of categories $`\{𝒞_\alpha ^{}\}_{\alpha \pi }`$. The objects of $`𝒞_\alpha ^{}`$ are the same as the objects of the category $`𝒞_\alpha _i𝒞^i`$ above. The operations on the objects and the unit object are defined by (1.4.a). By definition,
$$\text{Hom}_𝒞^{}(\{U_i\}_i,\{U_i^{}\}_i)=\underset{iI}{}\text{Hom}_{𝒞^i}(U_i,U_i^{}).$$
This $`K`$-module is additively generated by elements of type $`_i(f_i:U_iU_i^{})`$. The composition of morphisms is defined on the generators by
$$\underset{i}{}(f_i^{}:U_i^{}U_i^{\prime \prime })\underset{i}{}(f_i:U_iU_i^{})=\underset{i}{}(f_i^{}f_i:U_iU_i^{\prime \prime }).$$
This extends to arbitrary morphisms by $`K`$-linearity and makes $`𝒞^{}`$ a $`K`$-additive category. The tensor product of morphisms is defined on the generators by $`(_if_i)(_ig_i)=_i(f_ig_i)`$ and extends to arbitrary morphisms by $`K`$-linearity. Observe that there is a canonical functor $`_i𝒞^i𝒞^{}`$ which is the identity on the objects and sends a morphism $`\{f_i\}_i`$ in $`_i𝒞^i`$ into the morphism $`_if_i`$ in $`𝒞^{}`$. This functor is by no means $`K`$-linear but does preserve the tensor product. Applying this functor to the structural morphisms in $`_i𝒞^i`$ defined above we obtain structural morphisms in $`𝒞^{}`$ satisfying all the conditions of Section 1.1. In this way $`𝒞^{}`$ becomes a $`\pi `$-category.
2. Crossed, braided, and ribbon $`\pi `$-categories
2.1. Crossed $`\pi `$-categories. Let $`𝒞`$ be a $`K`$-additive monoidal category with left duality. By an automorphism of $`𝒞`$ we mean an invertible $`K`$-linear (on the morphisms) functor $`\phi :𝒞𝒞`$ which preserves the tensor product, the unit object, the duality and the structural morphisms $`a,l,r,b,d`$. Thus,
$$\phi (\text{1}\text{ I})=\text{1}\text{ I},\phi (UV)=\phi (U)\phi (V),\phi (U^{})=(\phi (U))^{},\phi (fg)=\phi (f)\phi (g)$$
for any objects $`U,V`$ and any morphisms $`f,g`$ in $`𝒞`$ and
$$\phi (a_{U,V,W})=a_{\phi (U),\phi (V),\phi (W)},\phi (l_U)=l_{\phi (U)},\phi (r_U)=r_{\phi (U)},$$
$$\phi (b_U)=b_{\phi (U)},\phi (d_U)=d_{\phi (U)}$$
for any objects $`U,V,W𝒞`$. The group of automorphisms of $`𝒞`$ is denoted by $`\text{Aut}(𝒞)`$.
A crossed $`\pi `$-category over $`K`$ is a $`\pi `$-category $`𝒞`$ endowed with a group homomorphism $`\phi :\pi \text{Aut}(𝒞)`$ such that for all $`\alpha ,\beta \pi `$ the functor $`\phi _\alpha =\phi (\alpha ):𝒞𝒞`$ maps $`𝒞_\beta `$ into $`𝒞_{\alpha \beta \alpha ^1}`$. For objects $`U𝒞_\alpha ,V𝒞_\beta `$, set
$${}_{}{}^{U}V=\phi _\alpha (V)𝒞_{\alpha \beta \alpha ^1}.$$
In particular, $`{}_{}{}^{U}U=\phi _\alpha (U)𝒞_\alpha `$ for any $`U𝒞_\alpha `$. Note the identities
$${}_{}{}^{U}(VW)={}_{}{}^{U}V{}_{}{}^{U}W$$
$`(2.1.a)`$
$${}_{}{}^{(UV)}W={}_{}{}^{U}({}_{}{}^{V}W),$$
$`(2.1.b)`$
$${}_{}{}^{U}(V^{})=({}_{}{}^{U}V)^{},$$
$`(2.1.c)`$
$${}_{}{}^{\text{1}\text{ I}}V={}_{}{}^{U}({}_{}{}^{U^{}}V)={}_{}{}^{U^{}}({}_{}{}^{U}V)=V,{}_{}{}^{U}\text{1}\text{ I}=\text{1}\text{ I},$$
$`(2.1.d)`$
for any $`U,V,W𝒞`$. Similarly, for an object $`U𝒞_\alpha `$ and a morphism $`f:VV^{}`$ in $`𝒞`$ set
$${}_{}{}^{U}f=\phi _\alpha (f):{}_{}{}^{U}V{}_{}{}^{U}(V^{}).$$
Note the identities
$${}_{}{}^{U}(f^{}f)={}_{}{}^{U}(f^{}){}_{}{}^{U}f,$$
$`(2.1.e)`$
$${}_{}{}^{U}(fg)={}_{}{}^{U}f{}_{}{}^{U}g,$$
$`(2.1.f)`$
$${}_{}{}^{U}(id_V)=id_{({}_{}{}^{U}V)},{}_{}{}^{U}(b_V)=b_{({}_{}{}^{U}V)},{}_{}{}^{U}(d_V)=d_{({}_{}{}^{U}V)},$$
$`(2.1.g)`$
$${}_{}{}^{U}(l_V)=l_{({}_{}{}^{U}V)},{}_{}{}^{U}(r_V)=r_{({}_{}{}^{U}V)},{}_{}{}^{U}(a_{V,W,X})=a_{({}_{}{}^{U}V),({}_{}{}^{U}W),({}_{}{}^{U}X)},$$
$${}_{}{}^{(UV)}f={}_{}{}^{U}({}_{}{}^{V}f),{}_{}{}^{\text{1}\text{ I}}f={}_{}{}^{U}({}_{}{}^{U^{}}f)={}_{}{}^{U^{}}({}_{}{}^{U}f)=f.$$
$`(2.1.h)`$
Examples of crossed $`\pi `$-categories will be given in Section 2.6 and in further sections.
2.2. Braiding in $`\pi `$-categories. Let $`𝒞`$ be a crossed $`\pi `$-category. A braiding in $`𝒞`$ is a system of invertible morphisms
$$\{c_{U,V}:UV{}_{}{}^{U}VU\}_{U,V𝒞}$$
$`(2.2.a)`$
satisfying the following three conditions:
(2.2.1) for any morphisms $`f:UU^{},g:VV^{}`$ such that $`U,U^{}`$ lie in the same component of $`𝒞`$, we have
$$c_{U^{},V^{}}(fg)=({}_{}{}^{U}gf)c_{U,V};$$
$`(2.2.b)`$
(2.2.2) for any objects $`U,V,W𝒞`$ we have
$$c_{UV,W}$$
$`(2.2.c)`$
$$=a_{{}_{}{}^{(UV)}W,U,V}(c_{U,{}_{}{}^{V}W}\text{id}_V)(a_{U,{}_{}{}^{V}W,V})^1(\text{id}_Uc_{V,W})a_{U,V,W},$$
$$c_{U,VW}$$
$`(2.2.d)`$
$$=(a_{({}_{}{}^{U}V),({}_{}{}^{U}W),U})^1(\text{id}_{({}_{}{}^{U}V)}c_{U,W})a_{({}_{}{}^{U}V),U,W}(c_{U,V}\text{id}_W)(a_{U,V,W})^1;$$
(2.2.3) the action of $`\pi `$ on $`𝒞`$ preserves the braiding, i.e., for any $`\alpha \pi `$ and any $`V,W𝒞`$ we have
$$\phi _\alpha (c_{V,W})=c_{\phi _\alpha (V),\phi _\alpha (W)}.$$
Note that if in (2.2.1) the objects $`U,U^{}`$ do not lie in the same component of $`𝒞`$ then both sides of (2.2.b) are equal to 0 and have the same source $`UV`$ but may have different targets. Formulas (2.1.a) and (2.1.b) imply that the targets of the morphisms on the left and right hand sides of (2.2.c) and (2.2.d) are the same so that these equalities make sense.
A crossed $`\pi `$-category endowed with a braiding is said to be braided. For $`\pi =1`$, we obtain the standard definition of a braided monoidal category.
A braiding in a crossed $`\pi `$-category $`𝒞`$ satisfies a version of the Yang-Baxter identity. Assume for simplicity that $`𝒞`$ is strict. Then for any braiding (2.2.a) in $`𝒞`$ and any objects $`U,V,W𝒞`$,
$$(c_{({}_{}{}^{U}V),({}_{}{}^{U}W)}\text{id}_U)(\text{id}_{({}_{}{}^{U}V)}c_{U,W})(c_{U,V}\text{id}_W)$$
$`(2.2.e)`$
$$=(\text{id}_{({}_{}{}^{UV}W)}c_{U,V})(c_{U,{}_{}{}^{V}W}\text{id}_V)(\text{id}_Uc_{V,W}).$$
Indeed, by (2.2.d) and (2.2.3),
$$(c_{({}_{}{}^{U}V),({}_{}{}^{U}W)}\text{id}_U)(\text{id}_{({}_{}{}^{U}V)}c_{U,W})(c_{U,V}\text{id}_W)=({}_{}{}^{U}(c_{V,W})\text{id}_U)c_{U,VW}.$$
Applying (2.2.b) to $`f=\text{id}_U,g=c_{V,W}`$ and using (2.2.d), we obtain
$$({}_{}{}^{U}(c_{V,W})\text{id}_U)c_{U,VW}=c_{U,{}_{}{}^{V}WV}(\text{id}_Uc_{V,W})$$
$$=(\text{id}_{({}_{}{}^{UV}W)}c_{U,V})(c_{U,{}_{}{}^{V}W}\text{id}_V)(\text{id}_Uc_{V,W}).$$
If $`𝒞`$ is strict, then applying (2.2.c), (2.2.d) to $`U=V=\text{1}\text{ I}`$ and $`V=W=\text{1}\text{ I}`$ and using the invertibility of $`c_{U,\text{1}\text{ I}},c_{\text{1}\text{ I},U}`$, we obtain
$$c_{U,\text{1}\text{ I}}=c_{\text{1}\text{ I},U}=\text{id}_U$$
$`(2.2.f)`$
for any object $`U𝒞`$.
2.3. Twist in $`\pi `$-categories. A twist in a braided (crossed) $`\pi `$-category $`𝒞`$ is a family of invertible morphisms $`\{\theta _U:U{}_{}{}^{U}U\}_{U𝒞}`$ satisfying the following conditions:
(2.3.1) for any morphism $`f:UV`$ with $`U,V`$ lying in the same component of $`𝒞`$ we have $`\theta _Vf=({}_{}{}^{U}f)\theta _U`$;
(2.3.2) for any $`U𝒞`$ we have $`(\theta _U\text{id}_U^{})b_U=(\text{id}_{({}_{}{}^{U}U)}\theta _{({}_{}{}^{U}U)^{}})b_{({}_{}{}^{U}U)}`$;
(2.3.3) for any objects $`U,V𝒞`$, we have
$$\theta _{UV}=c_{{}_{}{}^{(UV)}V,{}_{}{}^{U}U}c_{({}_{}{}^{U}U),({}_{}{}^{V}V)}(\theta _U\theta _V);$$
$`(2.3.a)`$
(2.3.4) the action of $`\pi `$ on $`𝒞`$ preserves the twist, i.e., for any $`\alpha \pi `$ and any $`V𝒞`$, we have $`\phi _\alpha (\theta _V)=\theta _{\phi _\alpha (V)}`$.
As an exercise, the reader may check that the morphisms on both sides of the equations (2.3.1) - (2.3.4) have the same source and target. If $`𝒞`$ is strict, then it follows from (2.2.f) and (2.3.3) that $`\theta _{\text{1}\text{ I}}=\text{id}_{\text{1}\text{ I}}`$.
A braided crossed $`\pi `$-category endowed with a twist is called a ribbon crossed $`\pi `$-category. For $`\pi =1`$, we obtain the standard definition of a ribbon monoidal category.
The neutral component $`𝒞_1`$ of a ribbon crossed $`\pi `$-category $`𝒞`$ is a ribbon category in the usual sense of the word. Note also that every ribbon crossed $`\pi `$-category is equivalent to a strict ribbon crossed $`\pi `$-category in a canonical way.
2.4. Dual morphisms. Condition (2.3.2) is better understood when it is rewritten in terms of dual morphisms. For a morphism $`f:UV`$ in a monoidal category with left duality, the dual (or transpose) morphism $`f^{}:V^{}U^{}`$ is defined by
$$f^{}=(r_U^{})^1(d_V\text{id}_U^{})(a_{V^{},V,U^{}})^1(\text{id}_V^{}(f\text{id}_U^{}))(\text{id}_V^{}b_U)l_V^{}.$$
It follows from (1.1.g) that $`(\text{id}_U)^{}=\text{id}_U^{}`$. It is well-known that $`(fg)^{}=g^{}f^{}`$ for composable morphisms $`f,g`$. Condition (2.3.2) can be shown to be equivalent to
$$(\theta _U)^{}=\theta _{{}_{}{}^{U}(U^{})}.$$
$`2.4.a`$
2.5. Operations on ribbon group-categories. The operations on group-categories defined in Sections 1.4 can be adopted to the setting of crossed (resp. braided, ribbon) group-categories. Having a group homomorphism $`q:\pi ^{}\pi `$ we can pull back any crossed $`\pi `$-category $`𝒞`$ into a crossed $`\pi ^{}`$-category $`𝒞^{}=q^{}(𝒞)`$ as in Section 1.4 with action of $`\pi ^{}`$ defined by $`\phi _\alpha =\phi _{q(\alpha )}:𝒞_\beta ^{}𝒞_{\alpha \beta \alpha ^1}^{}`$ where $`\alpha ,\beta \pi ^{}`$. A braiding (resp. twist) in $`𝒞`$ induces a braiding (resp. twist) in $`𝒞^{}`$ in the obvious way. In particular, if $`\pi ^{}\pi `$ is a subgroup of $`\pi `$, then any crossed (resp. braided, ribbon) $`\pi `$-category $`𝒞=_{\alpha \pi }𝒞_\alpha `$ induces a crossed (resp. braided, ribbon) $`\pi ^{}`$-category $`_{\alpha \pi ^{}}𝒞_\alpha `$.
A crossed (resp. braided, ribbon) group-category can be pushed forward along group epimorphisms whose kernel acts trivially on the category. Consider a group epimorphism $`q:\pi ^{}\pi `$ whose kernel acts as the identity on a crossed (resp. braided, ribbon) $`\pi ^{}`$-category $`𝒞^{}`$. Then the action of $`\pi ^{}`$ on $`𝒞^{}`$ induces an action of $`\pi `$ on the push-forward $`\pi `$-category $`q_{}(𝒞^{})`$ defined in Section 1.4. A braiding (resp. twist) in $`𝒞^{}`$ induces a braiding (resp. twist) in $`q_{}(𝒞^{})`$ in the obvious way.
Given a family of crossed $`\pi `$-categories $`\{𝒞^i\}_{iI}`$, the direct product $`_i𝒞^i`$ is a crossed $`\pi `$-category. The action of $`\alpha \pi `$ on objects and morphisms is defined by
$$\phi _\alpha (\{U_i\}_{iI})=\{\phi _\alpha (U_i)\}_{iI},\phi _\alpha (\{f_i\}_{iI})=\{\phi _\alpha (f_i)\}_{iI}.$$
$`(2.5.a)`$
If $`\{𝒞^i\}_{iI}`$ are braided (resp. ribbon) $`\pi `$-categories then $`_i𝒞^i`$ is a braided (resp. ribbon) $`\pi `$-category: the braiding and twist are defined coordinate-wisely and their coordinates are the braiding and twist in $`\{𝒞^i\}_{iI}`$, respectively.
For a finite family of crossed $`\pi `$-categories $`\{𝒞^i\}_{iI}`$, the tensor product $`_i𝒞^i`$ is a crossed $`\pi `$-category. The action of $`\alpha \pi `$ on objects is defined as in (2.5.a). The action of $`\alpha \pi `$ on morphisms is defined on the generators by $`\phi _\alpha (_if_i)=_i\phi _\alpha (f_i)`$ and extends to arbitrary morphisms by $`K`$-linearity. If $`\{𝒞^i\}_{iI}`$ are braided (resp. ribbon) $`\pi `$-categories then $`_i𝒞^i`$ is a braided (resp. ribbon) $`\pi `$-category: the braiding and twist are obtained from the corresponding morphisms in $`_i𝒞^i`$ via the canonical functor $`_i𝒞^i_i𝒞^i`$.
We define a transformation of crossed $`\pi `$-categories called reflection. Let $`𝒞=_{\alpha \pi }𝒞_\alpha `$ be a crossed $`\pi `$-category with tensor product $``$, duality $``$, structural morphisms $`a,l,r,b,d`$ and $`\pi `$-action $`\phi :\pi \text{Aut}(𝒞)`$. We define a crossed $`\pi `$-category $`\overline{𝒞}=_{\alpha \pi }\overline{𝒞}_\alpha `$ with tensor product $`\overline{}`$, duality $``$, structural morphisms $`\overline{a},\overline{l},\overline{r},\overline{b},\overline{d}`$ and $`\pi `$-action $`\overline{\phi }:\pi \text{Aut}(\overline{𝒞})`$ as follows:
\- $`\overline{𝒞}=𝒞`$ as categories (but not monoidal categories);
\- $`\overline{𝒞}_\alpha =𝒞_{\alpha ^1}`$ as categories for all $`\alpha \pi `$;
\- $`\text{1}\text{ I}_{\overline{𝒞}}=\text{1}\text{ I}_𝒞`$;
\- for objects $`U\overline{𝒞}_\alpha ,V\overline{𝒞}_\beta `$, set $`U\overline{}V=\phi _{\beta ^1}(U)V\overline{𝒞}_{\alpha \beta }`$;
\- for morphisms $`f:UU^{},g:VV^{}`$ in $`\overline{𝒞}`$ with $`U\overline{𝒞}_\alpha ,U^{}\overline{𝒞}_\alpha ^{},V\overline{𝒞}_\beta ,V^{}\overline{𝒞}_\beta ^{}`$, set
$$f\overline{}g=\{\begin{array}{cc}\phi _{\beta ^1}(f)g\text{Hom}_{\overline{𝒞}}(U\overline{}V,U^{}\overline{}V^{}),if\beta =\beta ^{},\hfill & \\ 0\text{Hom}_{\overline{𝒞}}(U\overline{}V,U^{}\overline{}V^{}),if\beta \beta ^{};\hfill & \end{array}$$
\- for $`U\overline{𝒞}_\alpha `$, set $`U^{}=\phi _\alpha (U^{})\overline{𝒞}_{\alpha ^1}`$ and
$$\overline{l}_U=l_U,\overline{r}_U=r_U,\overline{d}_U=d_U,\overline{b}_U=\phi _\alpha (b_U);$$
\- for objects $`U\overline{𝒞}_\alpha ,V\overline{𝒞}_\beta ,W\overline{𝒞}_\gamma `$, set
$$\overline{a}_{U,V,W}=a_{\phi _{\gamma ^1\beta ^1}(U),\phi _{\gamma ^1}(V),W};$$
\- for $`\alpha \pi `$ set $`\overline{\phi }_\alpha =\phi _\alpha `$.
A routine check shows that $`\overline{𝒞}`$ is a crossed $`\pi `$-category. Moreover, if $`c,\theta `$ are a braiding and a twist in $`𝒞`$ then the formulas
$$\overline{c}_{U,V}=(c_{V,U})^1,\overline{\theta }_U=(\theta _{\phi _\alpha (U)})^1$$
(where $`U\overline{𝒞}_\alpha `$) define a braiding and a twist in $`\overline{𝒞}`$. We call $`\overline{𝒞}`$ the mirror of $`𝒞`$. Its neutral component $`\overline{𝒞}_1`$ is the mirror of $`𝒞_1`$ in the sense of \[Tu2, Section I.1.4\]. It is easy to see that $`\overline{\overline{𝒞}}=𝒞`$.
2.6. Example. Consider the $`\pi `$-category $`𝒞`$ defined in Section 1.3 and assume that both $`a`$ and $`b`$ are invariant under conjugation, i.e.,
$$a_{\delta \alpha \delta ^1,\delta \beta \delta ^1,\delta \gamma \delta ^1}=a_{\alpha ,\beta ,\gamma },$$
$`(2.6.a)`$
and $`b_{\delta \alpha \delta ^1}=b_\alpha `$ for any $`\alpha ,\beta ,\gamma ,\delta \pi `$. Then $`𝒞`$ is a crossed $`\pi `$-category as follows. For $`\alpha ,\beta \pi `$, set $`\phi _\alpha (V_\beta )=V_{\alpha \beta \alpha ^1}`$. This extends to morphisms in $`𝒞`$ in the obvious way since all non-zero morphisms in $`𝒞`$ are proportional to the identity endomorphisms of objects. The resulting functor $`\phi _\alpha :𝒞𝒞`$ preserves all the structural morphisms in $`𝒞`$ since $`a`$ and $`b`$ are conjugation invariant. To construct specific examples we can take $`b=1`$. Finding conjugation invariant 3-cocylces is a delicate task. Obvious examples: the trivial cocycle $`a=1`$; any 3-cocycle in the case of abelian $`\pi `$.
It is clear that a braiding in $`𝒞`$ is given by a family $`\{c_{\alpha ,\beta }K^{}\}_{\alpha ,\beta \pi }`$ where $`c_{\alpha ,\beta }`$ determines the braiding morphism
$$V_{\alpha \beta }=V_\alpha V_\beta \phi _\alpha (V_\beta )V_\alpha =V_{\alpha \beta \alpha ^1}V_\alpha =V_{\alpha \beta }.$$
$`(2.6.b)`$
The conditions on the braiding can be reformulated as the following idenitites:
$$c_{\delta \alpha \delta ^1,\delta \beta \delta ^1}=c_{\alpha ,\beta },$$
$`(2.6.c)`$
$$c_{\alpha \beta ,\gamma }=c_{\beta ,\gamma }c_{\alpha ,\beta \gamma \beta ^1}a_{\alpha ,\beta ,\gamma }(a_{\alpha ,\beta \gamma \beta ^1,\beta })^1a_{\alpha \beta \gamma \beta ^1\alpha ^1,\alpha ,\beta },$$
$`(2.6.d)`$
$$c_{\alpha ,\beta \gamma }=c_{\alpha ,\beta }c_{\alpha ,\gamma }(a_{\alpha ,\beta ,\gamma })^1a_{\alpha \beta \alpha ^1,\alpha ,\gamma }(a_{\beta ,\gamma ,\alpha })^1,$$
$`(2.6.e)`$
for all $`\alpha ,\beta ,\gamma ,\delta \pi `$. The equality (2.6.d) can be rewritten in a more convenient form using (2.6.c). Namely, observe that $`c_{\alpha ,\beta \gamma \beta ^1}=c_{\delta ,\gamma }`$ with $`\delta =\beta ^1\alpha \beta `$. Now, $`\alpha \beta =\beta \delta `$ which gives the following equivalent form of (2.6.d):
$$c_{\beta \delta ,\gamma }=c_{\beta ,\gamma }c_{\delta ,\gamma }a_{\beta \delta \beta ^1,\beta ,\gamma }(a_{\delta ,\gamma ,\beta })^1a_{\delta \gamma \delta ^1,\delta ,\beta }.$$
$`(2.6.f)`$
A direct computation shows that (2.6.c) follows from (1.3.a), (2.6.a,e,f).
The definition of a twist in $`𝒞`$ considerably simplifies since $`{}_{}{}^{U}U=U`$ for any $`U𝒞`$. Given a braiding $`\{c_{\alpha ,\beta }K^{}\}_{\alpha ,\beta \pi }`$ in $`𝒞`$, a twist in $`𝒞`$ is determined by a family $`\{\theta _\alpha K^{}\}_{\alpha \pi }`$ (where $`\theta _\alpha `$ is the twist $`V_\alpha V_\alpha =\phi _\alpha (V_\alpha )`$) such that
$$\theta _{\alpha \beta }=c_{\alpha ,\beta }c_{\beta ,\alpha }\theta _\alpha \theta _\beta ,$$
$`(2.6.g)`$
$$\theta _{\alpha ^1}=\theta _\alpha $$
$`(2.6.h)`$
for all $`\alpha ,\beta \pi `$. The equality (2.6.g) implies that $`\theta _{\alpha \beta }=\theta _{\beta \alpha }`$ so that $`\theta `$ is conjugation invariant.
To sum up, a conjugation invariant tuple $`(a,b,c,\theta )`$ satisfying (1.3.a), (2.6.e-h) gives rise to a ribbon crossed $`\pi `$-category $`𝒞=𝒞(a,b,c,\theta )`$. Such tuples $`(a,b,c,\theta )`$ form a group under pointwise multiplication. This group operation corresponds to tensor multiplication of the $`\pi `$-categories $`𝒞(a,b,c,\theta )`$. The reflection of ribbon $`\pi `$-categories defined in Section 2.5 corresponds to the following involution in the set of tuples $`(a,b,c,\theta )`$:
$$\overline{a}_{\alpha ,\beta ,\gamma }=a_{\beta ^1\alpha ^1\beta ,\beta ^1,\gamma ^1},\overline{c}_{\alpha ,\beta }=(c_{\beta ^1,\alpha ^1})^1,$$
$`(2.6.i)`$
$$\overline{b}_\alpha =b_{\alpha ^1},\overline{\theta }_\alpha =(\theta _\alpha )^1$$
where $`\alpha ,\beta ,\gamma \pi `$. We shall further discuss equations (2.6.e-h) in Appendix 1.
If $`G\pi `$ is a subgroup of the center of $`\pi `$ then the action of $`G`$ on $`𝒞(a,b,c,\theta )`$ is trivial so that we can push $`𝒞(a,b,c,\theta )`$ forward along the projection $`\pi \pi /G`$. This gives a ribbon crossed $`(\pi /G)`$-category.
2.7. Remarks. 1. The objects of a crossed $`\pi `$-category $`(𝒞,\phi :\pi \text{Aut}(𝒞))`$ form a $`\pi `$-automorphic set in terminology of Brieskorn \[Br\] or a $`\pi `$-rack in terminology of Fenn and Rourke \[FR2\]. Recall that a $`\pi `$-rack is a set $`X`$ equipped with a left action of $`\pi `$ and a map $`:X\pi `$ such that $`(\alpha a)=\alpha (a)\alpha ^1`$ for all $`\alpha \pi ,aX`$. The underlying $`\pi `$-rack of $`𝒞`$ comprises the set of objects of $`𝒞`$, the action of $`\pi `$ on this set induced by $`\phi `$ and the map assigning to any object of $`𝒞_\alpha `$ the element $`\alpha \pi `$.
2. It is clear from definitions that a braiding in a crossed $`\pi `$-category in general is not a braiding in the underlying monoidal category in the usual sense of the word. There is one exceptional case. Namely, assume that $`\pi `$ is abelian and $`𝒞`$ is a $`\pi `$-category as in Section 1.2. The trivial homomorphism $`\phi =1:\pi \text{Aut}(𝒞)`$ makes $`𝒞`$ a crossed $`\pi `$-category. It is clear that a braiding (resp. twist) in this crossed $`\pi `$-category is a braiding (resp. twist) in $`𝒞`$ in the usual sense.
3. Colored $`\pi `$-tangles and their invariants
3.1. Colored $`\pi `$-links. It is well-known that a framed oriented link in $`S^3`$ whose components are colored with objects of a ribbon category gives rise to an invariant lying in the ground ring of this category, see \[Tu2\]. To adapt this theory to our present setting we introduce $`\pi `$-links and their colorings. For the sake of future references, we consider links in an arbitrary connected oriented 3-manifold $`M`$.
Let $`\mathrm{}=\mathrm{}_1\mathrm{}\mathrm{}_nM`$ be an oriented $`n`$-component link in $`M`$ with $`n0`$. Denote the open 3-manifold $`M\backslash \mathrm{}`$ by $`C_{\mathrm{}}`$ where $`C`$ stands for complement. We say that $`\mathrm{}`$ is framed if each its component $`\mathrm{}_i`$ is provided with a longitude $`\stackrel{~}{\mathrm{}}_iC_{\mathrm{}}`$ which goes very closely along $`\mathrm{}_i`$ (it may wind around $`\mathrm{}_i`$ several times). Set $`\stackrel{~}{\mathrm{}}=_{i=1}^n\stackrel{~}{\mathrm{}}_i`$. For a path $`\gamma :[0,1]C_{\mathrm{}}`$ connecting a point $`z=\gamma (0)`$ to a point $`\gamma (1)\stackrel{~}{\mathrm{}}_i`$, denote by $`\mu _\gamma \pi _1(C_{\mathrm{}},z)`$ the (homotopy) meridian of $`\mathrm{}_i`$ represented by the loop $`\gamma m_i\gamma ^1`$, where $`m_i`$ is a small loop encircling $`\mathrm{}_i`$ with linking number $`+1`$. We similarly define a (homotopy) longitude $`\lambda _\gamma =[\gamma \stackrel{~}{\mathrm{}}_i\gamma ^1]\pi _1(C_{\mathrm{}},z)`$ where the square brackets denote the homotopy class of a loop and the circle $`\stackrel{~}{\mathrm{}}_i`$ is viewed as a loop beginning and ending in $`\gamma (1)`$. Both $`\mu _\gamma `$ and $`\lambda _\gamma `$ are invariant under homotopies of $`\gamma `$ fixing $`\gamma (0)=z`$ and keeping $`\gamma (1)`$ on $`\stackrel{~}{\mathrm{}}`$. Clearly, $`\mu _\gamma `$ and $`\lambda _\gamma `$ commute in $`\pi _1(C_{\mathrm{}},z)`$. If $`\beta `$ is a loop in $`(C_{\mathrm{}},z)`$ (i.e., a path $`[0,1]C_{\mathrm{}}`$ beginning and ending in $`z`$) then
$$\mu _{\beta \gamma }=[\beta ]\mu _\gamma [\beta ]^1\text{and}\lambda _{\beta \gamma }=[\beta ]\lambda _\gamma [\beta ]^1.$$
By a $`\pi `$-link in $`M`$ we shall mean a triple (a framed oriented link $`\mathrm{}M`$, a base point $`zC_{\mathrm{}}`$, a group homomorphism $`g:\pi _1(C_{\mathrm{}},z)\pi `$).
Fix a crossed $`\pi `$-category $`(𝒞,\phi :\pi \text{Aut}(𝒞))`$ which we call the category of colors. A coloring of a $`\pi `$-link $`(\mathrm{},z,g)`$ is a function $`u`$ which assigns to every path $`\gamma :[0,1]C_{\mathrm{}}`$ with $`\gamma (0)=z,\gamma (1)\stackrel{~}{\mathrm{}}`$ an object $`u_\gamma 𝒞_{g(\mu _\gamma )}`$ such that
(i) $`u_\gamma `$ is preserved under homotopies of $`\gamma `$ fixing $`\gamma (0)=z`$ and keeping $`\gamma (1)`$ on $`\stackrel{~}{\mathrm{}}`$;
(ii) if $`\beta `$ is a loop in $`(C_{\mathrm{}},z)`$, then $`u_{\beta \gamma }=\phi _{g([\beta ])}(u_\gamma )`$.
Pushing the endpoint $`\gamma (1)\stackrel{~}{\mathrm{}}`$ of a path $`\gamma `$ as above along the corresponding component of $`\stackrel{~}{\mathrm{}}`$ we can deform $`\gamma `$ into a path homotopic to $`\lambda _\gamma \gamma `$. Conditions (i), (ii) imply that
$$\phi _{g(\lambda _\gamma )}(u_\gamma )=u_\gamma .$$
$`(3.1.a)`$
We shall see below (Lemma 3.2.1) that a coloring of an $`n`$-component $`\pi `$-link $`(\mathrm{}=\mathrm{}_1\mathrm{}\mathrm{}_n,z,g)`$ is uniquely determined by the objects associated to any given $`n`$ paths $`\gamma _1,\mathrm{},\gamma _n`$ connecting $`z`$ to $`\stackrel{~}{\mathrm{}}_1,\mathrm{},\stackrel{~}{\mathrm{}}_n`$, respectively. In the role of these objects we can take any objects of $`𝒞_{g(\mu _{\gamma _1})},\mathrm{},𝒞_{g(\mu _{\gamma _n})}`$ satisfying (3.1.a).
A $`\pi `$-link endowed with a coloring is said to be $`𝒞`$-colored or briefly colored. The notion of an ambient isotopy in $`M`$ applies to $`\pi `$-links and colored $`\pi `$-links in $`M`$ in the obvious way. This allows us to consider the (ambient) isotopy classes of such links.
The structure of a colored $`\pi `$-link $`(\mathrm{},z,g,u)`$ in $`M`$ can be transferred along paths in $`M\backslash \mathrm{}`$ relating various base points. More precisely, let $`\rho :[0,1]M\backslash \mathrm{}`$ be a path with $`\rho (0)=z`$. We define a new colored $`\pi `$-link $`(\mathrm{}^{}=\mathrm{},z^{},g^{}:\pi _1(C_{\mathrm{}},z^{})\pi ,u^{})`$ by $`z^{}=\rho (1)`$, $`g^{}([\alpha ])=g([\rho \alpha \rho ^1])`$ for any loop $`\alpha `$ in $`(M\backslash \mathrm{},z^{})`$, $`u_\gamma ^{}=u_{\rho \gamma }`$ for any path $`\gamma `$ in $`C_{\mathrm{}}`$ leading from $`z^{}`$ to $`\stackrel{~}{\mathrm{}}`$. It is clear that the transfers along homotopic paths (with the same endpoints) give the same results. The transfer preserves the ambient isotopy class of the colored $`\pi `$-link $`(\mathrm{},z,g,u)`$: it is ambiently isotopic to $`(\mathrm{},z^{},g^{},u^{})`$ via an isotopy of the identity map $`\text{id}_M:MM`$ which pushes $`z`$ along $`\rho `$ and is constant in a neighborhood of $`\mathrm{}`$.
Although we shall not need it, note that a $`𝒞`$-colored $`\pi `$-link can be defined in terms of $`\pi `$-racks (cf. Section 2.7) as a framed oriented link endowed with a homomorphism of its fundamental rack (as defined in \[FR2\]) into the underlying $`\pi `$-rack of $`𝒞`$. The notion of colored $`\pi `$-links can be reformulated also in terms of principal $`\pi `$-bundles over link complements.
3.2. Colored $`\pi `$-tangles. By a tangle with $`k0`$ inputs and $`l0`$ outputs we mean a tangle $`T^2\times [0,1]`$ with bottom endpoints (inputs) $`(r,0,0),r=1,\mathrm{},k`$ and top endpoints (outputs) $`(s,0,1),s=1,\mathrm{},l`$. The tangle $`T`$ consists of a finite number of mutually disjoint oriented embedded circles and arcs lying in the open strip $`^2\times ]0,1[`$ except the endpoints of the arcs. At the endpoints, $`T`$ should be orthogonal to the planes $`^2\times 0,^2\times 1`$. We denote the open 3-manifold $`(^2\times [0,1])\backslash T`$ by $`C_T`$.
We say that $`T`$ is framed if each its component $`t`$ is provided with a longitude $`\stackrel{~}{t}C_T`$ which goes very closely along $`t`$. Clearly, $`\stackrel{~}{t}`$ is an arc (resp. a circle) if $`t`$ is an arc (resp. a circle). We always assume that the longitudes of the arc components of $`T`$ have the endpoints $`(r,\delta ,0),r=1,\mathrm{},k`$ and $`(s,\delta ,1),s=1,\mathrm{},l`$ with small $`\delta >0`$. Set $`\stackrel{~}{T}=_t\stackrel{~}{t}`$ where $`t`$ runs over all the components of $`T`$.
As the base point of $`C_T`$ we always choose a point $`z`$ with a big negative second coordinate $`z_20`$ so that $`T\times [z_2+1,\mathrm{}]\times [0,1]`$. The set of such $`z`$ is contractible. This allows us to supress the base point from the notation for the fundamental group of $`C_T`$.
If $`T`$ is oriented, then for each path $`\gamma :[0,1]C_T`$ connecting the base point to $`\stackrel{~}{T}`$, we introduce a meridian $`\mu _\gamma \pi _1(C_T)`$ as in Section 3.1. If $`\gamma (1)`$ lies on a circle component of $`\stackrel{~}{T}`$ then we also have a longitude $`\lambda _\gamma \pi _1(C_T)`$. Both $`\mu _\gamma `$ and $`\lambda _\gamma `$ are invariant under homotopies of $`\gamma `$ fixing $`\gamma (0)`$ and keeping $`\gamma (1)`$ on $`\stackrel{~}{T}`$. Clearly, $`\mu _\gamma `$ and $`\lambda _\gamma `$ commute in $`\pi _1(C_T)`$.
A $`\pi `$-tangle is a pair (a framed oriented tangle $`T`$, a group homomorphism $`g:\pi _1(C_T)\pi `$). The definition of a coloring of a $`\pi `$-link extends to $`\pi `$-tangles word for word. A $`\pi `$-tangle endowed with a coloring is said to be colored.
As usual, we shall consider $`\pi `$-tangles and colored $`\pi `$-tangles up to ambient isotopy in $`^2\times [0,1]`$ constant on the endpoints. The $`\pi `$-tangles (resp. colored $`\pi `$-tangles) with 0 inputs and 0 outputs are nothing but $`\pi `$-links (resp. colored $`\pi `$-links) in $`^2\times ]0,1[`$.
Let $`T=(T,g,u)`$ be a colored $`\pi `$-tangle. With the $`r`$-th input $`(r,0,0)`$ of $`T`$ we associate a triple $`(\epsilon _r=\pm ,\alpha _r\pi ,U_r𝒞_{\alpha _r})`$ as follows. Set $`\epsilon _r=+`$ if the arc component of $`T`$ incident to the $`r`$-th input is directed out of $`^2\times [0,1]`$ and set $`\epsilon _r=`$ otherwise. Let $`\gamma _r`$ be the path in $`C_T`$ leading from the base point $`z=(z_1,z_2,z_3)`$ of $`C_T`$ to the $`r`$-th input of $`\stackrel{~}{T}`$ and defined as composition of the linear path from $`z`$ to $`(r,z_2,0)`$ with the linear path from $`(r,z_2,0)`$ to $`(r,\delta ,0)`$. We call $`\gamma _r`$ the canonical path associated to the $`r`$-th input of $`T`$. The meridian $`\mu _r=\mu _{\gamma _r}\pi _1(C_T)`$ is called the canonical meridian associated to the $`r`$-th input of $`T`$. Set $`\alpha _r=g(\mu _r)\pi `$ and $`U_r=u_{\gamma _r}𝒞_{\alpha _r}`$. The sequence $`(\epsilon _1,\alpha _1,U_1),\mathrm{},(\epsilon _k,\alpha _k,U_k)`$ (where $`k`$ is the number of inputs of $`T`$) is called the source of $`T`$. Similarly, with the $`s`$-th output $`(s,0,1)`$ of $`T`$ we associate a triple $`(\epsilon ^s=\pm ,\alpha ^s\pi ,U^s𝒞_{\alpha ^s})`$ as follows. Set $`\epsilon ^s=+`$ if the arc component of $`T`$ incident to the $`s`$-th output is directed inside $`^2\times [0,1]`$ and set $`\epsilon ^s=`$ otherwise. Let $`\gamma ^s`$ be the canonical path leading from the base point $`z=(z_1,z_2,z_3)`$ of $`C_T`$ to the $`s`$-th output $`(s,\delta ,1)`$ of $`\stackrel{~}{T}`$: It is defined as the composition of the linear path leading from $`z`$ to $`(s,z_2,1)`$ with the linear path leading from $`(s,z_2,1)`$ to $`(s,\delta ,1)`$. Set $`\alpha ^s=g(\mu _{\gamma ^s})\pi `$ and $`U^s=u_{\gamma ^s}𝒞_{\alpha ^s}`$. The sequence $`(\epsilon ^1,\alpha ^1,U^1),\mathrm{},(\epsilon ^l,\alpha ^l,U^l)`$ (where $`l`$ is the number of outputs of $`T`$) is called the target of $`T`$. Note the obvious equality
$$\underset{r=1}{\overset{k}{}}(\alpha _r)^{\epsilon _r}=\underset{s=1}{\overset{l}{}}(\alpha ^s)^{\epsilon ^s}.$$
A colored $`\pi `$-tangle $`(T,g:\pi _1(C_T)\pi ,u)`$ is said to be a colored $`\pi `$-braid if $`T`$ is a framed oriented braid. Let $`k`$ be the number of strings of $`T`$. Observe that the orientation of $`T`$, the homomorphism $`g`$ and the coloring $`u`$ are uniquely determined by the source $`(\epsilon _1,\alpha _1,U_1),\mathrm{},(\epsilon _k,\alpha _k,U_k)`$ of $`T`$. Indeed, the signs $`\epsilon _1,\mathrm{},\epsilon _k`$ determine the orientation of the strings. The group $`\pi _1(C_T)`$ is free on $`k`$ generators represented by the canonical meridians $`\mu _1,\mathrm{},\mu _k`$ corresponding to the inputs of $`T`$. Hence, the homomorphism $`g`$ is determined by $`\alpha _1=g(\mu _1),\mathrm{},\alpha _k=g(\mu _k)`$. It follows from definitions that the coloring $`u`$ is determined by $`U_1,\mathrm{},U_k`$. The next lemma implies that, conversely, given a framed braid $`T`$ on $`k0`$ strings and a finite sequence $`(\epsilon _1,\alpha _1,U_1),\mathrm{},(\epsilon _k,\alpha _k,U_k)`$ with $`\epsilon _r=\pm ,\alpha _r\pi ,U_r𝒞_{\alpha _r}`$ for $`r=1,\mathrm{},k`$, we can extend $`T`$ uniquely to a colored $`\pi `$-braid with source $`(\epsilon _1,\alpha _1,U_1),\mathrm{},(\epsilon _k,\alpha _k,U_k)`$.
3.2.1. Lemma. Let $`(T=t_1\mathrm{}t_n,g)`$ be an $`n`$-component $`\pi `$-tangle. Let $`\rho _i:[0,1]C_T`$ be a path connecting the base point of $`C_T`$ to a point of $`\stackrel{~}{t}_i`$ where $`i=1,\mathrm{},n`$. Let $`U_i𝒞_{g(\mu _{\rho _i})}`$ for $`i=1,\mathrm{},n`$. Assume that for each circle component $`t_i`$ of $`T`$ we have $`\phi _{g(\lambda _{\rho _i})}(U_i)=U_i`$. Then there is a unique coloring $`u`$ of $`T`$ such that $`u_{\rho _i}=U_i`$ for $`i=1,\mathrm{},n`$.
Proof. Any path $`\gamma :[0,1]C_T`$ connecting the base point $`z`$ to $`\stackrel{~}{t}_i`$ can be deformed fixing $`\gamma (0)=z`$ and keeping $`\gamma (1)\stackrel{~}{t}_i`$ so that $`\gamma (1)=\rho _i(1)`$. Then
$$u_\gamma =u_{\gamma \rho _i^1\rho _i}=\phi _{g([\gamma \rho _i^1])}(u_{\rho _i})=\phi _{g([\gamma \rho _i^1])}(U_i).$$
This proves the uniqueness. Conversely, our assumptions imply that the object
$$u_\gamma =\phi _{g([\gamma \rho _i^1])}(U_i)𝒞_{g(\mu _\gamma )}$$
$`(3.2.a)`$
does not depend on the choice of the deformation of $`\gamma `$ used to ensure $`\gamma (1)=\rho _i(1)`$. It is easy to check that formula (3.2.a) defines a coloring $`u`$ of $`\mathrm{}`$ such that $`u_{\rho _i}=U_i`$ for all $`i=1,\mathrm{},n`$.
3.3. Diagrams for colored $`\pi `$-tangles. It is standard in knot theory to present oriented framed tangles by plane pictures called tangle diagrams. A tangle diagram lies in a horizontal strip on the page of the picture identified with $`\times [0,1]=\times 0\times [0,1]^2\times [0,1]`$. We agree that the first axis is a horizontal line on the page of the picture directed to the right, the second axis is orthogonal to the plane of the picture and is directed from the reader towards this plane, the third axis is a vertical line on the plane of the picture directed from the bottom to the top. A tangle diagram consists of oriented immersed arcs and circles lying in general position with indication of over/undercrossings in all double points. The diagram has the same inputs and outputs as the corresponding tangle. The framing is given by shifting the tangle along the vector $`(0,\delta ,0)`$ with small $`\delta >0`$. Note that the points with negative second coordinate lie above the picture.
It is easy to extend the technique of tangle diagrams to present a $`\pi `$-tangle $`(T,g:\pi _1(C_T)\pi )`$. We first present $`T`$ by a tangle diagram, $`D`$. The undercrossings of $`D`$ split $`D`$ into disjoint oriented embedded arcs in $`\times [0,1]`$. (We do not break $`D`$ at the overcrossings). For each of these arcs, say $`e`$, consider the linear path in $`^2\times [0,1]`$ connecting the base point $`zC_T`$ to a point of $`e+(0,\delta ,0)`$. In the pictorial language, the point $`z`$ lies high above $`D`$ and the path in question is obtained by rushing from $`z`$ straight to a point lying slightly above $`e`$. Denote this path by $`\gamma (e)`$. We label $`e`$ with $`g_e=g(\mu _{\gamma (e)})\pi `$. In pictures, one usually puts $`g_e`$ on a small arrow drawn beneath $`e`$ and crossing $`e`$ from right to left. Knowing $`g_e`$ for all arcs $`e`$ of $`D`$, we can recover the homomorphism $`g`$ because the meridians $`\{\mu _{\gamma (e)}\}_e`$ generate $`\pi _1(C_T)`$. We say that the $`\pi `$-tangle $`(T,g)`$ is presented by the diagram $`D`$ whose arcs are labeled by elements of $`\pi `$ as above.
A tangle diagram $`D`$ whose arcs are labeled by elements of $`\pi `$ presents a $`\pi `$-tangle if and only if the following local condition is satisfied:
$`()`$. Encircling a double point of $`D`$ and multiplying the corresponding four elements of $`\pi `$ we always obtain $`1\pi `$.
It is understood that crossing an arc $`e`$ from right to left we read $`g_e`$ while crossing $`e`$ from left to right we read $`g_e^1`$.
To present a colored $`\pi `$-tangle $`(T,g,u)`$ we additionally endow each arc $`e`$ of $`D`$ with the object $`u_{\gamma (e)}𝒞_{g_e}`$. This data uniquely determines $`(T,g,u)`$.
Conversely, consider a tangle diagram $`D`$ whose arcs $`e`$ are labeled with pairs $`(g_e\pi ,U_e𝒞_{g_e})`$. Assume that condition $`()`$ is met. Then $`D`$ presents a colored $`\pi `$-tangle if and only if the following local condition is satisfied:
$`()`$. For any double point $`d`$ of $`D`$, consider the three arcs $`e,f,h`$ of $`D`$ incident to $`d`$ such that in a neighborhood of $`d`$ they appear as the overcrossing, the incoming undercrossing and the outgoing undercrossing, respectively. Then $`U_f=(\phi _{g_e})^\epsilon (U_h)`$ where $`\epsilon =\pm 1`$ is the sign of $`d`$.
The necessity of $`()`$ follows from the definition of a coloring and the obvious equality $`\gamma (f)=(\mu _{\gamma (e)})^\epsilon \gamma (h)`$. The sufficiency of $`()`$ follows from Lemma 3.2.1 and the fact that $`()`$ implies (3.1.a) for all circle components of $`T`$.
In the sequel by a diagam of a colored $`\pi `$-tangle we mean its diagram labeled as above so that conditions $`(),()`$ are satisfied. The technique of Reidemeister moves on diagrams of framed oriented tangles (see, for instance, \[Tu2, Section I.4\]) extends to the diagrams of colored $`\pi `$-tangles. The key point is that any Reidemeister move on the underlying unlabeled diagram extends uniquely to a move on the labels keeping all the labels outside of the 2-disc where the diagram is modified. As in the standard theory, two diagrams of colored $`\pi `$-tangles represent isotopic colored $`\pi `$-tangles if and only if they can be related by a finite sequence of (labeled) Reidemeister moves.
3.4. Category of colored $`\pi `$-tangles. We define a category of colored $`\pi `$-tangles $`𝒯=𝒯(\pi ,𝒞,K)`$ as follows. The objects of $`𝒯`$ are finite sequences $`\{(\epsilon _r,\alpha _r,U_r)\}_{r=1}^k`$ where $`k0,\epsilon _r=\pm ,\alpha _r\pi ,U_r𝒞_{\alpha _r}`$ for $`r=1,\mathrm{},k`$. A morphism of such sequences $`\{(\epsilon _r,\alpha _r,U_r)\}_{r=1}^k\{(\epsilon ^s,\alpha ^s,U^s)\}_{s=1}^l`$ in $`𝒯`$ is a finite formal sum $`_iz_iT_i`$ where $`z_iK`$ and $`T_i`$ is a colored $`\pi `$-tangle with $`k`$ inputs and $`l`$ outputs such that $`\{(\epsilon _r,\alpha _r,U_r)\}_{r=1}^k`$ is the source of $`T_i`$ and $`\{(\epsilon ^s,\alpha ^s,U^s)\}_{s=1}^l`$ is the target of $`T_i`$. If the sum $`z_iT_i`$ consists of one term $`1T`$ then we say that the corresponding morphism in $`𝒯`$ is represented by $`T`$. The composition of morphisms $`TT^{}`$ represented by colored $`\pi `$-tangles $`T,T^{}`$ is obtained by the gluing of $`T`$ on the top of $`T^{}`$; this extends by $`K`$-linearity to all morphisms. The identity morphism of an object $`\{(\epsilon _r,\alpha _r,U_r)\}_{r=1}^k`$ is represented by the trivial colored $`\pi `$-braid with zero framing and source (and target) $`\{(\epsilon _r,\alpha _r,U_r)\}_{r=1}^k`$. This completes the definition of $`𝒯`$.
3.4.1. Lemma. The category $`𝒯`$ is a strict ribbon crossed $`\pi `$-category.
Proof. The tensor product for the objects of $`𝒯`$ is the juxtaposition of sequences. The unit object is the empty sequence. The tensor product of the morphisms in $`𝒯`$ represented by colored $`\pi `$-tangles $`T,T^{}`$ is obtained by placing any diagram of $`T^{}`$ on the right of any diagram of $`T`$. Then the union of these two diagrams represents the colored $`\pi `$-tangle $`TT^{}`$. This extends to arbitrary morphisms in $`𝒯`$ by linearity. In this way, $`𝒯`$ becomes a strict monoidal category.
The dual of an object $`U=((\epsilon _1,\alpha _1,U_1),\mathrm{},(\epsilon _k,\alpha _k,U_k))𝒯`$ is by definition the object $`U^{}=((\epsilon _k,\alpha _k,U_k),(\epsilon _{k1},\alpha _{k1},U_{k1}),\mathrm{},(\epsilon _1,\alpha _1,U_1))`$. The duality morphisms $`b_U:\text{1}\text{ I}UU^{}`$ and $`d_U:U^{}U`$ are defined in the same way as in the usual theory where $`\pi =1`$. They are represented by tangle diagrams consisting of $`k`$ disjoint concentric cups (resp. caps). The orientations and labels on these cups (resp. caps) are uniquely determined by the data at the outputs (resp. inputs). Formulas (1.1.f) and (1.1.g) are straightforward. Note that the associativity morphisms and the structural morphisms $`l,r`$ in $`𝒯`$ are the identities so that formulas (1.1.f) and (1.1.g) simplify to
$$(\text{id}_Ud_U)(b_U\text{id}_U)=\text{id}_U,(d_U\text{id}_U^{})(\text{id}_U^{}b_U)=\text{id}_U^{}.$$
$`(3.4.a)`$
Given $`\alpha \pi `$, consider a full subcategory $`𝒯_\alpha `$ of $`𝒯`$ whose objects are sequences $`((\epsilon _1,\alpha _1,U_1),\mathrm{},(\epsilon _k,\alpha _k,U_k))`$ with $`_{r=1}^k(\alpha _r)^{\epsilon _r}=\alpha `$. It is obvious that $`𝒯=_{\alpha \pi }𝒯_\alpha `$. It follows from definitions that $`𝒯`$ is a $`\pi `$-category.
For each $`\alpha \pi `$ we define an automorphism $`\phi _\alpha `$ of $`𝒯`$ as follows. The action on the objects is given by
$$\phi _\alpha (\{(\epsilon _r,\alpha _r,U_r)\}_{r=1}^k)=\{(\epsilon _r,\alpha \alpha _r\alpha ^1,\phi _\alpha (U_r))\}_{r=1}^k$$
where on the right-hand side we use the action of $`\alpha `$ on the category of colors $`𝒞`$. The automorphism $`\phi _\alpha :𝒯𝒯`$ transforms a colored $`\pi `$-tangle $`(T,g:\pi _1(C_T)\pi ,u)`$ into the colored $`\pi `$-tangle $`(T,\alpha g\alpha ^1:\pi _1(C_T)\pi ,{}_{}{}^{\alpha }u)`$ where $`{}_{}{}^{\alpha }u`$ is the coloring of the $`\pi `$-tangle $`(T,\alpha g\alpha ^1)`$ assigning to any path $`\gamma :[0,1]C_T`$ leading from the base point to $`\stackrel{~}{T}`$ the object $`\phi _\alpha (u_\gamma )𝒞_{\alpha g(\mu _\gamma )\alpha ^1}`$. This extends to arbitrary morphisms in $`𝒯`$ by linearity and yields an automorphism $`\phi _\alpha `$ of $`𝒯`$. In this way $`𝒯`$ becomes a crossed $`\pi `$-category.
For objects $`U=\{(\epsilon _r,\alpha _r,U_r)\}_{r=1}^k𝒯_\alpha `$ and $`V=\{(\epsilon ^s,\alpha ^s,U^s)\}_{s=1}^l𝒯_\beta `$ with $`\alpha ,\beta \pi `$, the braiding $`UV{}_{}{}^{U}VU`$ is represented by the same (framed oriented) braid $`T=T_{k,l}`$ on $`k+l`$ strings as in the standard theory. The braid $`T`$ is given by a diagram consisting of two families of parallel linear intervals: one family consists of $`k`$ intervals going from $`k`$ leftmost inputs to $`k`$ rightmost outputs, the second family consists of $`l`$ intervals going from $`l`$ rightmost inputs to $`l`$ leftmost outputs and meeting from below all the intervals of the first family. We extend $`T`$ uniquely to a colored $`\pi `$-braid with source
$$UV=((\epsilon _1,\alpha _1,U_1),\mathrm{},(\epsilon _k,\alpha _k,U_k),(\epsilon ^1,\alpha ^1,U^1),\mathrm{},(\epsilon ^l,\alpha ^l,U^l)).$$
It follows from definitions that the target of $`T`$ is then
$$((\epsilon ^1,\alpha \alpha ^1\alpha ^1,\phi _\alpha (U^1)),\mathrm{},(\epsilon ^l,\alpha \alpha ^l\alpha ^1,\phi _\alpha (U^l)),(\epsilon _1,\alpha _1,U_1),\mathrm{},(\epsilon _k,\alpha _k,U_k))$$
$$=\phi _\alpha (V)U={}_{}{}^{U}VU.$$
Hence $`T`$ represents a morphism $`UV{}_{}{}^{U}VU`$ in $`𝒯`$ denoted $`c_{U,V}`$. The axioms of a braiding are straightforward.
For $`U=\{(\epsilon _r,\alpha _r,U_r)\}_{r=1}^k𝒯_\alpha `$, the twist
$$\theta _U:U{}_{}{}^{U}U=\phi _\alpha (U)=\{(\epsilon _r,\alpha \alpha _r\alpha ^1,\phi _\alpha (U_r))\}_{r=1}^k$$
is represented by the same framed braid $`t_k`$ as in the standard theory. It can be obtained from the trivial braid on $`k`$ strings with constant framing by one full right-hand twist. We extend $`t_k`$ uniquely to a colored $`\pi `$-braid with source $`U`$. It is easy to deduce from definitions that the target of this colored $`\pi `$-braid is $`\phi _\alpha (U)`$ so that it represents a morphism $`U{}_{}{}^{U}U`$ in $`𝒯`$. The axioms of a twist are straightforward. Thus, $`𝒯`$ is a ribbon crossed $`\pi `$-category.
3.5. Ribbon functors. To formulate our main theorem concerning the category of colored $`\pi `$-tangles we need the notion of a ribbon $`\pi `$-functor. Consider two crossed $`\pi `$-categories $`𝒞,𝒞^{}`$. A monoidal $`\pi `$-functor from $`𝒞^{}`$ to $`𝒞`$ is a covariant functor $`F:𝒞^{}𝒞`$ such that
(i) $`F`$ is $`K`$-linear on morphisms;
(ii) $`F(\text{1}\text{ I}_𝒞^{})=\text{1}\text{ I}_𝒞`$ and $`F(fg)=F(f)F(g)`$ for any morphisms or objects $`f,g`$ in $`𝒞^{}`$;
(iii) $`F`$ transforms the structural morphisms $`a,l,r`$ in $`𝒞^{}`$ into the corresponding morphisms in $`𝒞`$: for any $`U,V,W𝒞^{}`$,
$$F(a_{U,V,W})=a_{F(U),F(V),F(W)},F(l_U)=l_{F(U)},F(r_U)=r_{F(U)};$$
(iv) $`F`$ maps $`𝒞_\alpha ^{}`$ into $`𝒞_\alpha `$ for all $`\alpha \pi `$;
(v) $`F`$ is equivariant with respect to the action of $`\pi `$, i.e., $`F\phi _\alpha =\phi _\alpha F`$ for any $`\alpha \pi `$.
Note that we do not require monoidal functors to preserve duality. If $`𝒞`$ and $`𝒞^{}`$ are strict, then condition (iii) is superfluous since the structural morphisms in question are the identities.
Assume now that $`𝒞,𝒞^{}`$ are ribbon $`\pi `$-categories. A monoidal $`\pi `$-functor $`F:𝒞^{}𝒞`$ is said to be ribbon if it transforms the braiding and twist in $`𝒞^{}`$ into the braiding and twist in $`𝒞`$, i.e., for any $`U,V𝒞^{}`$,
$$F(c_{U,V})=c_{F(U),F(V)},F(\theta _U)=\theta _{F(U)}.$$
3.6. Theorem. If $`𝒞`$ is a strict ribbon crossed $`\pi `$-category, then there is a unique ribbon monoidal $`\pi `$-functor $`F:𝒯=𝒯(\pi ,𝒞,K)𝒞`$ such that
(i) for any length 1 object $`(\epsilon ,\alpha ,U)`$ of $`𝒯`$, we have $`F((\epsilon ,\alpha ,U))=U^\epsilon `$ where
$$U^\epsilon =\{\begin{array}{cc}U𝒞_\alpha ,if\epsilon =+,\hfill & \\ U^{}𝒞_{\alpha ^1},if\epsilon =;\hfill & \end{array}$$
(ii) for any $`\alpha \pi ,U𝒞_\alpha `$,
$$F(b_{(+,\alpha ,U)})=b_U:\text{1}\text{ I}_𝒞UU^{}\text{a}ndF(d_{(+,\alpha ,U)})=d_U:U^{}U\text{1}\text{ I}_𝒞.$$
Theorem 3.6 generalizes the results of \[Tu2, Chapter I\] where $`\pi =1`$.
The unicity in Theorem 3.6 is quite straightforward. The assumption that $`F`$ is monoidal and condition (i) determine $`F`$ on all objects:
$$F((\epsilon _1,\alpha _1,U_1),\mathrm{},(\epsilon _k,\alpha _k,U_k))=\underset{r=1}{\overset{k}{}}(U_r)^{\epsilon _r}.$$
Next we observe that the morphisms of type
$$(c_{(\epsilon ,\alpha ,U),(\epsilon ^{},\alpha ^{},U^{})})^{\pm 1},(\theta _{(\epsilon ,\alpha ,U)})^{\pm 1},b_{(+,\alpha ,U)},d_{(+,\alpha ,U)}$$
$`(3.6.a)`$
(where $`\epsilon ,\epsilon ^{}=\pm ,\alpha ,\alpha ^{}\pi ,U𝒞_\alpha ,U^{}𝒞_\alpha ^{}`$) generate the category $`𝒯`$ in the sense that any morphism in $`𝒯`$ can be obtained from these generators by taking tensor product and composition. Given the value of $`F`$ on these generators, we can compute $`F`$ on any morphism in $`𝒯`$. This proves the uniqueness of $`F`$.
The values of $`F`$ on the colored $`\pi `$-tangles $`b_{(,\alpha ,U)},d_{(,\alpha ,U)}`$ with $`\alpha \pi ,U𝒞_\alpha `$ can be computed from the following equalities in $`𝒯`$:
$$d_{(,\alpha ,U)}=d_{(+,\alpha ,{}_{}{}^{U}U)}c_{(+,\alpha ,{}_{}{}^{U}U),(,\alpha ,U)}(\theta _{(+,\alpha ,U)}\text{id}_{(,\alpha ,U)}),$$
$`(3.6.b)`$
$$b_{(,\alpha ,U)}=(\text{id}_{(,\alpha ,U)}(\theta _{(+,\alpha ,U)})^1)(c_{(,\alpha ,U),(+,\alpha ,{}_{}{}^{U}U)})^1b_{(+,\alpha ,U)}.$$
$`(3.6.c)`$
A colored $`\pi `$-link $`(\mathrm{},z,g,u)`$ in $`^2\times ]0,1[`$ represents an endomorphism of the unit object $`\text{1}\text{ I}_𝒯`$ in $`𝒯`$ (the empty sequence) and is mapped by $`F`$ into $`F(\mathrm{},z,g,u)\text{End}_𝒞(\text{1}\text{ I}_𝒞)`$. It is obvious that any colored $`\pi `$-link in $`S^3=^3\{\mathrm{}\}`$ is ambiently isotopic to a unique (up to ambient isotopy) colored $`\pi `$-link in $`^2\times ]0,1[`$. This allows us to apply $`F`$ to colored $`\pi `$-links in $`S^3`$. For instance, an empty $`\pi `$-link $`\mathrm{}`$ represents the identity endomorphism of $`\text{1}\text{ I}_𝒯`$ and therefore $`F(\mathrm{})=1\text{End}_𝒞(\text{1}\text{ I}_𝒞)`$.
In Section 4 we generalize Theorem 3.6 to colored $`\pi `$-graphs and briefly discuss a proof of these theorems.
4. Colored $`\pi `$-graphs and their invariants
The standard theory of colored tangles generalizes to so-called ribbon graphs. They are instrumental in the construction of 3-dimensional TQFT’s from modular categories, see \[Tu2\]. Here we discuss a similar generalization of colored $`\pi `$-tangles.
4.1. Colored $`\pi `$-graphs. We recall the definition of a ribbon graph referring to \[Tu2\] for details. A coupon is a rectangle with a distinguished side called the bottom base; the opposite side is called the top base. A ribbon graph $`\mathrm{\Omega }`$ in $`^2\times [0,1]`$ with $`k0`$ inputs $`\{(r,0,0)\}_{r=1}^k`$ and $`l0`$ outputs $`\{(s,0,1)\}_{s=1}^l`$ consists of a finite family of arcs, circles and coupons embedded in $`^2\times [0,1]`$. We call these arcs, circles and coupons the strata of $`\mathrm{\Omega }`$. The inputs and outputs of $`\mathrm{\Omega }`$ should be among the endpoints of the arcs, all the other endpoints of the arcs should lie on the bases of the coupons. Otherwise the strata of $`\mathrm{\Omega }`$ should be disjoint. At the inputs and outputs, the arcs should be orthogonal to the planes $`^2\times 0,^2\times 1`$. All the strata of $`\mathrm{\Omega }`$ should be oriented and framed so that their framings form a continuous non-singular vector field on $`\mathrm{\Omega }`$ transversal to $`\mathrm{\Omega }`$. This vector field is called the framing of $`\mathrm{\Omega }`$. At the inputs and outputs, the framing should be given by the vector $`(0,1,0)`$. On each coupon the framing should be transversal to the coupon and yield together with the orientation of the coupon the right-handed orientation of $`^2\times [0,1]`$. Slightly pushing $`\mathrm{\Omega }`$ along its framing we obtain a disjoint copy $`\stackrel{~}{\mathrm{\Omega }}`$ of $`\mathrm{\Omega }`$. Pushing an arc (resp. a circle, a coupon) $`t`$ of $`\mathrm{\Omega }`$ along the framing we obtain an arc (resp. a circle, a coupon) $`\stackrel{~}{t}\stackrel{~}{\mathrm{\Omega }}`$. We use here a language slightly different but equivalent to the one in \[Tu2\], where instead of the framings on arcs and circles we considered orthogonal 2-dimensional bands and annuli.
Let $`\mathrm{\Omega }^2\times [0,1]`$ be a ribbon graph. We provide its complement $`C_\mathrm{\Omega }=(^2\times [0,1])\backslash \mathrm{\Omega }`$ with the “canonical” base point $`z`$ with a big negative second coordinate $`z_20`$. As in Section 3, we can supress the base point from the notation for the fundamental group of $`C_\mathrm{\Omega }`$.
We define meridians of an arc or a circle of $`\mathrm{\Omega }`$ exactly as in Sections 3.1 and 3.2. With the inputs and outputs of $`\mathrm{\Omega }`$ we associate canonical meridians as in Section 3.2. We also define meridians of a coupon $`Q\mathrm{\Omega }`$ as follows. Choose an oriented interval $`qQ`$ leading from the top base of $`Q`$ to its bottom base. For a path $`\gamma :[0,1]C_\mathrm{\Omega }`$ connecting the base point to a point $`\gamma (1)\stackrel{~}{Q}`$, the meridian $`\mu _\gamma \pi _1(C_\mathrm{\Omega })`$ is represented by the loop $`\gamma m\gamma ^1`$ where $`mC_\mathrm{\Omega }`$ is a small loop encircling $`Q`$ (in a plane transversal to $`q`$) and having linking number $`+1`$ with $`q`$.
A $`\pi `$-graph is a ribbon graph $`\mathrm{\Omega }^2\times [0,1]`$ endowed with a homomorphism $`g:\pi _1(C_\mathrm{\Omega })\pi `$. Clearly, a $`\pi `$-graph without coupons is nothing but a $`\pi `$-tangle.
Fix a crossed $`\pi `$-category $`(𝒞,\phi :\pi \text{Aut}(𝒞))`$. A coloring of a $`\pi `$-graph $`(\mathrm{\Omega },g:\pi _1(C_\mathrm{\Omega })\pi )`$ consists of two functions $`u`$ and $`v`$. The function $`u`$ assigns to every arc or circle $`t`$ of $`\mathrm{\Omega }`$ and to every path $`\gamma :[0,1]C_\mathrm{\Omega }`$ connecting the base point $`\gamma (0)=zC_\mathrm{\Omega }`$ to a point $`\gamma (1)\stackrel{~}{t}\stackrel{~}{\mathrm{\Omega }}`$ a certain object $`u_\gamma 𝒞_{g(\mu _\gamma )}`$ where $`\mu _\gamma \pi _1(C_\mathrm{\Omega })`$ is the meridian of $`t`$ determined by $`\gamma `$. This function should satisfy the same conditions as in Section 3, i.e.,
(i) $`u_\gamma `$ is preserved under homotopies of $`\gamma `$ fixing $`\gamma (0)`$ and keeping $`\gamma (1)`$ on $`\stackrel{~}{t}`$;
(ii) if $`\beta `$ is a loop in $`(C_\mathrm{\Omega },z)`$ then $`u_{\beta \gamma }=\phi _{g([\beta ])}(u_\gamma )`$.
The function $`v`$ assigns to every coupon $`Q\mathrm{\Omega }`$ and every path $`\gamma `$ in $`C_\mathrm{\Omega }`$ connecting the base point $`\gamma (0)=zC_\mathrm{\Omega }`$ to a point $`\gamma (1)\stackrel{~}{Q}\stackrel{~}{\mathrm{\Omega }}`$ a certain morphism $`v_\gamma `$ in $`𝒞_{g(\mu _\gamma )}`$ satisfying three conditions:
(i) $`v_\gamma `$ is preserved under homotopies of $`\gamma `$ fixing $`\gamma (0)`$ and keeping $`\gamma (1)`$ on $`\stackrel{~}{Q}`$;
(ii) if $`\beta `$ is a loop in $`(C_\mathrm{\Omega },z)`$ then $`v_{\beta \gamma }=\phi _{g([\beta ])}(v_\gamma )`$.
To formulate the third condition on $`v_\gamma `$ we need some preparations. Let $`t_1,\mathrm{},t_m`$ be the arcs of $`\mathrm{\Omega }`$ incident to the bottom side of $`Q`$ in the order determined by the direction of this side induced by the orientation of $`Q`$. Set $`\epsilon _i=+1`$ if $`t_i`$ is directed out of $`Q`$ and $`\epsilon _i=1`$ otherwise. Let $`t^1,\mathrm{},t^n`$ be the arcs of $`\mathrm{\Omega }`$ incident to the top side of $`Q`$ in the order determined by the direction of this side opposite to the one induced by the orientation of $`Q`$. Set $`\epsilon ^j=1`$ if $`t^j`$ is directed out of $`Q`$ and $`\epsilon ^j=+1`$ otherwise. For $`i=1,\mathrm{},m`$, we can compose the path $`\gamma `$ with a path in $`\stackrel{~}{Q}`$ leading from $`\gamma (1)\stackrel{~}{Q}`$ to the endpoint of $`\stackrel{~}{t}_i`$ lying on the bottom side of $`\stackrel{~}{Q}`$. The resulting path, $`\gamma _i`$, leads from the base point of $`C_\mathrm{\Omega }`$ to a point of $`\stackrel{~}{t}_i`$. Similarly, for $`j=1,\mathrm{},n`$, we can compose $`\gamma `$ with a path in $`\stackrel{~}{Q}`$ leading from $`\gamma (1)`$ to the endpoint of $`\stackrel{~}{t}^j`$ lying on the top side of $`\stackrel{~}{Q}`$. The resulting path, $`\gamma ^j`$, leads from the base point of $`C_\mathrm{\Omega }`$ to a point of $`\stackrel{~}{t}^j`$. Note that
$$\mu _\gamma =\underset{i=1}{\overset{m}{}}(\mu _{\gamma _i})^{\epsilon _i}=\underset{j=1}{\overset{n}{}}(\mu _{\gamma ^j})^{\epsilon ^j}\pi _1(C_\mathrm{\Omega }).$$
Now we can state the third condition on $`v`$:
(iii) for any coupon $`Q`$ of $`\mathrm{\Omega }`$ and any path $`\gamma `$ in $`C_\mathrm{\Omega }`$ connecting the base point to a point of $`\stackrel{~}{Q}`$,
$$v_\gamma \text{Hom}_𝒞(\underset{i=1}{\overset{m}{}}(u_{\gamma _i})^{\epsilon _i},\underset{j=1}{\overset{n}{}}(u_{\gamma ^j})^{\epsilon ^j})$$
where $`u_{\gamma _i}𝒞_{g(\mu _{\gamma _i})}`$ and $`u_{\gamma ^j}𝒞_{g(\mu _{\gamma ^j})}`$ are the objects provided by the function $`u`$.
A $`\pi `$-graph endowed with a coloring is said to be colored. The definition of the source and the target given in Section 3.2 applies to colored $`\pi `$-graphs word for word. The technique of diagrams discussed in Section 3.3 generalizes to colored $`\pi `$-graphs in the obvious way. The coupons should be presented by small rectangles in $`\times [0,1]`$ with horizontal bottom and top bases such that the bottom base has smaller second coordinate than the top base and the orientation of the coupon is counter-clockwise.
We define a category of colored $`\pi `$-graphs $`𝒢=𝒢(\pi ,𝒞,K)`$ following the lines of Section 3.4. This category has the same objects as $`𝒯(\pi ,𝒞,K)`$. The morphisms in $`𝒢`$ are formal linear combinations over $`K`$ of (the isotopy classes of) colored $`\pi `$-graphs with the given source and target. The composition and tensor product are defined exactly as in $`𝒯`$. It is clear that $`𝒯`$ is a monoidal subcategory of $`𝒢`$.
4.2. Remark. Sometimes it is convenient to formulate the notions of a $`\pi `$-graph and a colored $`\pi `$-graph using the graph exteriors. The exterior $`E_\mathrm{\Omega }`$ of a ribbon graph $`\mathrm{\Omega }^2\times [0,1]`$ is the complement in $`^2\times [0,1]`$ of an open regular neighborhood of $`\mathrm{\Omega }`$. We choose this neighborhood so that $`\stackrel{~}{\mathrm{\Omega }}E_\mathrm{\Omega }`$. It is clear that $`E_\mathrm{\Omega }C_\mathrm{\Omega }`$ is a deformation retract of $`C_\mathrm{\Omega }`$ so that $`\pi _1(E_\mathrm{\Omega })=\pi _1(C_\mathrm{\Omega })`$. Repeating the definitions of Section 4.1 but considering only loops and paths in $`E_\mathrm{\Omega }`$ we obtain an equivalent definition of a colored $`\pi `$-graph in terms of the exterior.
4.3. Lemma. The category $`𝒢`$ is a strict ribbon crossed $`\pi `$-category.
The proof follows the lines of the proof of Lemma 3.4.1. In particular, given $`\alpha \pi `$, the category $`𝒢_\alpha `$ is defined as the full subcategory of $`𝒢`$ whose objects are sequences $`((\epsilon _1,\alpha _1,U_1),\mathrm{},(\epsilon _k,\alpha _k,U_k))`$ with $`_{r=1}^k(\alpha _r)^{\epsilon _r}=\alpha `$. The action of $`\alpha \pi `$ on $`𝒢`$ is obtained by applying $`\phi _\alpha `$ to the objects and morphisms forming the colorings of $`\pi `$-graphs. The braiding, twist and duality in $`𝒢`$ come from the corresponding morphisms in $`𝒯`$ via the inclusion $`𝒯𝒢`$.
4.4. Elementary colored $`\pi `$-graphs. Let $`V=((\epsilon _1,\alpha _1,V_1),\mathrm{},(\epsilon _k,\alpha _k,V_k))`$ and $`V^{}=((\epsilon ^1,\alpha ^1,V^1),\mathrm{},(\epsilon ^l,\alpha ^l,V^l))`$ be objects of $`𝒢_\alpha `$ with $`\alpha \pi `$. Let
$$f\text{Hom}_𝒞(\underset{i=1}{\overset{m}{}}(V_i)^{\epsilon _i},\underset{j=1}{\overset{n}{}}(V^j)^{\epsilon ^j}).$$
With this data we associate a colored $`\pi `$-graph $`\mathrm{\Omega }=\mathrm{\Omega }(V,V^{},f)`$ as follows. Geometrically, $`\mathrm{\Omega }`$ consists of one coupon, $`k`$ short vertical intervals attached to its bottom base and $`l`$ short vertical intervals attached to its top base. The coupon and the intervals lie in $`\times [0,1]`$, the framing is constant and orthogonal to $`\times [0,1]`$. The orientation of the bottom (resp. top) intervals is determined by the signs $`\epsilon _1,\mathrm{},\epsilon _k`$ (resp. $`\epsilon ^1,\mathrm{},\epsilon ^l`$). Note that the group $`\pi _1(C_\mathrm{\Omega })`$ is generated by the canonical meridians $`\mu _1,\mathrm{},\mu _k,\mu ^1,\mathrm{},\mu ^l\pi _1(C_\mathrm{\Omega })`$ corresponding to the inputs and outputs of $`\mathrm{\Omega }`$. They are subject to only one relation
$$(\mu _1)^{\epsilon _1}\mathrm{}(\mu _k)^{\epsilon _k}=(\mu ^1)^{\epsilon ^1}\mathrm{}(\mu ^l)^{\epsilon ^l}.$$
$`(4.4.a)`$
Therefore the formulas $`g(\mu _r)=\alpha _r,g(\mu ^s)=\alpha ^s`$ where $`r=1,\mathrm{},k`$ and $`s=1,\mathrm{},l`$ define a group homomorphism $`g:\pi _1(C_\mathrm{\Omega })\pi `$ sending both sides of (4.4.a) into $`\alpha `$. We define $`\mathrm{\Omega }(V,V^{},f)`$ to be the $`\pi `$-graph $`(\mathrm{\Omega },g)`$ endowed with the coloring $`(u,v)`$ such that
(i) the source of $`\mathrm{\Omega }(V,V^{},f)`$ is $`V`$ and the target of $`\mathrm{\Omega }(V,V^{},f)`$ is $`V^{}`$;
(ii) for the linear path $`\gamma `$ in $`C_\mathrm{\Omega }`$ leading from the base point to the point lying slightly above $`\mathrm{\Omega }`$, we have $`v_\gamma =f`$.
Such a coloring of $`\mathrm{\Omega }`$ exists and is unique. We call $`\mathrm{\Omega }(V,V^{},f)`$ the elementary colored $`\pi `$-graph associated with $`V,V^{},f`$.
4.5. Theorem. If $`𝒞`$ is a strict ribbon crossed $`\pi `$-category, then there is a unique ribbon monoidal $`\pi `$-functor $`F:𝒢=𝒢(\pi ,𝒞,K)𝒞`$ extending the functor $`𝒯(\pi ,𝒞,K)𝒞`$ of Theorem 3.6 and such that for any elementary colored $`\pi `$-graph $`\mathrm{\Omega }(V,V^{},f)`$ as above we have $`F(\mathrm{\Omega }(V,V^{},f))=f`$.
Theorem 4.5 generalizes Theorem I.2.5 in \[Tu2\] where $`\pi =1`$.
The unicity in Theorem 4.5 is straightforward: it suffices to observe that the morphisms (3.6.a) and the elementary colored $`\pi `$-graphs generate $`𝒢`$. The proof of the existence follows the proof of \[Tu2, Theorem I.2.5\] with natural changes. The key role in the proof given in \[Tu2\] is played by certain commutative diagrams in the target category $`𝒞`$. We give here versions of these diagrams in our setting: for any objects $`V,W𝒞`$ the following two diagrams are commutative
$$\begin{array}{ccccc}V=V\text{1}\text{ I}& \stackrel{\text{id}_Vb_W}{}& VWW^{}& \stackrel{c_{V,W}\text{id}_W^{}}{}& {}_{}{}^{V}WVW^{}\\ \text{c}_{V,\text{1}\text{ I}}=id_V& & c_{V,WW^{}}& & \text{id}_{\left({}_{}{}^{V}W\right)}c_{V,W^{}}& & \\ V=\text{1}\text{ I}V& \stackrel{b_{({}_{}{}^{V}W)}\text{id}_V}{}& {}_{}{}^{V}W{}_{}{}^{V}(W^{})V& =& {}_{}{}^{V}W{}_{}{}^{V}(W^{})V,\end{array}$$
$$\begin{array}{ccccc}{}_{}{}^{V}(W^{})VW& \stackrel{c_{V,W^{}}\text{id}_W}{}& VW^{}W& \stackrel{\text{id}_Vd_W}{}& V\text{1}\text{ I}=V\\ \text{id}_{{}_{}{}^{V}\left(W^{}\right)}c_{V,W}& & c_{V,W^{}W}& & \text{c}_{V,\text{1}\text{ I}}=id_V& & \\ {}_{}{}^{V}(W^{}){}_{}{}^{V}WV& =& {}_{}{}^{V}(W^{}){}_{}{}^{V}WV& \stackrel{d_{({}_{}{}^{V}W)}\text{id}_V}{}& \text{1}\text{ I}V=V.\end{array}$$
One should also use two similar diagrams where the middle vertical arrow is replaced with $`c_{W^{}W,V}:W^{}WVVW^{}W`$ and also the diagrams obtained by replacing $`V,W`$ with $`W^{},V`$, respectively.
4.6. Corollary. For any object $`U`$ of a strict ribbon crossed $`\pi `$-category $`𝒞`$, there is a canonical isomorphism $`\eta _U:UU^{}=(U^{})^{}`$ in $`𝒞`$. For any morphism $`f:UV`$ in $`𝒞`$, the following diagram is commutative:
$$\begin{array}{ccc}U& \stackrel{\eta _U}{}& U^{}\\ f& & f^{}& & \\ V& \stackrel{\eta _V}{}& V^{}.\end{array}$$
Proof. If $`U𝒞_\alpha `$ with $`\alpha \pi `$ then the morphisms
$$\eta _U=(F(d_{(,\alpha ,U)})\text{id}_U^{})(\text{id}_Ub_U^{}):UU^{},$$
$$(d_U^{}\text{id}_U)(\text{id}_U^{}F(b_{(,\alpha ,U)})):U^{}U$$
are mutually inverse isomorphisms. This can be deduced directly from definitions using (3.4.a). A pictorial proof uses Theorem 4.5 and followes the lines of the proof of Corollary I.2.6.1 in \[Tu2\]. The second claim is an exercise.
4.7. Properties of $`F`$. We discuss here three operations on a colored $`\pi `$-graph $`(\mathrm{\Omega },g,u,v)`$ preserving $`F(\mathrm{\Omega },g,u,v)`$. Fix a circle stratum $`\mathrm{}`$ of $`\mathrm{\Omega }`$.
1. The invariant $`F(\mathrm{\Omega },g,u,v)`$ does not change when we replace the colors of the paths leading to $`\mathrm{}`$ by isomorphic colors. More precisely, fix a path $`\gamma `$ leading from the canonical base point to the longitude $`\stackrel{~}{\mathrm{}}`$ of $`\mathrm{}`$. Let $`V`$ be an object of the category $`𝒞_{g(\mu _\gamma )}`$ isomorphic to $`u_\gamma 𝒞_{g(\mu _\gamma )}`$. We define a new coloring $`(u^{},v)`$ of $`\mathrm{\Omega }`$ as follows. For the paths leading from the base point to the strata of $`\mathrm{\Omega }`$ distinct from $`\mathrm{}`$, the associated colors are the same as in $`(u,v)`$. Set $`u_\gamma ^{}=V`$. This data extends uniquely to a coloring $`(u^{},v)`$ of $`(\mathrm{\Omega },g)`$. Then $`F(\mathrm{\Omega },g,u,v)=F(\mathrm{\Omega },g,u^{},v)`$. To prove this equality observe that $`F(\mathrm{\Omega },g,u,v)`$ does not change when we insert into $`\mathrm{}`$ two coupons whose colors corresponding to $`\gamma `$ are mutually inverse isomorphisms $`u_\gamma V`$ and $`Vu_\gamma `$. Pushing one of the coupons along $`\mathrm{}`$ and eventually cancelling it with the second coupon we obtain $`F(\mathrm{\Omega },g,u^{},v)`$.
2. The invariant $`F(\mathrm{\Omega },g,u,v)`$ does not change when we invert the orientation of $`\mathrm{}`$ and simultaneously replace the colors associated with $`\mathrm{}`$ by dual objects. The latter means that for any path $`\gamma `$ leading from the base point to $`\stackrel{~}{\mathrm{}}`$, we set $`u_\gamma ^{}=(u_\gamma )^{}`$. For the paths leading to other strata of $`\mathrm{\Omega }`$, the color is preserved. This yields a coloring $`(u^{},v)`$ of the $`\pi `$-graph $`(\mathrm{\Omega }^{},g)`$ obtained from $`(\mathrm{\Omega },g)`$ by inverting the orientation of $`\mathrm{}`$. Then $`F(\mathrm{\Omega },g,u,v)=F(\mathrm{\Omega }^{},g,u^{},v)`$. The proof follows the lines of the proof of \[Tu2, Corollary I.2.8.1\].
3. Let $`\mathrm{}_1,\mathrm{}_2`$ be two parallel copies of $`\mathrm{}`$ determined by the framing and going very closely to $`\mathrm{}`$. Consider the ribbon graph $`\mathrm{\Omega }^{}=(\mathrm{\Omega }\backslash \mathrm{})\mathrm{}_1\mathrm{}_2`$. To describe the relevant colorings of $`\mathrm{\Omega }^{}`$ we use the language of exteriors. The exterior $`E^{}=E_\mathrm{\Omega }^{}`$ of $`\mathrm{\Omega }^{}`$ can be obtained from the exterior $`E=E_\mathrm{\Omega }`$ of $`\mathrm{\Omega }`$ by gluing (2-disc with two holes) $`\times S^1`$ along the 2-torus in $`E`$ corresponding to $`\mathrm{}`$. Denote the inclusion homomorphism $`\pi _1(E)\pi _1(E^{})`$ by $`h`$. Assume that there are a homomorphism $`g^{}:\pi _1(E^{})\pi `$ and a coloring $`(u^{},v^{})`$ of the $`\pi `$-graph $`(\mathrm{\Omega }^{},g^{})`$ satisfying the following conditions:
(i) $`g=g^{}h:\pi _1(E)\pi `$;
(ii) on the paths leading from the base point to all strata of $`\mathrm{\Omega }\backslash \mathrm{}`$ the colorings $`(u,v)`$ and $`(u^{},v^{})`$ coincide;
(iii) if $`\gamma `$ is a path in $`E`$ leading from the base point to $`\stackrel{~}{\mathrm{}}E`$, then composing $`\gamma `$ with short paths in $`E^{}\backslash E`$ we obtain paths $`\gamma _1,\gamma _2`$ in $`E^{}`$ leading to $`\stackrel{~}{\mathrm{}}_1,\stackrel{~}{\mathrm{}}_2`$, respectively, such that $`h(\mu _\gamma )=\mu _{\gamma _1}\mu _{\gamma _2}`$ and $`u_\gamma =u_{\gamma _1}^{}u_{\gamma _2}^{}`$ where $`u_{\gamma _i}^{}𝒞_{g^{}(\mu _{\gamma _i})}`$ for $`i=1,2`$.
We say then that the colored $`\pi `$-graph $`(\mathrm{\Omega }^{},g^{},u^{},v^{})`$ is obtained from $`(\mathrm{\Omega },g,u,v)`$ by doubling of $`\mathrm{}`$. It is easy to check that $`F(\mathrm{\Omega },g,u,v)=F(\mathrm{\Omega }^{},g^{},u^{},v^{})`$, cf. \[Tu2, Corollary I.2.8.3\].
5. Trace, dimension, and algebra of colors
We use Theorem 4.5 to define the trace, dimension, and algebra of colors associated with a ribbon crossed $`\pi `$-category $`𝒞`$.
5.1. Trace. We define the trace of an endomorphism $`f:UU`$ of an object $`U𝒞`$ by
$$\text{tr}(f)=d_{({}_{}{}^{U}U)}c_{({}_{}{}^{U}U),U^{}}(\theta _Uf\text{id}_U^{})b_U\text{End}_𝒞(\text{1}\text{ I})=\text{Hom}_𝒞(\text{1}\text{ I},\text{1}\text{ I}).$$
It is clear that for any $`kK`$, we have $`\text{tr}(kf)=k\text{tr}(f)`$. For any $`\beta \pi `$, we have $`\text{tr}(\phi _\beta (f))=\phi _\beta (\text{tr}(f))`$ where on the right hand side $`\phi _\beta `$ acts on $`\text{End}_𝒞(\text{1}\text{ I})`$.
If $`U𝒞_\alpha `$ with $`\alpha \pi `$ then we can rewrite the definition of $`\text{tr}(f:UU)`$ using (3.6.b):
$$\text{tr}(f)=F(d_{(,\alpha ,U)})(f\text{id}_U^{})b_U\text{End}_𝒞(\text{1}\text{ I}).$$
Theorem 4.5 gives a geometric interpretation of $`\text{tr}(f)`$. Consider the elementary colored $`\pi `$-graph $`\mathrm{\Omega }_f=\mathrm{\Omega }((+,\alpha ,U),(+,\alpha ,U),f)`$ with one input and one output. As in the standard theory we can close $`\mathrm{\Omega }_f`$ to obtain a colored $`\pi `$-graph $`\widehat{\mathrm{\Omega }}_f`$ with 0 inputs and 0 outputs. A diagram of $`\widehat{\mathrm{\Omega }}_f`$ is obtained from any (labeled) diagram of $`\mathrm{\Omega }_f`$ by connecting the top and bottom endpoints by an arc disjoint from the rest of the diagram. More formally,
$$\widehat{\mathrm{\Omega }}_f=d_{(,\alpha ,U)}(\mathrm{\Omega }_f\text{id}_{(,\alpha ,U)})b_{(+,\alpha ,U)}.$$
Hence,
$$F(\widehat{\mathrm{\Omega }}_f)=F(d_{(,\alpha ,U)})(f\text{id}_U^{})b_U=\text{tr}(f).$$
Using this geometric interpretation and following the lines of \[Tu2, Section I.2.7\] we obtain the following lemma.
5.1.1. Lemma. (i) For any morphisms $`f:UV`$, $`g:VU`$ in $`𝒞`$, we have $`\text{tr}(fg)=\text{tr}(gf);`$
(ii) for any endomorphisms $`f,g`$ of objects of $`𝒞`$, we have $`\text{tr}(f^{})=\text{tr}(f)`$ and $`\text{tr}(fg)=\text{tr}(f)\text{tr}(g).`$
5.2. Dimension. We define the dimension of an object $`U𝒞`$ by
$$\text{dim}(U)=\text{tr}(\text{id}_U)=d_{({}_{}{}^{U}U)}c_{({}_{}{}^{U}U),U^{}}(\theta _U\text{id}_U^{})b_U\text{End}_𝒞(\text{1}\text{ I}).$$
If $`U𝒞_\alpha `$, then $`\text{dim}(U)=F(d_{(,\alpha ,U)})b_{(+,\alpha ,U)}`$ is the value of $`F`$ on a colored $`\pi `$-knot represented by a diagram consisting of an embedded oriented circle labeled with $`(\alpha ,U)`$. This knot is an (oriented) unknot $`KS^3`$ endowed with homomorphism $`\pi _1(C_K)=\pi `$ sending the meridian to $`\alpha `$.
The properties of $`F`$ established in Section 4.7 (or Lemma 5.1.1) imply that isomorphic objects have equal dimensions and for any objects $`U,V`$ and $`\beta \pi `$, we have $`\text{dim}(U^{})=\text{dim}(U)`$, $`\text{dim}(\phi _\beta (U))=\phi _\beta (\text{dim}(U))`$ and $`\text{dim}(UV)=\text{dim}(U)\text{dim}(V)`$.
Note that for morphisms and objects of the neutral component $`𝒞_1𝒞`$, the definitions above coincide with the standard definition of the trace and dimension in a ribbon category. This implies that for any $`f\text{End}_𝒞(\text{1}\text{ I})`$, we have $`\text{tr}(f)=f`$. In particular, $`\text{dim}(\text{1}\text{ I}_𝒞)=\text{tr}(\text{id}_{\text{1}\text{ I}})=\text{id}_{\text{1}\text{ I}}`$.
5.3. Algebra of colors. We define an algebra of colors or Verlinde algebra $`L=L(𝒞)`$ of the ribbon crossed $`\pi `$-category $`𝒞`$ as follows. Consider the $`K`$-module $`_{U𝒞}\text{End}_𝒞(U)`$ where $`U`$ runs over all objects of $`𝒞`$. The additive generator of this module represented by $`f\text{End}_𝒞(U)`$ will be denoted by $`U,f`$ or briefly by $`f`$. We factorize this module by the following relations: for any morphisms $`f:UV`$, $`g:VU`$ in $`𝒞`$,
$$V,fg=U,gf.$$
$`(5.3.a)`$
Denote the quotient $`K`$-module by $`L`$. We provide $`L`$ with multiplication by the formula $`ff^{}=ff^{}`$. Clearly, $`L`$ is an associative $`K`$-algebra with unit $`\text{id}_{\text{1}\text{ I}}L`$. Every object $`U𝒞`$ determines an element $`U=\text{id}_UL`$.
The algebra $`L`$ is $`\pi `$-graded: $`L=_{\alpha \pi }L_\alpha `$ where $`L_\alpha `$ is the submodule of $`L`$ additively generated by elements $`U,f`$ with $`U𝒞_\alpha `$. We have $`L_\alpha L_\beta L_{\alpha \beta }`$ for all $`\alpha ,\beta \pi `$. The formula $`\phi _\alpha (f)=\phi _\alpha (f)`$ defines an action of $`\pi `$ on $`L`$ by algebra endomorphisms. Clearly, $`\phi _\alpha (L_\beta )=L_{\alpha \beta \alpha ^1}`$ for all $`\alpha ,\beta \pi `$. The existence of the braiding implies that $`ab=\phi _\alpha (b)a`$ for any $`aL_\alpha ,bL_\beta `$. The existence of the twist implies that $`\phi _\alpha |_{L_\alpha }=\text{id}`$ for any $`\alpha \pi `$.
The trace of morphisms in $`𝒞`$ defines a $`K`$-linear homomorphism $`L\text{End}_𝒞(\text{1}\text{ I})`$ sending any generator $`f`$ as above to $`\text{tr}(f)\text{End}_𝒞(\text{1}\text{ I})`$. We denote this functional by dim. In particular, $`\text{dim}(U)=\text{tr}(\text{id}_U)=\text{dim}(U)`$ for any $`U𝒞`$. It follows from the properties of the trace that dim is an algebra homomorphism.
Sending a generator $`U,f`$ to the generator $`U^{},f^{}:U^{}U^{}`$ we define a $`K`$-linear homomorphism $`LL`$ denoted by $``$. It follows from definitions that $``$ is an involutive anti-automorphism of the algebra $`L`$ sending each $`L_\alpha `$ onto $`L_{\alpha ^1}`$ and commuting with $`\text{dim}:L\text{End}_𝒞(\text{1}\text{ I})`$.
5.4. Special $`\pi `$-links. The importance of the algebra of colors $`L=L(𝒞)`$ is due to the fact that its elements can be used to color so-called special $`\pi `$-links. Let us call a $`\pi `$-link $`(\mathrm{},z,g)`$ special if the longitudes of all the components of $`\mathrm{}`$ are sent by $`g`$ to $`1\pi `$. A generalized coloring of a $`\pi `$-link $`(\mathrm{},z,g)`$ is a function which assigns to every path $`\gamma :[0,1]C_{\mathrm{}}`$ with $`\gamma (0)=z,\gamma (1)\stackrel{~}{\mathrm{}}`$ an element $`a_\gamma L_{g(\mu _\gamma )}`$ such that
(i) $`a_\gamma `$ is preserved under homotopies of $`\gamma `$ fixing $`\gamma (0)`$ and keeping $`\gamma (1)`$ on $`\stackrel{~}{\mathrm{}}`$;
(ii) if $`\beta `$ is a loop in $`(C_{\mathrm{}},z)`$, then $`a_{\beta \gamma }=\phi _{g([\beta ])}(a_\gamma )`$.
For instance, a coloring $`u`$ of $`(\mathrm{},z,g)`$ as defined in Section 3.1 gives rise to a generalized coloring by $`a_\gamma =u_\gamma `$.
The invariant $`F`$ extends to generalized colorings $`a`$ of a special $`\pi `$-link $`(\mathrm{}=\mathrm{}_1\mathrm{}\mathrm{}_n,z,g)`$. To define this extension, fix a path $`\gamma _i`$ leading from $`z`$ to a point of $`\stackrel{~}{\mathrm{}}_i`$ for $`i=1,\mathrm{},n`$. We transform $`\mathrm{}`$ into a $`\pi `$-graph $`\mathrm{\Omega }`$ by inserting one coupon into each component $`\mathrm{}_i`$ near $`\gamma _i(1)`$. If $`a_{\gamma _i}=U_i,f_iL_{g(\mu _{\gamma _i})}`$ for $`i=1,\mathrm{},n`$ then we set $`u_{\gamma _i}=U_i,v_{\gamma _i}=f_i`$ for $`i=1,\mathrm{},n`$. This extends uniquely to a coloring $`(u,v)`$ of $`\mathrm{\Omega }`$ (it is here that we need $`\mathrm{}`$ to be special, cf. Lemma 3.2.1). Set $`F(\mathrm{},z,g,a)=F(\mathrm{\Omega },g,u,v)`$. This does not depend on the choice of $`\gamma _i`$ and extends to arbitrary $`a`$ by additivity. Note that if $`a_{\gamma _i}=U_i`$ then there is no need to introduce the coupon on $`\mathrm{}_i`$; it suffices to keep $`\mathrm{}_i`$ as a stratum and to set $`u_{\gamma _i}=U_i`$. This gives the same invariant $`F`$.
Similarly, we call a $`\pi `$-graph special if the longitudes of all its circle strata are sent to $`1\pi `$. The elements of $`L`$ can be used to color circle strata of special $`\pi `$-graphs. The invariant $`F`$ extends to such generalized colorings of $`\pi `$-graphs exactly as above. The properties of $`F`$ established in Section 4.7 extend to the generalized colorings in the obvious way.
6. Modular crossed $`\pi `$-categories
6.1. Modular crossed $`\pi `$-categories. Let $`𝒞`$ be a crossed $`\pi `$-category. An object $`V`$ of $`𝒞`$ is said to be simple if $`\text{End}_𝒞(V)=K\text{id}_V`$. It is clear that an object isomorphic or dual to a simple object is itself simple. The action of $`\pi `$ on $`𝒞`$ transforms simple objects into simple objects.
We say that an object $`U`$ of $`𝒞`$ is dominated by simple objects if there exist a finite set of simple objects $`\{V_r\}_r`$ of $`𝒞`$ (possibly with repetitions) and morphisms $`\{f_r:V_rU,g_r:UV_r\}_r`$ such that $`\text{id}_U=_rf_rg_r`$. Clearly, if $`U𝒞_\alpha `$ then without loss of generality we can assume that $`V_r𝒞_\alpha `$ for all $`r`$.
We say that a ribbon crossed $`\pi `$-category $`𝒞`$ is modular if it satisfies the following five axioms:
(6.1.1) the unit object $`\text{1}\text{ I}_𝒞`$ is simple;
(6.1.2) for each $`\alpha \pi `$, the set $`I_\alpha `$ of the isomorphism classes of simple objects of $`𝒞_\alpha `$ is finite;
(6.1.3) for each $`\alpha \pi `$, any object of $`𝒞_\alpha `$ is dominated by simple objects of $`𝒞_\alpha `$;
(6.1.4) if $`V,W`$ are non-isomorphic simple objects of $`𝒞`$ then $`\text{Hom}_𝒞(V,W)=0`$;
To formulate the last axiom we need some notation. For $`i,jI_1`$, choose simple objects $`V_i,V_j𝒞_1`$ representing $`i,j`$, respectively, and set
$$S_{i,j}=\text{tr}(c_{V_j,V_i}c_{V_i,V_j}:V_iV_jV_jV_i)\text{End}_𝒞(\text{1}\text{ I})=K.$$
It follows from the properties of the trace that $`S_{i,j}`$ does not depend on the choice of $`V_i,V_j`$.
(6.1.5). The square matrix $`S=[S_{i,j}]_{i,jI_1}`$ is invertible over $`K`$.
These axioms generalize the axioms of a modular ribbon category given in \[Tu2\] where $`\pi =1`$.
It follows from axioms (6.1.1)-(6.1.5) that the neutral component $`𝒞_1`$ of $`𝒞`$ is a modular category in the sense of \[Tu2\]. (Note that (6.1.5) involves only $`𝒞_1`$.) Recall the following property of modular categories: if $`U,V`$ are objects of a modular category over $`K`$ then $`\text{Hom}(U,V)`$ is a projective $`K`$-module of finite type. For any objects $`U,V`$ of $`𝒞`$ belonging to the same component of $`𝒞`$, we have $`\text{Hom}_𝒞(U,V)=\text{Hom}_𝒞(\text{1}\text{ I},VU^{})=\text{Hom}_{𝒞_1}(\text{1}\text{ I},VU^{})`$ so that $`\text{Hom}_𝒞(U,V)`$ is a projective $`K`$-module of finite type. Set
$$\mu _{U,V}=\text{Dim}(\text{Hom}_𝒞(U,V))K$$
where Dim denotes the usual dimension of projective $`K`$-modules of finite type, see for instance \[Tu2, Appendix 1\]. We have
$$\mu _{V,U}=\text{Dim}(\text{Hom}_𝒞(V,U))=\text{Dim}(\text{Hom}_𝒞(U,V))=\mu _{U,V}$$
where the second equality follows from Lemma 6.4 below and the fact that the dimensions of dual projective $`K`$-modules are equal. Note also the identity
$$\mu _{U,VW}=\mu _{UW^{},V}$$
for any $`U,V,W𝒞`$.
Axiom (6.1.1) and the equality $`\phi _\alpha (\text{id}_{\text{1}\text{ I}})=\text{id}_{\text{1}\text{ I}}`$ imply that any $`\alpha \pi `$ acts in $`\text{End}_𝒞(\text{1}\text{ I})=K`$ as the identity. Therefore the dimension of objects of $`𝒞`$ is invariant under the action of $`\pi `$: for any $`V𝒞,\alpha \pi `$, we have $`\text{dim}(\phi _\alpha (V))=\phi _\alpha (\text{dim}(V))=\text{dim}(V)`$.
If the ground ring $`K`$ is a field then axiom (6.1.4) is redundant. It is easy to show (using for instance Lemma 6.4 below) that in this case any non-zero morphism between simple objects is an isomorphism.
We need three lemmas concerning a modular crossed $`\pi `$-category $`𝒞`$. The first lemma computes the algebra of colors $`L(𝒞)=_{\alpha \pi }L_\alpha `$ as a $`K`$-module.
6.2. Lemma. Let $`\alpha \pi `$ and $`\{V_i^\alpha \}_{iI_\alpha }`$ be representatives of the isomorphism classes of simple objects in the category $`𝒞_\alpha `$. Then $`L_\alpha `$ is a free $`K`$-module with basis $`\{V_i^\alpha \}_{iI_\alpha }`$. For any $`U𝒞_\alpha `$,
$$U=\underset{iI_\alpha }{}\mu _{V_i^\alpha ,U}V_i^\alpha ,$$
$`(6.2.a)`$
$$\text{dim}(U)=\underset{iI_\alpha }{}\mu _{V_i^\alpha ,U}\text{dim}(V_i^\alpha ),$$
$`(6.2.b)`$
Proof. Let $`U𝒞_\alpha ,f:UU`$ be a generator of $`L_\alpha `$. By (6.1.3), there is a finite system of objects $`\{V_{i(r)}\}_r`$ belonging to the family $`\{V_i^\alpha \}_{iI_\alpha }`$ (possibly with repetitions) and morphisms $`\{f_r:V_{i(r)}U,g_r:UV_{i(r)}\}_r`$ such that $`\text{id}_U=_rf_rg_r`$. Then $`f=_rff_rg_r`$ and
$$U,f=\underset{r}{}U,ff_rg_r=\underset{r}{}V_{i(r)},g_rff_r.$$
Since $`V_{i(r)}`$ is a simple object, its endomorphism $`g_rff_r`$ equals $`k_r\text{id}_{V_{i(r)}}`$ for a certain $`k_rK`$. Therefore
$$U,f=\underset{r}{}V_{i(r)},k_r\text{id}_{V_{i(r)}}=\underset{r}{}k_rV_{i(r)},\text{id}_{V_{i(r)}}=\underset{r}{}k_rV_{i(r)}.$$
Therefore the elements $`\{V_i^\alpha \}_{iI_\alpha }`$ generate $`L_\alpha `$ over $`K`$. To prove their linear independence we define for each $`iI_\alpha `$ a linear functional $`t_i:LK`$ as follows. Let $`U𝒞,f:UU`$ be a generator of $`L`$. Denote by $`f_i`$ the automorphism of $`\text{Hom}_𝒞(V_i^\alpha ,U)`$ sending each $`h\text{Hom}_𝒞(V_i^\alpha ,U)`$ into $`fh`$. Set $`t_i(f)=\text{Tr}(f_i)K`$ where Tr denotes the trace of $`K`$-endomorphisms of projective $`K`$-modules, see for instance \[Tu2, Appendix 1\]. (If $`K`$ is a field then Tr is the usual trace of matrices.) It follows from the standard properties of the trace, that $`t_i`$ annihilates the relation (5.3.a) and defines thus a linear functional $`t_i:LK`$. Note that $`t_i(U)=\text{Dim}(\text{Hom}_𝒞(V_i^\alpha ,U))=\mu _{V_i^\alpha ,U}`$. By (6.1.4), $`t_i(V_j^\alpha )=\delta _j^iK`$ where $`\delta _j^i`$ is the Kronecker delta. This implies that $`L_\alpha `$ is a free $`K`$-module with basis $`\{V_i^\alpha \}_{iI_\alpha }`$.
For any object $`U𝒞`$ and any $`f\text{End}_𝒞(U)`$, there is a unique decomposition $`U,f=_{iI_\alpha }r_iV_i^\alpha `$ with $`r_iK`$. Applying $`t_i`$ we obtain $`r_i=\text{Tr}(f_i)`$ for all $`i`$. Substituting $`f=\text{id}_U`$, we obtain (6.2.a). Applying the functional $`\text{dim}:L\text{End}_𝒞(\text{1}\text{ I})=K`$ to both sides of (6.2.a) we obtain (6.2.b).
6.3. Corollary. $`L_\alpha =0`$ if and only if the category $`𝒞_\alpha `$ is void.
By a void category we mean a category with empty sets of objects and morphisms. Clearly, if $`𝒞_\alpha `$ is void then $`L_\alpha =0`$. Conversely, if $`L_\alpha =0`$ then by Lemma 6.2, the category $`𝒞_\alpha `$ has no simple objects. By (6.1.3), it is void.
6.4. Lemma. For any objects $`U,V`$ of $`𝒞`$, the pairing $`(x,y)\text{tr}(yx):\text{Hom}_𝒞(U,V)_K\text{Hom}_𝒞(V,U)K`$ is non-degenerate.
We say that a $`K`$-bilinear pairing $`H_1_KH_2K`$ is non-degenerate if the adjoint homomorphisms $`H_1\text{Hom}_K(H_2,K)`$ and $`H_2\text{Hom}_K(H_1,K)`$ are isomorphisms.
Proof of Lemma (cf. the proof of Lemma II.4.2.3 in \[Tu2\]). Consider first the case $`U=\text{1}\text{ I}𝒞_1`$. If $`V𝒞_1`$ then the claim of the lemma follows from the standard properties of the modular category $`𝒞_1`$. If $`V𝒞_\alpha `$ with $`\alpha 1`$ then the claim of the lemma is obvious since $`\text{Hom}_𝒞(U,V)=\text{Hom}_𝒞(V,U)=0`$. In the case of an arbitrary $`U`$ we have canonical isomorphisms
$$\alpha :\text{Hom}_𝒞(U,V)\text{Hom}_𝒞(\text{1}\text{ I},VU^{}),\beta :\text{Hom}_𝒞(V,U)\text{Hom}_𝒞(VU^{},\text{1}\text{ I})$$
such that $`\text{tr}(yx)=\text{tr}(\beta (y)\alpha (x))`$ for any $`x\text{Hom}_𝒞(U,V),y\text{Hom}_𝒞(V,U)`$. Therefore the general case of the lemma follows from the case $`U=\text{1}\text{ I}`$.
6.5. Lemma. For any simple object $`V`$ of $`𝒞`$, its dimension $`\text{dim}(V)`$ is invertible in $`K`$.
Proof. With respect to the generator $`\text{id}_V\text{Hom}_𝒞(V,V)=K`$ the bilinear form $`(x,y)\text{tr}(yx)K`$ on $`\text{Hom}_𝒞(V,V)`$ is presented by the $`(1\times 1)`$-matrix $`[\text{tr}(\text{id}_V)]=[\text{dim}(V)]`$. The non-degeneracy of this form implies that $`\text{dim}(V)K^{}`$.
6.6. The canonical color. Let $`𝒞`$ be a modular crossed $`\pi `$-category over $`K`$. For each $`\alpha \pi `$, choose representatives $`\{V_i^\alpha \}_{iI_\alpha }`$ of the isomorphism classes of simple objects in the category $`𝒞_\alpha `$. We define a canonical color $`\omega _\alpha L_\alpha `$ by
$$\omega _\alpha =\underset{iI_\alpha }{}\text{dim}(V_i^\alpha )V_i^\alpha .$$
$`(6.6.a)`$
It is clear that $`\omega _\alpha `$ does not depend on the choice of representatives $`\{V_i^\alpha \}_{iI_\alpha }`$. The fact that the duality $`VV^{}`$ transforms simple objects in $`𝒞_\alpha `$ into simple objects in $`𝒞_{\alpha ^1}`$ and preserves the dimension implies that
$$(\omega _\alpha )^{}=\omega _{\alpha ^1}$$
$`(6.6.b)`$
for all $`\alpha \pi `$. Since the action of any $`\alpha \pi `$ on $`𝒞`$ transforms simple objects in $`𝒞_\beta `$ into simple objects in $`𝒞_{\alpha \beta \alpha ^1}`$ we have
$$\phi _\alpha (\omega _\beta )=\omega _{\alpha \beta \alpha ^1}.$$
$`(6.6.c)`$
The next lemma generalizes a property of canonical colors well-known in the case $`\pi =1`$. We follow the proof given in \[Bru\].
6.6.1. Lemma. For any $`\alpha ,\beta \pi `$ and $`U𝒞_\beta `$, $`\omega _\alpha U=\text{dim}(U)\omega _{\alpha \beta }`$ and $`U\omega _\alpha =\text{dim}(U)\omega _{\beta \alpha }`$.
Proof. We prove only the first equality, the second one is similar. We have
$$\omega _\alpha U=\underset{iI_\alpha }{}\text{dim}(V_i^\alpha )V_i^\alpha U=\underset{iI_\alpha }{}\text{dim}(V_i^\alpha )V_i^\alpha U$$
$$=\underset{iI_\alpha }{}\underset{jI_{\alpha \beta }}{}\text{dim}(V_i^\alpha )\mu _{V_j^{\alpha \beta },V_i^\alpha U}V_j^{\alpha \beta }=\underset{iI_\alpha }{}\underset{jI_{\alpha \beta }}{}\text{dim}(V_i^\alpha )\mu _{V_i^\alpha ,V_j^{\alpha \beta }U^{}}V_j^{\alpha \beta }$$
$$=\underset{jI_{\alpha \beta }}{}(\underset{iI_\alpha }{}\text{dim}(V_i^\alpha )\mu _{V_i^\alpha ,V_j^{\alpha \beta }U^{}})V_j^{\alpha \beta }=\underset{jI_{\alpha \beta }}{}\text{dim}(V_j^{\alpha \beta }U^{})V_j^{\alpha \beta }$$
$$=\underset{jI_{\alpha \beta }}{}\text{dim}(V_j^{\alpha \beta })\text{dim}(U^{})V_j^{\alpha \beta }=\text{dim}(U^{})\omega _{\alpha \beta }=\text{dim}(U)\omega _{\alpha \beta }.$$
6.7. Elements $`𝒟,\mathrm{\Delta }_\pm `$ of $`K`$. In the sequel we shall need several elements of $`K`$ associated with the neutral component $`𝒞_1`$ of a modular crossed $`\pi `$-category $`𝒞`$ over $`K`$. Let $`\{V_i\}_{iI_1}`$ be representatives of the isomorphism classes of simple objects in $`𝒞_1`$. A rank of $`𝒞_1`$ is an element $`𝒟K`$ such that
$$𝒟^2=\underset{iI_1}{}(\text{dim}(V_i))^2K.$$
The existence of a rank is a minor technical condition which does not reduce the range of our constructions.
Since each $`V_i𝒞_1`$ is a simple object, the twist $`\theta _{V_i}:V_i\phi _1(V_i)=V_i`$ equals $`v_i\text{id}_{V_i}`$ with $`v_iK`$. Since $`\theta _{V_i}`$ is invertible, $`v_iK^{}`$. Set
$$\mathrm{\Delta }_\pm =\underset{iI_1}{}v_i^{\pm 1}(\text{dim}(V_i))^2K.$$
We can interpret $`\mathrm{\Delta }_\pm K`$ as the invariant $`F`$ of an unknot $`\mathrm{}S^3`$ endowed with framing $`\pm 1`$, trivial homomorphism $`\pi _1(C_{\mathrm{}})1\pi `$ and (generalized) color $`\omega _1L_1`$. It is known that $`𝒟,\mathrm{\Delta }_\pm `$ are invertible in $`K`$ and that $`\mathrm{\Delta }_+\mathrm{\Delta }_{}=𝒟^2`$ (see \[Tu2, Formula II.2.4.a\]).
6.8. Examples and constructions of modular $`\pi `$-categories. The ribbon $`\pi `$-category $`𝒞=𝒞(a,b,c,\theta )`$ constructed in Section 2.6 is modular by the obvious reasons: each category $`𝒞_\alpha `$ has only one object which is simple and the matrix $`S=[1]`$ is the unit $`(1\times 1)`$-matrix. If $`G\pi `$ is a finite subgroup of the center of $`\pi `$ then pushing $`𝒞(a,b,c,\theta )`$ forward along the projection $`\pi \pi /G`$ we obtain a ribbon crossed $`(\pi /G)`$-category satisfying axioms (6.1.1)-(6.1.4) but possibly not (6.1.5). This is a special case of the following easy fact: if the kernel of a group epimorphism $`q:\pi ^{}\pi `$ is finite and acts as the identity on a modular $`\pi ^{}`$-category $`𝒞^{}`$ then the push-forward $`\pi `$-category $`q_{}(𝒞^{})`$ (cf. Sections 1.4, 2.5) satisfies all axioms of a modular $`\pi `$-category except possibly (6.1.5). On the other hand, the pull-back of any modular $`\pi `$-category $`𝒞`$ along a group homomorphism $`\pi ^{}\pi `$ is always a modular $`\pi ^{}`$-category.
A tensor product of a finite family of modular crossed $`\pi `$-categories is always modular. The direct product of $`n2`$ modular crossed $`\pi `$-categories is not modular because the unit object in such a direct product is not simple, $`\text{End}(\text{1}\text{ I})=K^n`$.
It is easy to check that the mirror of a modular $`\pi `$-category is a modular $`\pi `$-category, cf. \[Tu2, Exercise II.1.9.2\].
Examples of modular $`\pi `$-categories with $`\pi =/2`$ are provided by the categories of representations of $`U_q(sl_2())`$ at roots of unity. We use a topological description of these categories given in \[Tu2\] (cf. \[Th\]). Let $`r3`$ be an odd integer and $`a`$ be a primitive complex root of unity of order $`4r`$. In \[Tu2\] the author used the tangles and the Jones-Wenzl idempotents to define a ribbon category $`𝒱(a)`$ whose objects are finite sequences $`(j_1,\mathrm{},j_l)`$ of integer numbers belonging to the set $`\{1,2,\mathrm{},r2\}`$. We label such an object with $`j_1+\mathrm{}+j_l(\text{mod}\mathrm{\hspace{0.17em}2})/2=\pi `$. The category $`𝒱(a)`$ splits as a disjoint union of two full subcategories comprising objects labelled by 0 and 1, respectively. The crossing isomorphisms $`\{\phi _\alpha \}_{\alpha \pi }`$ are the identity maps. The standard ribbon structure in $`𝒱(a)`$ makes $`𝒱(a)`$ a ribbon crossed $`\pi `$-category. Its simple objects are the same as in the standard theory. The results of \[Tu2, Section XII.7.5\] show that if $`r`$ is odd then $`𝒱(a)`$ a modular $`\pi `$-category. More general examples of modular $`\pi `$-categories with abelian $`\pi `$ associated with quantum groups will be discussed in \[LT\].
7. Invariants of 3-dimensional $`\pi `$-manifolds
7.1. Principal $`\pi `$-bundles and $`\pi `$-manifolds. Let $`\pi `$ be a (discrete) group. A principal $`\pi `$-bundle over a space $`M`$ is a regular covering $`\stackrel{~}{M}M`$ with group of automorphisms $`\pi `$. The spaces $`\stackrel{~}{M}`$ and $`M`$ are called the total space and the base of the bundle. Two principal $`\pi `$-bundles over $`M`$ are isomorphic if there is a homeomorphism of their total spaces commuting with the action of $`\pi `$ and inducing the identity map $`MM`$.
The isomorphism classes of principal $`\pi `$-bundles over a manifold $`M`$ are classified by the homotopy classes of maps from $`M`$ to the Eilenberg-MacLane space $`K(\pi ,1)`$. The monodromy of a principal $`\pi `$-bundle $`\xi `$ over $`M`$ at a point $`zM`$ is the homomorphism $`\pi _1(M,z)\pi `$ induced by the classifying map $`MK(\pi ,1)`$ of $`\xi `$. The monodromy of $`\xi `$ at $`z`$ is defined up to conjugation by an element of $`\pi `$. The monodromy can be computed geometrically as follows. Choose a point $`\stackrel{~}{z}`$ lying over $`z`$ in the total space of $`\xi `$. Any loop $`\alpha `$ in $`(M,z)`$ lifts to a path in the total space of $`\xi `$ beginning in $`\stackrel{~}{z}`$ and ending in $`g\stackrel{~}{z}`$ with $`g=g(\alpha )\pi `$. The monodromy of $`\xi `$ is then the group homomorphism $`\pi _1(M,z)\pi `$ defined by $`[\alpha ]g(\alpha )`$. A different choice of $`\stackrel{~}{z}`$ leads to a conjugated homomorphism. If $`M`$ is connected then we obtain thus a bijective correspondence between the isomorphism classes of principal $`\pi `$-bundles over $`M`$ and the group homomorphisms $`\pi _1(M,z)\pi `$ considered up to conjugation. The monodromies of $`\xi `$ at two points $`y,zM`$ and the isomorphism $`\pi _1(M,y)\pi _1(M,z)`$ induced by any path connecting $`y`$ to $`z`$ form a diagram
$$\begin{array}{ccc}\pi _1(M,y)& & \pi _1(M,z)\\ & & & & \\ \pi & =& \pi \end{array}$$
commutative up to conjugation in $`\pi `$.
By a $`\pi `$-manifold we mean a pair $`(M,\xi )`$ where $`M`$ is a manifold and $`\xi `$ is a principal $`\pi `$-bundle over $`M`$. We say that a $`\pi `$-manifold $`(M,\xi )`$ is $`m`$-dimensional (resp. closed, connected, oriented, etc.) if $`M`$ is $`m`$-dimensional (resp. closed, connected, oriented, etc.). A homeomorphism of $`\pi `$-manifolds $`(M,\xi )(M^{},\xi ^{})`$ is a homeomorphism of the total spaces of $`\xi ,\xi ^{}`$ commuting with the action of $`\pi `$. Such a homeomorphism induces a homeomorphism $`MM^{}`$. We will work in the category of oriented manifolds and consider only homeomorphisms of $`\pi `$-manifolds $`(M,\xi )(M^{},\xi ^{})`$ inducing orientation preserving homeomorphisms $`MM^{}`$.
7.2. Invariant $`\tau _𝒞`$ of 3-dimensional $`\pi `$-manifolds. Let $`𝒞`$ be a modular crossed $`\pi `$-category over a commutative unital ring $`K`$. We shall derive from $`𝒞`$ a homeomorphism invariant $`\tau _𝒞`$ of closed oriented 3-dimensional $`\pi `$-manifolds. Let $`L=_{\alpha \pi }L_\alpha `$ be the algebra of colors of $`𝒞`$ and let $`𝒟,\mathrm{\Delta }_\pm `$ be the elements of $`K^{}`$ associated with $`𝒞`$ in Section 6.7.
Let $`(M,\xi )`$ be a closed connected oriented 3-dimensional $`\pi `$-manifold. Present $`M`$ as the result of surgery on $`S^3`$ along a framed link $`\mathrm{}`$ with $`\mathrm{\#}\mathrm{}`$ components. Recall that $`M`$ is obtained by gluing $`\mathrm{\#}\mathrm{}`$ solid tori to the exterior $`E=E_{\mathrm{}}`$ of $`\mathrm{}`$. Take any point $`zEM`$. The inclusion $`EM`$ induces an epimorphism $`\pi _1(C_{\mathrm{}},z)=\pi _1(E,z)\pi _1(M,z)`$ where $`C_{\mathrm{}}=S^3\backslash \mathrm{}`$. Composing it with a homomorphism $`\pi _1(M,z)\pi `$ representing the monodromy of $`\xi `$ at $`z`$ we obtain a group homomorphism, $`g:\pi _1(C_{\mathrm{}},z)\pi `$. We fix an arbitrary orientation of $`\mathrm{}`$. The triple $`(\mathrm{},z,g)`$ is thus a $`\pi `$-link. Clearly, it is a special $`\pi `$-link in the sense of Section 5.4. We provide $`(\mathrm{},z,g)`$ with the following (generalized) coloring. To every path $`\gamma :[0,1]C_{\mathrm{}}`$ with $`\gamma (0)=z,\gamma (1)\stackrel{~}{\mathrm{}}`$, we assign $`\omega _{g(\mu _\gamma )}L_{g(\mu _\gamma )}`$. By (6.6.c), this satisfies conditions (i), (ii) of Section 5.4 and defines a canonical coloring of $`(\mathrm{},z,g)`$. Denote the resulting colored $`\pi `$-link by $`\mathrm{}_{can}`$. Set
$$\tau _𝒞(M,\xi )=\mathrm{\Delta }_{}^{\sigma (\mathrm{})}𝒟^{\sigma (\mathrm{})\mathrm{\#}\mathrm{}1}F(\mathrm{}_{can})K$$
$`(7.2.a)`$
where $`\sigma (\mathrm{})`$ is the signature of the compact oriented 4-manifold $`W_{\mathrm{}}`$ bounded by $`M`$ and obtained from the 4-ball $`B^4`$ by attaching $`n`$ 2-handles along tubular neighborhoods of the components of $`\mathrm{}`$ in $`S^3=B^4`$. Here the orientation of $`W_{\mathrm{}}`$ is induced by the one of $`M`$. We use the “outward vector first” convention for the induced orientation: at any point of $`M=W_{\mathrm{}}`$ the orientation of $`W_{\mathrm{}}`$ is determined by the tuple (a tangent vector directed outwards, a positive basis in the tangent space of $`M`$).
7.3. Theorem. $`\tau _𝒞(M,\xi )`$ is a homeomorphism invariant of $`(M,\xi )`$.
Proof. We should prove that $`\tau _𝒞(M,\xi )`$ does not depend on the choices made in its definition. Under the conjugation of the monodromy $`\pi _1(M,z)\pi `$ by an element $`\alpha \pi `$, the homomorphism $`g`$ is replaced with $`\alpha g\alpha ^1`$. It follows from (6.6.c) that the colored $`\pi `$-links $`\mathrm{}_{can}`$ corresponding to $`g`$ and $`\alpha g\alpha ^1`$ are related by the transformation $`\phi _\alpha `$ defined in the proof of Lemma 3.4.1. By Theorem 3.6 and (3.5.(v)), the corresponding values $`F(\mathrm{}_{can})\text{End}_𝒞(\text{1}\text{ I})=K`$ are also related via $`\phi _\alpha `$. Since $`\phi _\alpha `$ acts in $`\text{End}_𝒞(\text{1}\text{ I})`$ as the identity, $`F(\mathrm{}_{can})`$ does not depend on the choice of monodromy in its conjugacy class.
The independence of the choice of $`z`$ follows from the invariance of $`F(\mathrm{}_{can})`$ under transfers described in Section 3.1.
Let $`\mathrm{}^{}`$ be obtained from $`\mathrm{}`$ by reversing the orientation on one of the components. By (6.6.b), $`\mathrm{}_{can}`$ and $`\mathrm{}_{can}^{}`$ are related by the transformation described in Section 4.7.2 and therefore $`F(\mathrm{}_{can})=F(\mathrm{}_{can}^{})`$. Therefore $`\tau _𝒞(M,\xi )`$ does not depend on the choice of orientation on $`\mathrm{}`$.
To prove the independence of the choice of $`\mathrm{}`$ we use the Kirby theory of moves on framed links. It is shown in \[Ki\] that any two framed links in $`S^3`$ yielding after surgery homeomorphic 3-manifolds can be related by certain transformations called Kirby moves. There are moves of two kinds. The first move adds to a framed link $`\mathrm{}S^3`$ a distant unknot $`\mathrm{}^\pm `$ with framing $`\pm 1`$; under this move the 4-manifold $`W_{\mathrm{}}`$ is transformed into its connected sum with $`CP^2`$. The second move preserves $`W_{\mathrm{}}`$ and is induced by a sliding of a 2-handle of $`W_{\mathrm{}}`$ across another 2-handle. We need a more precise version of this theory. For a framed link $`\mathrm{}S^3`$, denote the result of surgery on $`\mathrm{}`$ by $`M_{\mathrm{}}`$. A surgery presentation of a closed connected oriented 3-manifold $`M`$ is a pair (a framed link $`\mathrm{}S^3`$, an isotopy class of degree $`+1`$ homeomorphisms $`f:MM_{\mathrm{}}`$). Note that any framing preserving isotopy of $`\mathrm{}`$ onto itself induces a homeomorphism $`j_0:M_{\mathrm{}}M_{\mathrm{}}`$. Clearly, $`(\mathrm{},j_0f)`$ is a surgery presentation of $`M`$; we say that it is obtained from $`(\mathrm{},f)`$ by isotopy. The first Kirby move $`\mathrm{}\mathrm{}^{}=\mathrm{}\mathrm{}^\pm `$ induces a homeomorphism $`j_1:M_{\mathrm{}}M_{\mathrm{}^{}}`$ which is the identity outside a small 3-ball containing $`\mathrm{}^\pm `$. The second Kirby move $`\mathrm{}\mathrm{}^{}`$ induces a diffeomorphisms $`W_{\mathrm{}}W_{\mathrm{}^{}}`$ which restricts to a homeomorphism $`j_2:M_{\mathrm{}}M_{\mathrm{}^{}}`$. In both cases we say that the surgery presentation $`(\mathrm{}^{},j_kf:MM_{\mathrm{}^{}})`$ (where $`k=1,2`$) is obtained from $`(\mathrm{},f:MM_{\mathrm{}})`$ by the $`k`$-th Kirby move. The arguments in \[Ki, Section 2\] show that for any surgery presentations $`(\mathrm{}_1,f_1:M_1M_\mathrm{}_1)`$ and $`(\mathrm{}_2,f_2:M_2M_\mathrm{}_2)`$ of closed connected oriented 3-manifolds $`M_1,M_2`$ and for any isotopy class of degree $`+1`$ homeomorphisms $`f:M_1M_2`$ there is a sequence of Kirby moves and isotopies transforming $`(\mathrm{}_1,f_1)`$ into $`(\mathrm{}_2,f_2f)`$.
The result $`M_{\mathrm{}}`$ of surgery on a special $`\pi `$-link $`\mathrm{}S^3`$ is a $`\pi `$-manifold in the obvious way. Any Kirby move on a special $`\pi `$-link $`\mathrm{}S^3`$ yields a special $`\pi `$-link $`\mathrm{}^{}S^3`$ where the homomorphism $`\pi _1(C_{\mathrm{}^{}})\pi `$ is the composition of the inclusion homomorphism $`\pi _1(C_{\mathrm{}^{}})\pi _1(M_{\mathrm{}^{}})`$, the isomorphism $`\pi _1(M_{\mathrm{}^{}})=\pi _1(M_{\mathrm{}})`$ induced by the homeomorphism $`M_{\mathrm{}}M_{\mathrm{}^{}}`$ mentioned above and the homomorphism $`\pi _1(M_{\mathrm{}})\pi `$ induced by the given homomorphism $`\pi _1(C_{\mathrm{}})\pi `$. The results of the previous paragraph imply that if two special $`\pi `$-links in $`S^3`$ yield after surgery homeomorphic $`\pi `$-manifolds then these $`\pi `$-links can be related by a finite sequence of Kirby moves and isotopies.
It is clear that $`\tau _𝒞(M,\xi )`$ is invariant under isotopies on $`\mathrm{}`$. To prove the theorem it is thus enough to show that $`\tau _𝒞(M,\xi )`$ is invariant under the Kirby moves on $`\mathrm{}`$. We begin with the first Kirby move $`\mathrm{}\mathrm{}^{}=\mathrm{}\mathrm{}^\pm `$. The meridian of the unknot $`\mathrm{}^\pm `$ is contractible in $`M`$ and therefore the $`\pi `$-link $`\mathrm{}^{}`$ is a disjoint union of $`\mathrm{}`$ and the framed unknot $`\mathrm{}^\pm `$ endowed with the trivial homomorphism to $`\pi `$. The colored $`\pi `$-link $`\mathrm{}_{can}^{}`$ is a disjoint union of $`\mathrm{}_{can}`$ and the framed unknot $`\mathrm{}_{can}^\pm `$ endowed with the trivial homomorphism to $`\pi `$ and the color $`\omega _1L_1`$. We have
$$F(\mathrm{}_{can}^{})=F(\mathrm{}_{can}^\pm )F(\mathrm{}_{can})=\mathrm{\Delta }_\pm F(\mathrm{}_{can}).$$
Therefore the invariance of $`\tau _𝒞(M,\xi )`$ under the first Kirby move follows from the formulas $`\mathrm{\#}(\mathrm{}^{})=\mathrm{\#}\mathrm{}+1,\sigma (\mathrm{}^{})=\sigma (\mathrm{})\pm 1`$, $`\mathrm{\Delta }_+\mathrm{\Delta }_{}=𝒟^2`$.
We consider the second Kirby moves in the restricted form studied by Fenn and Rourke, \[FR1\]. The Kirby-Fenn-Rourke moves split into positive and negative ones. It is explained in \[RT\] that (modulo the first Kirby moves) it is enough to consider only one of these families. Consider for concreteness a negative Kirby-Fenn-Rourke move $`\mathrm{}\mathrm{}^{}`$. It replaces a piece $`T`$ of $`\mathrm{}`$ lying in a ball by another piece $`T^{}`$ lying in the same ball and having the same endpoints. Here $`T`$ is a system of parallel strings with parallel framing and $`T^{}`$ is obtained from $`T`$ by applying a full left-hand twist and adding an unknotted component $`t`$ which encircles $`T`$ and has framing $`1`$. Note that $`\mathrm{\#}\mathrm{}^{}=\mathrm{\#}\mathrm{}+1`$ and $`\sigma (\mathrm{}^{})=\sigma (\mathrm{})1`$. We need to prove that $`F(\mathrm{}_{can}^{})=\mathrm{\Delta }_{}F(\mathrm{}_{can})`$. This equality follows from a “local” equality involving only $`T`$ and $`T^{}`$. To formulate this local equality we first position $`T`$ as a trivial braid in $`^2\times [0,1]`$ with constant framing. The framed tangle $`T^{}=Tt^2\times [0,1]`$ is obtained from $`T`$ as explained above. We orient $`T`$ and $`t`$. Note that $`C_T=(^2\times [0,1])\backslash T`$ is obtained from $`C_T^{}=(^2\times [0,1])\backslash T^{}`$ by the surgery on $`t`$. Therefore any group homomorphism $`g:\pi _1(C_T)\pi `$ induces a group homomorphism $`g^{}:\pi _1(C_T^{})\pi `$ such that $`g^{}`$ maps the homotopy class of the $`(1)`$-longitude of $`t`$ into $`1\pi `$. Any coloring $`u`$ of $`(T,g)`$ induces a unique coloring $`u^{}`$ of $`(T^{},g^{})`$ such that the sources and targets of $`T`$ and $`T^{}`$ coincide and the component $`t`$ of $`T^{}`$ has the canonical color. The local equality mentioned above says that for any orientation of $`Tt`$, any group homomorphism $`g:\pi _1(C_T)\pi `$, and any coloring $`u`$ of $`(T,g)`$, we have
$$F(T^{},g^{},u^{})=\mathrm{\Delta }_{}F(T,g,u).$$
$`(7.3.a)`$
Let us prove this formula. Let $`(\epsilon _1,\alpha _1,U_1),\mathrm{},(\epsilon _k,\alpha _k,U_k)`$ be the source of $`T`$. Using the standard technique of coupons colored with identity morphisms we can reduce the general case to the case where $`T`$ consists of only one string oriented from top to bottom and colored with object $`_{r=1}^k(U_r)^{\epsilon _r}𝒞`$. Using a decomposition of the identity endomorphism of this object provided by axiom (6.1.3), we can further reduce ourselves to the case where the string $`T`$ is colored with a simple object of $`𝒞`$. Thus we can assume that $`T`$ is a string oriented from top to bottom and the source and target of $`T`$ (and $`T^{}`$) are a $`1`$-term sequence $`(+,\alpha ,V)`$ where $`\alpha \pi `$ and $`V`$ is a simple object of $`𝒞_\alpha `$. By the argument used above, the invariant $`F(T^{},g^{},u^{})`$ does not change if we invert the orientation of $`t`$. Therefore we can assume that $`t`$ is oriented so that its linking number with $`T`$ equals $`1`$. Let us denote the colored $`\pi `$-tangles $`T,T^{}`$ by $`T_V,T_V^{}`$. Clearly, $`F(T_V)=\text{id}_V`$. Since the object $`V`$ is simple, $`F(T_V^{})=k\text{id}_V`$ with $`kK`$. We have to prove only that $`k=\mathrm{\Delta }_{}`$. To this end we close $`T_V^{}`$ into a colored 2-component $`\pi `$-link $`\widehat{T}_V^{}`$. As in the standard theory (cf. \[Tu2, Corollary I.2.7.1\]),
$$F(\widehat{T}_V^{})=\text{tr}(F(T_V^{}))=\text{tr}(k\text{id}_V)=k\text{tr}(\text{id}_V)=k\text{dim}(V).$$
On the other hand, the link $`\widehat{T}_V^{}`$ is obtained by doubling described in Section 4.7.3 from an unknot $`\mathrm{}^{}S^3`$ endowed with framing $`1`$, trivial homomorphism $`\pi _1(C_{\mathrm{}^{}})1\pi `$ and color $`\omega _{\alpha ^1}VL_1`$. By Lemma 6.6.1 and the results of Section 6.7 the invariant $`F`$ of this colored $`\pi `$-unknot $`(\mathrm{}^{},\omega _{\alpha ^1}V)`$ is computed by
$$F(\mathrm{}^{},\omega _{\alpha ^1}V)=F(\mathrm{}^{},\text{dim}(V)\omega _1)=\text{dim}(V)F(\mathrm{}^{},\omega _1)=\text{dim}(V)\mathrm{\Delta }_{}.$$
Since $`F`$ is preserved under the doubling, we have
$$F(\widehat{T}_V^{})=F(\mathrm{}^{},\omega _{\alpha ^1}V)=\text{dim}(V)\mathrm{\Delta }_{}=\mathrm{\Delta }_{}\text{dim}(V).$$
Comparing these two computations of $`F(\widehat{T}_V^{})`$ and using Lemma 6.5, we obtain $`k=\mathrm{\Delta }_{}`$.
7.4. Computations and remarks. 1. The 3-sphere $`S^3`$ is simply connected and therefore admits a unique structure of a $`\pi `$-manifold. Presenting $`S^3`$ as the result of surgery on $`S^3`$ along an empty link we obtain $`\tau _𝒞(S^3)=𝒟^1`$.
2. For each $`\alpha \pi `$, there is a unique $`\pi `$-structure $`\xi _\alpha `$ on $`S^1\times S^2`$ whose monodromy along $`S^1\times pt`$ equals $`\alpha `$. We prove in Section 7.7 that
$$\tau _𝒞(S^1\times S^2,\xi _\alpha )=\{\begin{array}{cc}0,\text{if \hspace{0.17em} the \hspace{0.17em} category}𝒞_\alpha \text{is\hspace{0.17em} void},\hfill & \\ 1,\text{otherwise}.\hfill & \end{array}$$
$`(7.4.a)`$
3. Formula (7.2.a) can be rewritten in a more symmetric form:
$$\tau _𝒞(M,\xi )=𝒟^{b_1(M)1}\mathrm{\Delta }_{}^\sigma _{}\mathrm{\Delta }_+^{\sigma _+}F(\mathrm{}_{can})$$
where $`b_1(M)=\mathrm{\#}\mathrm{}\sigma _+\sigma _{}`$ is the first Betti number of $`M`$ and $`\sigma _+`$ (resp. $`\sigma _{}`$) is the number of positive (resp. negative) squares in the diagonal decomposition of the intersection form $`H_2(W_{\mathrm{}})\times H_2(W_{\mathrm{}})`$. This shows that the invariant
$$\tau _𝒞^{}(M,\xi )=\mathrm{\Delta }_{}^\sigma _{}\mathrm{\Delta }_+^{\sigma _+}F(\mathrm{}_{can})=𝒟^{b_1(M)+1}\tau _𝒞(M,\xi )$$
does not depend on the choice of $`𝒟`$. Note that $`\tau _𝒞^{}(M,\xi )`$ is defined for a wider class of ribbon crossed $`\pi `$-categories $`𝒞`$ satisfying (6.1.1)-(6.1.4) and such that $`\mathrm{\Delta }_+,\mathrm{\Delta }_{}K^{}`$. The latter condition is a weakened form of (6.1.5): it follows from (6.1.1)-(6.1.5) but in general does not imply (6.1.5). However, the invertibility of the matrix $`S`$ is needed for the construction of a TQFT.
4. It is easy to deduce from definitions that the invariant $`𝒟\tau _𝒞`$ is multiplicative with respect to the connected sum of $`\pi `$-manifolds. In other words, for closed connected oriented 3-dimensional $`\pi `$-manifolds $`M,N`$
$$\tau _𝒞(M\mathrm{\#}N)=𝒟\tau _𝒞(M)\tau _𝒞(N)$$
where the structure of a $`\pi `$-manifold on $`M\mathrm{\#}N`$ is defined so that its monodromy extends the monodromies of $`M`$ and $`N`$.
5. We extend $`\tau _𝒞(M,\xi )`$ to non-connected closed oriented 3-dimensional $`\pi `$-manifolds by multiplicativity so that $`\tau _𝒞(MN,\xi )=\tau _𝒞(M,\xi |_M)\tau _𝒞(N,\xi |_N)`$.
6. Let $`𝒞=𝒱(a)`$ be the modular $`(/2)`$-category associated with a primitive complex root of unity $`a`$ of order $`4r`$ with odd $`r3`$ discussed in Section 6.8. The principle $`/2`$-bundles over $`M`$ are numerated by elements $`\xi H^1(M;/2)`$. The corresponding invariants $`\tau _𝒞(M,\xi )`$ were first introduced in \[Bl\], \[KM\], \[Tu1\]. The invariant $`\tau _𝒞(M,\xi )`$ corresponding to $`\xi =0H^1(M;/2)`$ is called the quantum $`SO(3)`$-invariant of $`M`$. This example suggests that there should be similar categorical structures yielding invariants of spin 3-manifolds or more generally 3-manifolds endowed with principal bundles over their spaces of tangent frames. The author hopes to consider this elsewhere.
7.5. Extended $`\pi `$-manifolds. The invariant $`\tau _𝒞`$ defined above can be generalized to 3-manifolds with partial $`\pi `$-structure, i.e., with a principal $`\pi `$-bundle on the complement of a framed oriented link or more generally on the complement of a ribbon graph. This graph should be colored over $`𝒞`$. We consider here only 3-manifolds without boundary and ribbon graphs without inputs or outputs, the manifolds with boundary will be discussed in Section 10. We proceed to precise definitions.
Let $`M`$ be a closed connected oriented 3-manifold. A ribbon graph in $`M`$ consists of a finite number of framed oriented embedded arcs, circles and coupons which are disjoint, except that the endpoints of the arcs lie on the bases of the coupons, and the framing satisfies the same conditions as in Section 4.1. Since $`M`$ is connected, the complement of a ribbon graph in $`M`$ is connected. A $`\pi `$-graph in $`M`$ is a ribbon graph $`\mathrm{\Omega }M`$ whose complement is endowed with a base point $`zM\backslash \mathrm{\Omega }`$ and a homomorphism $`\pi _1(M\backslash \mathrm{\Omega },z)\pi `$. This data determines a principal $`\pi `$-bundle over $`M\backslash \mathrm{\Omega }`$ which may extend or not to $`M`$. We shall use the language of monodromies rather than principal bundles; that is why we pay attention to the base points. Note that a $`\pi `$-graph without coupons is a $`\pi `$-link as defined in Section 3.1.
Fix a modular crossed $`\pi `$-category $`𝒞`$. A $`\pi `$-graph in $`M`$ is colored (over $`𝒞`$) if it is equipped with two functions $`u,v`$ satisfying the same conditions as in Section 4.1. The notion of an ambient isotopy in $`M`$ applies to $`\pi `$-graphs and colored $`\pi `$-graphs in $`M`$ in the obvious way. This allows us to consider the (ambient) isotopy classes of such graphs. As in Section 3.1, we can transfer the structure of a colored $`\pi `$-graph $`\mathrm{\Omega }`$ along paths in $`M\backslash \mathrm{\Omega }`$ relating various base points. The transfers preserve the ambient isotopy class of a colored $`\pi `$-graph.
A pair consisting of a closed connected oriented 3-manifold $`M`$ and a colored $`\pi `$-graph in $`M`$ is called a connected 3-dimensional extended $`\pi `$-manifold (without boundary). Let $`(M,\mathrm{\Omega })`$, $`(M^{},\mathrm{\Omega }^{})`$ be two connected 3-dimensional extended $`\pi `$-manifolds without boundary. Here $`\mathrm{\Omega }=(\mathrm{\Omega }M,z,g:\pi _1(M\backslash \mathrm{\Omega },z)\pi ,u,v)`$ is a colored $`\pi `$-graph in $`M`$ and $`\mathrm{\Omega }^{}=(\mathrm{\Omega }^{}M^{},z^{},g^{}:\pi _1(M^{}\backslash \mathrm{\Omega }^{},z^{})\pi ,u^{},v^{})`$ is a colored $`\pi `$-graph in $`M^{}`$. By an $`e`$-homeomorphism $`(M,\mathrm{\Omega })(M^{},\mathrm{\Omega }^{})`$ we mean a degree $`+1`$ homeomorphism of triples $`f:(M,\mathrm{\Omega },z)(M^{},\mathrm{\Omega }^{},z^{})`$ preserving the framing, the orientation and the splitting of $`\mathrm{\Omega },\mathrm{\Omega }^{}`$ into strata and such that $`g^{}f_\mathrm{\#}=g:\pi _1(M\backslash \mathrm{\Omega },z)\pi `$ and for any path $`\gamma `$ in $`M\backslash \mathrm{\Omega }`$ leading from $`z`$ to an arc or a circle of $`\stackrel{~}{\mathrm{\Omega }}`$ (resp. a coupon of $`\stackrel{~}{\mathrm{\Omega }}`$) we have $`u_{f\gamma }^{}=u_\gamma `$ (resp. $`v_{f\gamma }^{}=v_\gamma `$). In particular, if $`M^{}=M`$ and $`\mathrm{\Omega }^{}`$ is obtained from $`\mathrm{\Omega }`$ via an ambient isotopy $`\{f_t:MM\}_{t[0,1]}`$ (where $`f_0=\text{id}_M`$) then $`f_1`$ is an $`e`$-homeomorphism $`(M,\mathrm{\Omega })(M^{},\mathrm{\Omega }^{})`$.
The invariant $`\tau _𝒞`$ generalizes to extended $`\pi `$-manifolds as follows. Let $`\mathrm{\Omega }=(\mathrm{\Omega },z,g:\pi _1(M\backslash \mathrm{\Omega },z)\pi ,u,v)`$ be a colored $`\pi `$-graph in a closed connected oriented 3-manifold $`M`$. Present $`M`$ as the result of surgery on $`S^3`$ along a framed link $`\mathrm{}`$. As above, $`M`$ is obtained by gluing $`\mathrm{\#}\mathrm{}`$ solid tori to the exterior $`E`$ of $`\mathrm{}`$ in $`S^3`$. Applying isotopy to $`\mathrm{\Omega }M`$ we can deform it into $`EM`$. Similarly, we can push the base point $`zM\backslash \mathrm{\Omega }`$ into $`E`$. Thus, we may assume that $`\mathrm{\Omega }E`$ and $`zE`$. The inclusion $`EM`$ induces an epimorphism $`\pi _1(S^3\backslash (\mathrm{}\mathrm{\Omega }),z)=\pi _1(E\backslash \mathrm{\Omega },z)\pi _1(M\backslash \mathrm{\Omega },z)`$. Composing this with $`g`$ we obtain a homomorphism, $`\stackrel{~}{g}:\pi _1(S^3\backslash (\mathrm{}\mathrm{\Omega }),z)\pi `$. Fix an arbitrary orientation of $`\mathrm{}`$. The triple $`(\mathrm{}\mathrm{\Omega },z,\stackrel{~}{g})`$ is thus a $`\pi `$-graph in $`S^3`$. We equip $`\mathrm{}`$ with the canonical color as in Section 7.2 and keep the given coloring of $`\mathrm{\Omega }`$. (By the inclusion $`EM`$, any path in $`E`$ is also a path in $`M`$.) Denote the resulting colored $`\pi `$-graph in $`S^3`$ by $`\mathrm{}_{can}\mathrm{\Omega }`$. Set
$$\tau _𝒞(M,\mathrm{\Omega })=\mathrm{\Delta }_{}^{\sigma (\mathrm{})}𝒟^{\sigma (\mathrm{})\mathrm{\#}\mathrm{}1}F(\mathrm{}_{can}\mathrm{\Omega })K.$$
$`(7.5.a)`$
7.6. Theorem. Let $`\mathrm{\Omega }`$ be a colored $`\pi `$-graph in a closed connected oriented 3-manifold $`M`$. Then $`\tau _𝒞(M,\mathrm{\Omega })`$ does not depend on the choices made in its definition and is an $`e`$-homeomorphism invariant of the pair $`(M,\mathrm{\Omega })`$.
Theorem 7.6 includes Theorem 7.3 as a special case $`\mathrm{\Omega }=\mathrm{}`$. In this case the homomorphism $`g:\pi _1(M\backslash \mathrm{\Omega },z)=\pi _1(M,z)\pi `$ provides $`M`$ with a structure, $`\xi `$, of a $`\pi `$-manifold and $`\tau _𝒞(M,\mathrm{\Omega })=\tau _𝒞(M,\xi )`$. Theorem 7.6 implies that $`\tau _𝒞(M,\mathrm{\Omega })`$ is invariant under ambient isotopies of $`\mathrm{\Omega }`$. In particular, $`\tau _𝒞(M,\mathrm{\Omega })`$ is invariant under transfers of the base point.
The proof of Theorem 7.6 reproduces the proof of Theorem 7.3 with natural changes (cf. \[Tu2, Section II.3\]).
Theorem 7.6 implies that $`\tau (M,\mathrm{\Omega })`$ is an isotopy invariant of $`\mathrm{\Omega }`$. Restricting ourselves to colored $`\pi `$-graphs consisting of circles we obtain an isotopy invariant of colored $`\pi `$-links in $`M`$.
If $`M=S^3`$ then $`\tau _𝒞(M,\mathrm{\Omega })=𝒟^1F(\mathrm{\Omega })`$ (we may take $`\mathrm{}=\mathrm{}`$ to compute $`\tau _𝒞(S^3,\mathrm{\Omega })`$). The invariant $`\tau _𝒞(M,\mathrm{\Omega })`$ satisfies the following multiplicativity law:
$$\tau _𝒞(M_1\mathrm{\#}M_2,\mathrm{\Omega }_1\mathrm{\Omega }_2)=𝒟\tau _𝒞(M_1,\mathrm{\Omega }_1)\tau _𝒞(M_2,\mathrm{\Omega }_2)$$
$`(7.6.a)`$
where $`\mathrm{\Omega }_1,\mathrm{\Omega }_2`$ are colored $`\pi `$-graphs in closed connected oriented 3-manifolds $`M_1,M_2`$ respectively. The properties of the invariant $`F`$ of colored $`\pi `$-graphs in $`S^3`$ established in Section 4.7 generalize in the obvious way to colored $`\pi `$-graphs in closed 3-manifolds.
Taking disjoint unions of connected extended $`\pi `$-manifolds we can obtain non-connected extended $`\pi `$-manifolds. The invariant $`\tau _𝒞`$ extends to them by multiplicativity.
7.7. Proof of (7.4.a). Set
$$d_\alpha =\text{dim}(\omega _\alpha )=\underset{iI_\alpha }{}(\text{dim}(V_i^\alpha ))^2K$$
where $`I_\alpha `$ is the set of isomorphism classes of simple objects in $`𝒞_\alpha `$ and $`V_i^\alpha `$ is a simple object of $`𝒞_\alpha `$ representing $`iI_\alpha `$. We can obtain $`(S^1\times S^2,\xi _\alpha )`$ by surgery on $`S^3`$ along a $`\pi `$-unknot $`\mathrm{}`$ with zero framing and with homomorphism $`\pi _1(C_{\mathrm{}})\pi `$ sending a meridian of $`\mathrm{}`$ into $`\alpha \pi `$. Clearly, $`\sigma (\mathrm{})=0`$. By definition,
$$\tau _𝒞(S^1\times S^2,\xi _\alpha )=𝒟^2F(\mathrm{}_{can})=𝒟^2d_\alpha .$$
If the category $`𝒞_\alpha `$ is void then $`\tau _𝒞(S^1\times S^2,\xi _\alpha )=d_\alpha =0`$. Assume that $`𝒞_\alpha `$ is non-void. It suffices to prove that $`d_\alpha =𝒟^2`$.
By the assumption and axiom (6.1.3), the category $`𝒞_\alpha `$ has at least one simple object. Choose a simple object $`U𝒞_\alpha `$. Consider a link diagram in $`^2`$ obtained as disjoint union of two small embedded circles $`\mathrm{}_1,\mathrm{}_2`$ equipped with clockwise orientation. We label $`\mathrm{}_1,\mathrm{}_2`$ with $`(\alpha ,U)`$ and $`(\beta ,\omega _\beta )`$, respectively, where $`\beta \pi `$. This determines a special colored $`\pi `$-link in $`S^3`$. Apply the surgery on $`S^3`$ along $`\mathrm{}_2`$. Then $`\mathrm{}_1`$ represents a colored $`\pi `$-knot (in fact an unknot), $`\mathrm{\Omega }=\mathrm{\Omega }(\alpha ,\beta ,U)`$, in the result of the surgery $`S^1\times S^2`$. It follows from definitions (using multiplicativity of $`F`$ with respect to disjoint union) that
$$\tau _𝒞(S^1\times S^2,\mathrm{\Omega })=𝒟^2\text{dim}(U)d_\beta .$$
This formula follows also from (7.6.a) if we observe that $`(S^1\times S^2,\mathrm{\Omega })`$ is a connected sum of the $`\pi `$-manifold $`(S^1\times S^2,\xi _\beta )`$ and the extended $`\pi `$-manifold $`(S^3,\mathrm{}_1)`$. Observe now that the extended $`\pi `$-manifolds $`(S^1\times S^2,\mathrm{\Omega }(\alpha ,\beta ,U))`$ and $`(S^1\times S^2,\mathrm{\Omega }(\alpha ,\beta \alpha ,U))`$ are $`e`$-homeomorphic. To see this, we identify $`\mathrm{\Omega }`$ with the equatorial circle in the 2-sphere $`pt\times S^2S^1\times S^2`$. This circle splits $`pt\times S^2`$ into two half-discs whose regular neighborhoods are 3-balls, containing $`\mathrm{\Omega }`$. These two 3-balls give rise to two splittings of $`(S^1\times S^2,\mathrm{\Omega })`$ into connected sum
$$(S^1\times S^2,\mathrm{\Omega })=(S^1\times S^2,\xi _\beta )\mathrm{\#}(S^3,\mathrm{}_1)$$
and
$$(S^1\times S^2,\mathrm{\Omega })=(S^1\times S^2,\xi _{\beta \alpha })\mathrm{\#}(S^3,\mathrm{}_1).$$
Hence $`(S^1\times S^2,\mathrm{\Omega }(\alpha ,\beta ,U))`$ and $`(S^1\times S^2,\mathrm{\Omega }(\alpha ,\beta \alpha ,U))`$ are $`e`$-homeomorphic. This implies
$$𝒟^2\text{dim}(U)d_\beta =𝒟^2\text{dim}(U)d_{\beta \alpha }.$$
Since $`\text{dim}(U)K^{}`$, we obtain $`d_\beta =d_{\beta \alpha }`$. For $`\beta =1`$, this gives $`d_\alpha =d_1=𝒟^2`$.
7.8. Remark. The argument in Section 7.7 shows that the set of $`\alpha \pi `$ such that $`𝒞_\alpha \mathrm{}`$ is a subgroup of $`\pi `$ (in fact a normal subgroup).
8. A 2-dimensional homotopy modular functor
A 3-dimensional topological quantum field theory (TQFT) derived from a modular category comprises two ingredients: a modular functor assigning $`K`$-modules (called modules of conformal blocks) to surfaces and an invariant of 3-dimensional cobordisms. The surfaces in this theory have a certain additional structure consisting of a finite family of marked points and a Lagrangian space in real 1-dimensional homology, such surfaces are said to be extended. It turns out that a modular crossed $`\pi `$-category gives rise to a homotopy quantum field theory (HQFT) for $`\pi `$-surfaces and $`\pi `$-cobordisms. This has two ingredients: a homotopy modular functor assigning $`K`$-modules to extended $`\pi `$-surfaces and an invariant of 3-dimensional $`\pi `$-cobordisms. In this section we discuss the homotopy modular functor. In the case $`\pi =1`$ we recover the standard theory.
8.1. Preliminaries. We shall use the language of pointed homotopy theory. A topological space is pointed if all its connected components are provided with base points. A map between pointed spaces is a continuous map sending base points into base points and considered up to homotopy constant on the base points.
We fix an Eilenberg-MacLane space $`X=K(\pi ,1)`$ associated with $`\pi `$ and a base point $`xX`$. We assume that $`X`$ is a CW-space. Note that maps from a pointed connected CW-complex $`Y`$ into $`X`$ bijectively correspond to group homomorphisms $`\pi _1(Y)\pi =\pi _1(X,x)`$. The language of maps to $`X`$ is essentially equivalent to but slightly more convenient than the language of group homomorphisms into $`\pi `$ because it allows to treat connected and non-connected $`Y`$ on the same footing.
We fix a modular crossed $`\pi `$-category $`𝒞`$.
8.2. Extended $`\pi `$-surfaces. We first define extended $`\pi `$-surfaces without marks. An extended $`\pi `$-surface without marks is a pointed closed oriented surface $`\mathrm{{\rm Y}}`$ endowed with a map $`\mathrm{{\rm Y}}X=K(\pi ,1)`$ and with a Lagrangian space $`\lambda H_1(\mathrm{{\rm Y}};)`$. According to our conventions, the map $`\mathrm{{\rm Y}}X`$ sends the base points of all the components of $`\mathrm{{\rm Y}}`$ into $`xX`$ and is considered up to homotopy constant on the base points. Recall that a Lagrangian space in $`H_1(\mathrm{{\rm Y}};)`$ is a linear subspace of maximal dimension (equal to $`\frac{1}{2}\text{dim}H_1(\mathrm{{\rm Y}};)`$) on which the homological intersection form $`H_1(\mathrm{{\rm Y}};)\times H_1(\mathrm{{\rm Y}};)`$ is zero.
Now we define more general extended $`\pi `$-surfaces with marks. Let $`\mathrm{{\rm Y}}`$ be a pointed closed oriented surface. A point $`p\mathrm{{\rm Y}}`$ is marked if is equipped with a sign $`\epsilon _p=\pm 1`$ and a tangent direction, i.e., a ray $`_+v`$ where $`v`$ is a non-zero tangent vector at $`p`$. A marking of $`\mathrm{{\rm Y}}`$ is a finite (possibly void) set of distinct marked points $`P\mathrm{{\rm Y}}`$ disjoint from the base points (of the components) of $`\mathrm{{\rm Y}}`$. Pushing slightly a marked point $`pP`$ in the given tangent direction we obtain another point $`\stackrel{~}{p}\mathrm{{\rm Y}}`$ which in analogy with knot theory can be viewed as a “longitude” of $`p`$. Set $`\stackrel{~}{P}=_{pP}\stackrel{~}{p}\mathrm{{\rm Y}}\backslash P`$. Clearly, $`\text{card}(\stackrel{~}{P})=\text{card}(P)`$. For a path $`\gamma :[0,1]\mathrm{{\rm Y}}\backslash P`$ connecting a point $`z=\gamma (0)`$ to $`\stackrel{~}{p}\stackrel{~}{P}`$, denote by $`\mu _\gamma \pi _1(\mathrm{{\rm Y}}\backslash P,z)`$ the homotopy class of the loop $`(\gamma m_p\gamma ^1)^{\epsilon _p}=\gamma m_p^{\epsilon _p}\gamma ^1`$, where $`m_p`$ is a small loop beginning and ending in $`\stackrel{~}{p}`$ and encircling the point $`pP`$ in the clockwise direction. (The clockwise direction is opposite to the one induced by the orientation of $`\mathrm{{\rm Y}}`$.)
A $`\pi `$-marking of $`\mathrm{{\rm Y}}`$ is a marking $`P\mathrm{{\rm Y}}`$ endowed with a map $`g:\mathrm{{\rm Y}}\backslash PX=K(\pi ,1)`$ sending the base points of $`\mathrm{{\rm Y}}`$ into $`xX`$ and considered up to homotopy constant on the base points. A $`\pi `$-marking $`P\mathrm{{\rm Y}}`$ is said to be colored if it is equipped with a function $`u`$ which assigns to every path $`\gamma :[0,1]\mathrm{{\rm Y}}\backslash P`$ leading from a base point, $`z\mathrm{{\rm Y}}`$, to $`\stackrel{~}{P}`$ an object $`u_\gamma 𝒞`$ such that
(i) $`u_\gamma `$ is preserved under homotopies of $`\gamma `$ in $`\mathrm{{\rm Y}}\backslash P`$ fixing $`\gamma (0),\gamma (1)`$;
(ii) $`u_\gamma 𝒞_{g_\mathrm{\#}(\mu _\gamma )}`$ where $`g_\mathrm{\#}:\pi _1(\mathrm{{\rm Y}}\backslash P,z)\pi =\pi _1(X,x)`$ is the homomorphism induced by $`g`$;
(iii) if $`\beta `$ is a loop in $`(\mathrm{{\rm Y}}\backslash P,z)`$, then $`u_{\beta \gamma }=\phi _{g_\mathrm{\#}([\beta ])}(u_\gamma )`$.
An extended $`\pi `$-surface comprises a pointed closed oriented surface $`\mathrm{{\rm Y}}`$, a colored $`\pi `$-marking $`P\mathrm{{\rm Y}}`$, and a Lagrangian space $`\lambda =\lambda (\mathrm{{\rm Y}})H_1(\mathrm{{\rm Y}};)`$. In the case $`P=\mathrm{}`$ we obtain an extended $`\pi `$-surface without marks as above. A disjoint union of a finite number of extended $`\pi `$-surfaces is an extended $`\pi `$-surface in the obvious way. The empty set is considered as an empty extended $`\pi `$-surface.
A weak $`e`$-homeomorphism of extended $`\pi `$-surfaces $`(\mathrm{{\rm Y}},P,g,u,\lambda )(\mathrm{{\rm Y}}^{},P^{},g^{},u^{}`$, $`\lambda ^{})`$ is a degree $`+1`$ homeomorphism of pairs $`f:(\mathrm{{\rm Y}},P)(\mathrm{{\rm Y}}^{},P^{})`$ such that
\- $`f`$ preserves the base points, the signs of the marked points and their tangent directions;
\- $`g^{}f=g:\mathrm{{\rm Y}}\backslash PX`$ and for any path $`\gamma `$ in $`\mathrm{{\rm Y}}\backslash P`$ leading from a base point to $`\stackrel{~}{P}`$ we have $`u_{f\gamma }^{}=u_\gamma `$.
A weak $`e`$-homeomorphism $`f`$ as above is called an $`e`$-homeomorphism if the induced isomorphism $`f_{}:H_1(\mathrm{{\rm Y}};)H_1(\mathrm{{\rm Y}}^{};)`$ maps $`\lambda `$ onto $`\lambda ^{}`$.
For any extended $`\pi `$-surface $`\mathrm{{\rm Y}}`$, the opposite $`e`$-surface $`\mathrm{{\rm Y}}`$ is obtained from $`\mathrm{{\rm Y}}`$ by reversing the orientation of $`\mathrm{{\rm Y}}`$ and multiplying the signs of all the marked points by $`1`$ while keeping the rest of the data. Clearly, $`(\mathrm{{\rm Y}})=\mathrm{{\rm Y}}`$. The transformation $`\mathrm{{\rm Y}}\mathrm{{\rm Y}}`$ is natural in the sense that any (weak) $`e`$-homeomorphism $`f:\mathrm{{\rm Y}}\mathrm{{\rm Y}}^{}`$ gives rise to a (weak) $`e`$-homeomorphism $`f:\mathrm{{\rm Y}}\mathrm{{\rm Y}}^{}`$ which coincides with $`f`$ as a mapping.
8.3. Homotopy modular functor. The modular crossed $`\pi `$-category $`𝒞`$ gives rise to a 2-dimensional homotopy modular functor $`𝒯=𝒯_𝒞`$. This functor assigns
\- to each extended $`\pi `$-surface $`\mathrm{{\rm Y}}`$ a projective $`K`$-module of finite type $`𝒯(\mathrm{{\rm Y}})`$;
\- to each weak $`e`$-homeomorphism of extended $`\pi `$-surfaces $`f:\mathrm{{\rm Y}}\mathrm{{\rm Y}}^{}`$ an isomorphism $`f_\mathrm{\#}:𝒯(\mathrm{{\rm Y}})𝒯(\mathrm{{\rm Y}}^{})`$.
A construction of $`𝒯`$ will be outlined in Section 10. We state here a few simple properties of $`𝒯`$:
(8.3.1) for disjoint extended $`\pi `$-surfaces $`\mathrm{{\rm Y}},\mathrm{{\rm Y}}^{}`$, there is a natural isomorphism $`𝒯(\mathrm{{\rm Y}}\mathrm{{\rm Y}}^{})=𝒯(\mathrm{{\rm Y}})_K𝒯(\mathrm{{\rm Y}}^{})`$;
(8.3.2) $`𝒯(\mathrm{})=K`$;
(8.3.3) the isomorphism $`f_\mathrm{\#}`$ associated to a weak $`e`$-homeomorphism $`f`$ is invariant under isotopy of $`f`$ in the class of weak $`e`$-homeomorphisms;
(8.3.4) for any weak $`e`$-homeomorphisms $`f:\mathrm{{\rm Y}}\mathrm{{\rm Y}}^{}`$ and $`f^{}:\mathrm{{\rm Y}}^{}\mathrm{{\rm Y}}^{\prime \prime }`$, we have
$$(f^{}f)_\mathrm{\#}=(𝒟\mathrm{\Delta }_{}^1)^{\mu (f_{}(\lambda (\mathrm{{\rm Y}})),\lambda (\mathrm{{\rm Y}}^{}),(f^{})_{}^1(\lambda (\mathrm{{\rm Y}}^{\prime \prime })))}f_\mathrm{\#}^{}f_\mathrm{\#}$$
$`(8.3.a)`$
where $`f_{},f_{}^{}`$ denote the action of $`f,f^{}`$ in the real 1-homology and $`\mu `$ denotes the Maslov index of triples of Lagrangian subspaces of $`H_1(\mathrm{{\rm Y}}^{};)`$;
(8.3.5) for any extended $`\pi `$-surface $`\mathrm{{\rm Y}}`$, there is a non-degenerate bilinear pairing $`d_\mathrm{{\rm Y}}:𝒯(\mathrm{{\rm Y}})_K𝒯(\mathrm{{\rm Y}})K`$. The pairings $`\{d_\mathrm{{\rm Y}}\}_\mathrm{{\rm Y}}`$ are natural with respect to weak $`e`$-homeomorphisms, multiplicative with respect to disjoint unions, and symmetric in the sense that $`d_\mathrm{{\rm Y}}`$ is the composition of $`d_\mathrm{{\rm Y}}`$ with the standard flip $`𝒯(\mathrm{{\rm Y}})_K𝒯(\mathrm{{\rm Y}})𝒯(\mathrm{{\rm Y}})_K𝒯(\mathrm{{\rm Y}})`$.
The pairing $`d_\mathrm{{\rm Y}}`$ can be used to identify $`(𝒯(\mathrm{{\rm Y}}))^{}=\text{Hom}_K(𝒯(\mathrm{{\rm Y}}),K)`$ with $`𝒯(\mathrm{{\rm Y}})`$.
If two extended $`\pi `$-surfaces $`\mathrm{{\rm Y}},\mathrm{{\rm Y}}^{}`$ differ only by the choice of a Lagrangian space in homology then the identity homeomorphism $`\text{id}:\mathrm{{\rm Y}}\mathrm{{\rm Y}}^{}`$ is a weak $`e`$-homeomorphism and defines thus an isomorphism $`𝒯(\mathrm{{\rm Y}})𝒯(\mathrm{{\rm Y}}^{})`$. This together with (8.3.a) shows that the projective space associated with $`𝒯(\mathrm{{\rm Y}})`$ does not depend on the choice of $`\lambda (\mathrm{{\rm Y}})`$. Note also that the numerical factor in (8.3.a) equals to $`1`$ if $`f`$ or $`f^{}`$ is an $`e`$-homeomorphism: in this case $`\mu (f_{}(\lambda (\mathrm{{\rm Y}})),\lambda (\mathrm{{\rm Y}}^{}),(f^{})_{}^1(\lambda (\mathrm{{\rm Y}}^{\prime \prime })))=0`$.
A connected extended $`\pi `$-surface and the corresponding module can be explicitly described (at least up to isomorphism) as follows. Let $`\mathrm{{\rm Y}}`$ be a closed connected oriented surface of genus $`n0`$ with base point $`z`$ and $`m0`$ marked points $`P=\{p_1,\mathrm{},p_m\}\mathrm{{\rm Y}}\backslash \{z\}`$. Let $`\epsilon _r=\pm 1`$ be the sign of $`p_r`$. For $`r=1,\mathrm{},m`$, choose an arc $`\gamma _r`$ in $`\mathrm{{\rm Y}}\backslash P`$ leading from $`z`$ to $`\stackrel{~}{p}_r`$. We assume that these $`m`$ arcs are embedded and disjoint except in their common endpoint $`z`$. Recall the homotopy class $`\mu _{\gamma _r}\pi _1(\mathrm{{\rm Y}}\backslash P,z)`$ of the loop encircling $`p_r`$, see Section 8.2. The group $`\pi _1(\mathrm{{\rm Y}}\backslash P,z)`$ is known to be generated by $`\mu _{\gamma _1},\mathrm{},\mu _{\gamma _m}`$ and $`2n`$ elements $`a_1,b_1,\mathrm{},a_n,b_n`$ subject to the only relation
$$(\mu _{\gamma _1})^{\epsilon _1}\mathrm{}(\mu _{\gamma _m})^{\epsilon _m}[a_1,b_1]\mathrm{}[a_n,b_n]=1$$
$`(8.3.b)`$
where $`[a,b]=aba^1b^1`$. Choose $`m+2n`$ elements $`\mu _1,\mathrm{},\mu _m,\alpha _1,\beta _1,\mathrm{},\alpha _n,\beta _n\pi `$ satisfying
$$(\mu _1)^{\epsilon _1}\mathrm{}(\mu _m)^{\epsilon _m}[\alpha _1,\beta _1]\mathrm{}[\alpha _n,\beta _n]=1.$$
Then the formulas $`\mu _{\gamma _r}\mu _r,a_s\alpha _s,b_s\beta _s`$ with $`r=1,\mathrm{},m;s=1,\mathrm{},n`$ define a group homomorphism $`\pi _1(\mathrm{{\rm Y}}\backslash P,z)\pi `$ or equivalently a map, $`g:\mathrm{{\rm Y}}\backslash PX`$. This makes $`P`$ a $`\pi `$-marking. For any sequence of objects $`\{U_r𝒞_{\mu _r}\}_{r=1}^m`$, there is a unique coloring $`u`$ of $`P`$ such that $`u_{\gamma _r}=U_r`$ for all $`r`$. The linear subspace $`\lambda `$ of $`H_1(\mathrm{{\rm Y}};)`$ generated by the homology classes of $`a_1,\mathrm{},a_n`$ is a Lagrangian space. The tuple $`(\mathrm{{\rm Y}},P,g,u,\lambda )`$ is an extended $`\pi `$-surface. Recall the notation $`U_r^+=U_r,U_r^{}=U_r^{}`$. Then
$$𝒯(\mathrm{{\rm Y}},P,g,u,\lambda )$$
$`(8.3.c)`$
$$=\underset{i_1I_{\alpha _1},\mathrm{},i_nI_{\alpha _n}}{}\text{Hom}_𝒞(\text{1}\text{ I},(U_1)^{\epsilon _1}\mathrm{}(U_m)^{\epsilon _m}\underset{s=1}{\overset{n}{}}\left(V_{i_s}^{\alpha _s}(\phi _{\beta _s}(V_{i_s}^{\alpha _s}))^{}\right))$$
where for $`\alpha \pi `$ we denote by $`I_\alpha `$ the set of isomorphism classes of simple objects in the category $`𝒞_\alpha `$ and denote by $`\{V_i^\alpha \}_{iI_\alpha }`$ certain representatives of these classes. Note that the group of the isotopy classes of (weak) $`e`$-homeomorphisms of $`(\mathrm{{\rm Y}},P,g,u,\lambda )`$ onto itself acts (projectively) on $`𝒯(\mathrm{{\rm Y}},P,g,u,\lambda )`$.
It is clear that any extended $`\pi `$-surface $`\mathrm{{\rm Y}}`$ is weakly $`e`$-homeomorphic to an extended $`\pi `$-surface $`(\mathrm{{\rm Y}},P,g,u,\lambda )`$ of the type described in the previous paragraph. Formula (8.3.c) allows thus to compute $`𝒯(\mathrm{{\rm Y}})`$.
Most properties of the modular functors known for $`\pi =1`$ extend to the general case. We leave a detailed discussion to another place. Note only that the splitting formula for the modules of conformal blocks along simple closed curves on surfaces extends to our setting. Instead of giving a detailed statement we formulate the key algebraic fact underlying this formula.
8.4. Lemma. Let $`\alpha \pi `$ and $`\{V_i^\alpha \}_{iI_\alpha }`$ be representatives of the isomorphism classes of simple objects in the category $`𝒞_\alpha `$. For any objects $`V𝒞_\alpha ,W𝒞_{\alpha ^1}`$, there is a canonical isomorphism
$$\text{Hom}_𝒞(\text{1}\text{ I},VW)=\underset{iI_\alpha }{}\left(\text{Hom}_𝒞(\text{1}\text{ I},V(V_i^\alpha )^{})_K\text{Hom}_𝒞(\text{1}\text{ I},V_i^\alpha W)\right).$$
Proof. This is equivalent to
$$\text{Hom}_𝒞(W^{},V)=\underset{iI_\alpha }{}\left(\text{Hom}_𝒞(V_i^\alpha ,V)_K\text{Hom}_𝒞(W^{},V_i^\alpha )\right).$$
$`(8.4.a)`$
This isomorphism sends $`hh^{}`$ with $`h\text{Hom}_𝒞(V_i^\alpha ,V),h^{}\text{Hom}_𝒞(W^{},V_i^\alpha )`$ into $`hh^{}`$. Equality (8.4.a) involves only morphisms in the category $`𝒞_\alpha `$ and follows from axiom (6.1.3), cf. \[Tu2, Lemma II.4.2.2\].
8.5. Action of $`\pi `$. There is a canonical left action of $`\pi `$ on extended $`\pi `$-surfaces and (weak) $`e`$-homeomorphisms. The action of $`\alpha \pi `$ transforms an extended $`\pi `$-surface $`\mathrm{{\rm Y}}=(\mathrm{{\rm Y}},P,g,u,\lambda )`$ into $`{}_{}{}^{\alpha }\mathrm{{\rm Y}}=(\mathrm{{\rm Y}},P,\alpha _{}g,\phi _\alpha u,\lambda )`$ where $`\alpha _{}:(X,x)(X,x)`$ is the map inducing the conjugation $`\beta \alpha \beta \alpha ^1`$ in $`\pi _1(X,x)=\pi `$ and $`\phi _\alpha `$ is the given action of $`\alpha `$ on $`𝒞`$. For a (weak) $`e`$-homeomorphism of extended $`\pi `$-surfaces $`f:\mathrm{{\rm Y}}\mathrm{{\rm Y}}^{}`$, we define $`{}_{}{}^{\alpha }f`$ as the same map $`f`$ viewed as a (weak) $`e`$-homeomorphism $`{}_{}{}^{\alpha }\mathrm{{\rm Y}}{}_{}{}^{\alpha }\mathrm{{\rm Y}}_{}^{}`$.
The modular functor $`𝒯`$ can be enriched as follows: for every extended $`\pi `$-surface $`\mathrm{{\rm Y}}`$ and each $`\alpha \pi `$ there is a canonical isomorphism $`\alpha _{}:𝒯(\mathrm{{\rm Y}})𝒯({}_{}{}^{\alpha }\mathrm{{\rm Y}})`$, see Section 10.3. For $`\alpha ,\beta \pi `$ we have $`(\alpha \beta )_{}=\alpha _{}\beta _{}`$. For a weak $`e`$-homeomorphism of extended $`\pi `$-surfaces $`f:\mathrm{{\rm Y}}\mathrm{{\rm Y}}^{}`$, we have a commutative diagram
$$\begin{array}{ccc}𝒯(\mathrm{{\rm Y}})& \stackrel{f_\mathrm{\#}}{}& 𝒯(\mathrm{{\rm Y}}^{})\\ \alpha _{}& & \alpha _{}& & \\ 𝒯({}_{}{}^{\alpha }\mathrm{{\rm Y}})& \stackrel{({}_{}{}^{\alpha }f)_\mathrm{\#}}{}& 𝒯({}_{}{}^{\alpha }\mathrm{{\rm Y}}_{}^{}).\end{array}$$
8.6. Computations on the torus. Consider formula (8.3.c) in the case where $`\mathrm{{\rm Y}}`$ is a torus without marked points, i.e., $`m=0,n=1`$. The group $`\pi _1(\mathrm{{\rm Y}},z)`$ is generated by two elements $`a,b`$ subject to $`[a,b]=1`$. A map $`g:\mathrm{{\rm Y}}X`$ is determined by $`g_\mathrm{\#}(a)=\alpha ,g_\mathrm{\#}(b)=\beta `$ where $`\alpha ,\beta `$ are commuting elements of $`\pi `$. The 1-dimensional subspace $`\lambda `$ of $`H_1(\mathrm{{\rm Y}};)`$ generated by the homology class of $`a`$ is a Lagrangian space. By (8.3.c),
$$𝒯(\mathrm{{\rm Y}},g,\lambda )=\underset{iI_\alpha }{}\text{Hom}_𝒞(\text{1}\text{ I},V_i^\alpha (\phi _\beta (V_i^\alpha ))^{})=\underset{iI_\alpha }{}\text{Hom}_𝒞(\phi _\beta (V_i^\alpha ),V_i^\alpha ).$$
Observe that for any simple objects $`U,U^{}`$ of $`𝒞`$,
$$\text{Hom}_𝒞(U,U^{})=\{\begin{array}{cc}K,\text{if}U\text{is\hspace{0.17em} isomorphic \hspace{0.17em} to}U^{},\hfill & \\ 0,\text{otherwise}.\hfill & \end{array}$$
Note also that $`\phi _\beta `$ maps $`𝒞_\alpha `$ into itself and induces a permutation on the set $`I_\alpha `$. Therefore $`𝒯(\mathrm{{\rm Y}},g,\lambda )=K^N`$ where $`N=\text{card}\{iI_\alpha |\phi _\beta (i)=i\}`$. For instance if $`\beta =1`$ then $`N=\text{card}(I_\alpha )`$. More generally, if $`(\mathrm{{\rm Y}},g,\lambda )`$ is an extended $`\pi `$-torus without marks such that the image of $`g_\mathrm{\#}:\pi _1(\mathrm{{\rm Y}})\pi `$ is a cyclic group generated by $`\alpha \pi `$, then
$$𝒯(\mathrm{{\rm Y}},g,\lambda )=K^{\text{card}(I_\alpha )}.$$
Indeed, we can choose the generators $`a,b\pi _1(\mathrm{{\rm Y}})`$ as above so that $`\alpha =g_\mathrm{\#}(a),\beta =g_\mathrm{\#}(b)=1`$.
Consider now the case where $`\mathrm{{\rm Y}}`$ is a torus with one marked point $`p`$, i.e., $`m=n=1`$. We use the notation of Section 8.3 but omit the index 1 and write $`p,\epsilon ,\gamma ,a,b,\mu ,\alpha ,\beta ,U`$ for $`p_1,\epsilon _1,\gamma _1,a_1,b_1,\mu _1,\alpha _1,\beta _1,U_1`$. Thus $`P=\{p\}`$ and the group $`\pi _1(\mathrm{{\rm Y}}\backslash P,z)`$ is generated by three elements $`\mu _\gamma ,a,b`$ subject to $`(\mu _\gamma )^\epsilon [a,b]=1`$. The map $`g:\mathrm{{\rm Y}}\backslash PX`$ is determined by $`g_\mathrm{\#}(\mu _\gamma )=\mu ,g_\mathrm{\#}(a)=\alpha ,g_\mathrm{\#}(b)=\beta `$ where $`\mu ,\alpha ,\beta \pi `$ satisfy $`\mu ^\epsilon [\alpha ,\beta ]=1`$. The coloring $`u`$ of $`P`$ is determined by $`u_\gamma =U`$. The 1-dimensional subspace $`\lambda `$ of $`H_1(\mathrm{{\rm Y}};)`$ generated by the homology class of $`a`$ is a Lagrangian space. Then
$$𝒯(\mathrm{{\rm Y}},P,g,u,\lambda )=\underset{iI_\alpha }{}\text{Hom}_𝒞(\text{1}\text{ I},U^\epsilon V_i^\alpha (\phi _\beta (V_i^\alpha ))^{}).$$
In general, taking different systems of generators (8.3.b) of $`\pi _1(\mathrm{{\rm Y}}\backslash P,z)`$ we can obtain different descriptions of one and the same extended $`\pi `$-surface. The modules appearing on the right-hand side of (8.3.c) are then isomorphic. For instance, in the case $`m=n=1`$ the identity $`[a,b]=[aba^1,a^1]`$ implies that the group $`\pi _1(\mathrm{{\rm Y}}\backslash P,z)`$ is generated by three elements $`\mu _\gamma ,aba^1,a^1`$ subject to $`(\mu _\gamma )^\epsilon [aba^1,a^1]=1`$. The same map $`g:\mathrm{{\rm Y}}\backslash PX`$ as above is determined by $`g_\mathrm{\#}(\mu _\gamma )=\mu ,g_\mathrm{\#}(aba^1)=\alpha \beta \alpha ^1,g_\mathrm{\#}(a^1)=\alpha ^1`$. The same coloring $`u`$ is determined by $`u_\gamma =U`$. Thus under these two choices of generators of $`\pi _1(\mathrm{{\rm Y}}\backslash P,z)`$ we obtain two extended $`\pi `$-surfaces which differ only in the Lagrangian spaces generated by the homology classes of $`a`$ and $`b`$, respectively. The corresponding $`K`$-modules are then isomorphic:
$$\underset{iI_\alpha }{}\text{Hom}_𝒞(\text{1}\text{ I},U^\epsilon V_i^\alpha (\phi _\beta (V_i^\alpha ))^{})$$
$`(8.6.a)`$
$$=\underset{jI_{\alpha \beta \alpha ^1}}{}\text{Hom}_𝒞(\text{1}\text{ I},U^\epsilon V_j^{\alpha \beta \alpha ^1}(\phi _{\alpha ^1}(V_j^{\alpha \beta \alpha ^1}))^{})$$
$$=\underset{kI_\beta }{}\text{Hom}_𝒞(\text{1}\text{ I},U^\epsilon \phi _\alpha (V_k^\beta )(V_k^\beta )^{})$$
where the last equality follows from the fact that $`\phi _\alpha `$ induces a bijection $`I_\beta I_{\alpha \beta \alpha ^1}`$.
9. A 2-dimensional HQFT
We discuss the 2-dimensional homotopy quantum field theory (HQFT) underlying the homotopy modular functor of Section 8.
9.1. Preliminaries on 2-dimensional HQFT’s. We recall briefly the notion of a 2-dimensional HQFT referring for details to \[Tu3\]. A 2-dimensional HQFT with target $`X=K(\pi ,1)`$ has two ingredients: a family $`\{L_\alpha \}_{\alpha \pi }`$ of projective $`K`$-modules of finite type numerated by the elements of $`\pi `$ and a function $`\tau `$ assigning to surfaces equipped with a map to $`X`$ certain $`K`$-linear homomorphisms. More precisely, let $`W`$ be a compact oriented surface with pointed oriented boundary. A component $`c`$ of $`W`$ is positive (resp. negative) if its given orientation is opposite to the one induced from $`W`$ (resp. coincides with the one induced from $`W`$). We write $`\epsilon _c=+`$ and $`\epsilon _c=`$, respectively. We view $`W`$ as a cobordism between $`_{c,\epsilon _c=}c`$ and $`_{c,\epsilon _c=+}c`$. Let $`g:WX`$ be a map sending the base points of all the components of $`W`$ into a base point $`xX`$. Each component $`cW`$ is pointed and oriented so that the map $`g|_c:cX`$ is a loop in $`(X,x)`$. Denote its homotopy class by $`\alpha _c\pi `$. The function $`\tau `$ assigns to each such pair $`(W,g)`$ a $`K`$-homomorphism
$$\tau (W,g):\underset{c,\epsilon _c=}{}L_{\alpha _c}\underset{c,\epsilon _c=+}{}L_{\alpha _c}.$$
$`(9.1.a)`$
The homomorphism $`\tau (W,g)`$ should satisfy a few axioms. It should be preserved under homotopy of $`g`$ relative to the base points on $`W`$. It should be multiplicative under disjoint union of cobordisms. The gluing of cobordisms should correspond to composition of homomorphisms. The value of $`\tau `$ on a cylinder should be the identity homomorphism.
In particular, if $`W`$ is a closed oriented surface endowed with a map $`g:WX`$ then the homomorphism $`\tau (W,g):KK`$ is multiplication by a certain element of $`K`$. This element, denoted also by $`\tau (W,g)`$, is a homotopy invariant of $`g`$.
The algebraic counterparts of 2-dimensional HQFT’s are crossed $`\pi `$-algebras over $`K`$. A $`\pi `$-graded algebra or, briefly, a $`\pi `$-algebra over $`K`$ is an associative algebra $`L`$ over $`K`$ endowed with a splitting $`L=_{\alpha \pi }L_\alpha `$ such that each $`L_\alpha `$ is a projective $`K`$-module of finite type, $`L_\alpha L_\beta L_{\alpha \beta }`$ for any $`\alpha ,\beta \pi `$, and $`L`$ has a (right and left) unit $`1_LL_1`$ where $`1`$ is the neutral element of $`\pi `$. A crossed $`\pi `$-algebra over $`K`$ is a $`\pi `$-algebra $`L`$ over $`K`$ endowed with a symmetric $`K`$-bilinear form (inner product) $`\eta :L\times LK`$ and a group homomorphism $`\phi :\pi \text{Aut}(L)`$ satisfying the following conditions:
(9.1.1) $`\eta (L_\alpha L_\beta )=0`$ if $`\alpha \beta 1`$; the restriction of $`\eta `$ to $`L_\alpha L_{\alpha ^1}`$ is non-degenerate for all $`\alpha \pi `$, and $`\eta (ab,c)=\eta (a,bc)`$ for any $`a,b,cL`$;
(9.1.2) for all $`\alpha \pi `$, $`\phi _\alpha =\phi (\alpha )`$ is an algebra automorphism of $`L`$ preserving $`\eta `$ and such that $`\phi _\alpha (L_\beta )L_{\alpha \beta \alpha ^1}`$ for all $`\beta \pi `$;
(9.1.3) $`\phi _\alpha |_{L_\alpha }=\text{id}`$, for all $`\alpha \pi `$ and $`\phi _\alpha (b)a=ab`$ for any $`aL_\alpha ,bL_\beta `$;
(9.1.4) for any $`\alpha ,\beta \pi `$ and any $`cL_{\alpha \beta \alpha ^1\beta ^1}`$,
$$\text{Tr}(c\phi _\beta :L_\alpha L_\alpha )=\text{Tr}(\phi _{\alpha ^1}c:L_\beta L_\beta ).$$
Here the homomorphism on the left-hand side sends any $`aL_\alpha `$ into $`c\phi _\beta (a)L_\alpha `$ and the homomorphism on the right-hand side sends any $`bL_\beta `$ into $`\phi _{\alpha ^1}(cb)L_\beta `$.
According to \[Tu3\], for any 2-dimensional HQFT $`(\{L_\alpha \}_{\alpha \pi },\tau )`$ with target $`X=K(\pi ,1)`$ the direct sum $`L=_{\alpha \pi }L_\alpha `$ has the structure of a crossed $`\pi `$-algebra. Moreover, this establishes a bijective correspondence between the isomorphism classes of $`2`$-dimensional HQFT’s with target $`K(\pi ,1)`$ and the isomorphism classes of crossed $`\pi `$-algebras.
9.2. Underlying 2-dimensional HQFT. The 2-dimensional homotopy modular functor $`𝒯=𝒯_𝒞`$ derived from a modular crossed $`\pi `$-category $`𝒞`$ has an “underlying” 2-dimensional HQFT $`(\{L_\alpha \}_{\alpha \pi },\tau )`$. Here $`L=_{\alpha \pi }L_\alpha `$ is the algebra of colors of $`𝒞`$ defined in Section 5.3. The values of $`\tau `$ are obtained, roughly speaking, by taking the dimensions of the modules assigned by $`𝒯`$ to the surfaces. In particular, for a closed oriented surface $`W`$ and a map $`g:WX`$ we have $`\tau (W,g)=\text{Dim}𝒯(\mathrm{{\rm Y}}_{W,g})K`$ where $`\mathrm{{\rm Y}}_{W,g}`$ is an extended $`\pi `$-surface without marks obtained from $`W`$ by choosing an arbitrary Lagrangian space in $`H_1(W;)`$, choosing arbitrary base points on the components of $`W`$ and deforming $`g`$ so that it maps these base points into $`x`$. According to the results of Section 8, the isomorphism class of the module $`𝒯(\mathrm{{\rm Y}}_{W,g})`$ and therefore its dimension do not depend on the choices in the construction of $`\mathrm{{\rm Y}}_{W,g}`$.
To describe the homomorphism (9.1.a) for any pair $`(W,g:WX)`$ as in Section 9.1 we proceed as follows. For each component $`c`$ of $`W`$, the $`K`$-module $`L_{\alpha _c}`$ is free with basis given by Lemma 6.2. (Recall that $`\alpha _c\pi `$ is represented by the loop $`g|_c`$ in $`(X,x)`$.) We present $`\tau (W,g)`$ by a matrix with respect to the tensor products of these bases. Fix for all $`c`$ a simple object $`V_c𝒞_{\alpha _c}`$. Consider the basis elements
$$\underset{c,\epsilon _c=}{}V_c\underset{c,\epsilon _c=}{}L_{\alpha _c}\text{and }\underset{c,\epsilon _c=+}{}V_c\underset{c,\epsilon _c=+}{}L_{\alpha _c}.$$
It suffices to describe the corresponding matrix term of $`\tau (W,g)`$. We denote it by $`\tau (W,g)|\{V_c\}_c`$.
We upgrade $`W`$ to an extended $`\pi `$-surface with marks as follows. Assume for simplicity that $`W`$ is connected and choose a base point $`zW\backslash W`$. We deform $`g:WX`$ relative to $`W`$ so that $`g(z)=x`$. We cap $`W`$ with 2-discs by gluing to each component $`c`$ of $`W`$ a copy of the unit complex 2-disc $`D=\{a||a|1\}`$. The gluing is effected so that the point $`1D`$ is identified with the base point of $`c`$. This results in a closed orientable surface $`\mathrm{{\rm Y}}`$ which we provide with orientation extending the one in $`W\mathrm{{\rm Y}}`$. The centers (corresponding to $`a=0D`$) of the glued 2-discs form a finite set $`P\mathrm{{\rm Y}}`$. We provide each $`pP`$ with sign $`+`$ and tangent direction corresponding to $`_+`$. We agree that pushing $`pP`$ along its tangent direction we obtain the base point $`\stackrel{~}{p}`$ of the corresponding component, $`c_p`$, of $`W`$.
Clearly, $`W`$ is a deformation retract of $`\mathrm{{\rm Y}}\backslash P`$. Therefore the map $`g:WX`$ extends to a map $`\mathrm{{\rm Y}}\backslash P`$ denoted also by $`g`$. This makes $`P`$ a $`\pi `$-marking on $`\mathrm{{\rm Y}}`$. We color $`P`$ as follows. For a path $`\gamma :[0,1]\mathrm{{\rm Y}}\backslash P`$ leading from $`z`$ to $`\stackrel{~}{p}`$, the path $`g\gamma `$ is a loop in $`(X,x)`$ so that we can consider its homotopy class $`[g\gamma ]\pi `$. Set
$$u_\gamma =\{\begin{array}{cc}\phi _{[g\gamma ]}(V_{c_p}),\text{if}\epsilon _{c_p}=+1,\hfill & \\ \phi _{[g\gamma ]}(V_{c_p}^{}),\text{if}\epsilon _{c_p}=1.\hfill & \end{array}$$
$`(9.2.a)`$
Conditions (i) - (iii) of Section 8.2 are straightforward. In particular, Condition (ii) follows from the equality
$$g_\mathrm{\#}(\mu _\gamma )=[g\gamma ](\alpha _{c_p})^{\epsilon _{c_p}}[g\gamma ]^1.$$
Choosing an arbitrary Lagrangian space $`\lambda H_1(\mathrm{{\rm Y}};)`$ we obtain an extended $`\pi `$-surface $`(\mathrm{{\rm Y}},g,P,u,\lambda )`$. Set
$$\tau (W,g)|\{V_c\}_c=\text{Dim}𝒯(\mathrm{{\rm Y}},g,P,u,\lambda ).$$
One can check that $`(\{L_\alpha \}_{\alpha \pi },\tau )`$ is a 2-dimensional HQFT with target $`X`$. We describe here the structure of a crossed $`\pi `$-algebra in $`L=_{\alpha \pi }L_\alpha `$ underlying this 2-dimensional HQFT. First, define a pairing $`\eta :L\times LK`$ by $`\eta (U,f,U^{},f^{})=\text{Tr}(ff^{})_{}`$ where $`U,f,U^{},f^{}`$ are arbitrary additive generators of $`L`$ as in Section 5.3 and $`(ff^{})_{}`$ denotes the endomorphism of $`\text{Hom}_𝒞(\text{1}\text{ I},UU^{})`$ sending each $`h\text{Hom}_𝒞(\text{1}\text{ I},UU^{})`$ into $`(ff^{})h`$. It is easy to check that $`\eta `$ is a well-defined bilinear pairing.
9.3. Theorem. The $`\pi `$-graded algebra $`L=_{\alpha \pi }L_\alpha `$ with form $`\eta `$ and the action $`\phi `$ of $`\pi `$ defined in Section 5.3 is a crossed $`\pi `$-algebra.
Proof. For each $`\alpha \pi `$, fix representatives $`\{V_i^\alpha \}_{iI_\alpha }`$ of the isomorphism classes of simple objects in the category $`𝒞_\alpha `$. By Lemma 6.2, $`L_\alpha `$ is a free $`K`$-module with basis $`\{V_i^\alpha \}_{iI_\alpha }`$.
Let us check (9.1.1). Observe that for any objects $`U,U^{}`$ of $`𝒞`$,
$$\eta (U,U^{})=\text{Dim}(\text{Hom}_𝒞(\text{1}\text{ I},UU^{}))=\text{Dim}(\text{Hom}_𝒞(U^{},U^{})).$$
In particular if the objects $`U,U^{}`$ are simple then
$$\eta (U,U^{})=\{\begin{array}{cc}1,\text{if}U^{}\text{is\hspace{0.17em} isomorphic \hspace{0.17em} to}U^{},\hfill & \\ 0,\text{otherwise}.\hfill & \end{array}$$
$`(9.3.a)`$
This implies that $`\eta (L_\alpha L_\beta )=0`$ if $`\alpha \beta 1`$ and the bases $`\{V_i^\alpha \}_{iI_\alpha }`$ and $`\{V_i^{\alpha ^1}\}_{iI_{\alpha ^1}}`$ are dual to each other with respect to $`\eta `$. Hence the non-degeneracy of $`\eta `$. Formula $`\eta (ab,c)=\eta (a,bc)`$ follows from the definition of $`\eta `$.
By (9.3.a) and Corollary 4.6, the form $`\eta `$ is symmetric. Formula (9.3.a) implies that $`\eta `$ is invariant under the action of $`\pi `$. All other conditions in (9.1.2), (9.1.3) were checked in Section 5.3.
To check (9.1.4) let $`c=U`$ where $`U`$ is an object of $`𝒞_{\alpha \beta \alpha ^1\beta ^1}`$. The homomorphism $`c\phi _\beta :L_\alpha L_\alpha `$ sends $`V_i^\alpha `$ into
$$U\phi _\beta (V_i^\alpha )=\underset{jI_\alpha }{}\mu _{V_j^\alpha ,U\phi _\beta (V_i^\alpha )}V_j^\alpha .$$
Therefore
$$\text{Tr}(c\phi _\beta )=\underset{iI_\alpha }{}\mu _{V_i^\alpha ,U\phi _\beta (V_i^\alpha )}$$
$`(9.3.b)`$
$$=\underset{iI_\alpha }{}\mu _{U\phi _\beta (V_i^\alpha ),V_i^\alpha }=\underset{iI_\alpha }{}\text{Dim}(\text{Hom}_𝒞(\text{1}\text{ I},U^{}V_i^\alpha (\phi _\beta (V_i^\alpha ))^{})).$$
A similar computation shows that
$$\text{Tr}(\phi _{\alpha ^1}c:L_\beta L_\beta )=\underset{kI_\beta }{}\text{Dim}(\text{Hom}_𝒞(\text{1}\text{ I},U^{}\phi _\alpha (V_k^\beta )(V_k^\beta )^{})).$$
$`(9.3.c)`$
By (8.6.a) (where $`U`$ should be replaced with $`U^{}`$), the right-hand sides of (9.3.b) and (9.3.c) are the dimensions of isomorphic $`K`$-modules and therefore they are equal.
9.4. Remarks. 1. The proof of (9.1.4) in Theorem 9.3 uses the existence of the homotopy modular functor associated with $`𝒞`$. It would be useful to have a direct algebraic proof.
2. The neutral component $`L_1`$ of the algebra of colors of $`𝒞`$ is known to be a semisimple commutative $`K`$-algebra (cf. \[Tu2, Section IV.12.4\]). More precisely, the algebra $`L_1`$ splits as a direct sum of $`\text{card}(I_1)`$ copies of $`K`$. This and the results of \[Tu3\] imply that if $`K`$ is a field of characteristic 0 then the HQFT $`(\{L_\alpha \}_{\alpha \pi },\tau )`$ is semi-cohomological, i.e., is determined by a finite family of $`K`$-valued weights and 2-dimensional cohomology classes of subgroups of $`\pi `$ of finite index.
10. A 3-dimensional HQFT
Each modular crossed $`\pi `$-category $`𝒞`$ gives rise to a 3-dimensional homotopy quantum field theory (HQFT) with target space $`K(\pi ,1)`$. As in the standard case $`\pi =1`$, this HQFT has several equivalent versions. We describe here one of these versions formulated in terms of extended $`\pi `$-manifolds.
Throughout Section 10 we fix a modular crossed $`\pi `$-category $`𝒞`$ and an Eilenberg-MacLane CW-complex $`X=K(\pi ,1)`$ with base point $`xX`$.
10.1. Extended $`\pi `$-manifolds with boundary. Recall the notion of a connected extended $`\pi `$-manifold without boundary introduced in Section 7.5. In the present context it is more convenient to use (pointed) maps to $`X=K(\pi ,1)`$ rather than group homomorphisms to $`\pi `$. Thus, instead of a group homomorphism $`\pi _1(M\backslash \mathrm{\Omega },z)\pi `$ as in Section 7.5, we shall consider a (pointed) map $`g:M\backslash \mathrm{\Omega }X`$ inducing this homomorphism. (In the definition of a coloring we should then replace $`g`$ by $`g_\mathrm{\#}`$.) After this minor modification of the definitions of Section 7.5 we can extend them to manifolds with boundary.
Let $`M`$ be a compact connected oriented 3-manifold with non-void pointed boundary. A ribbon graph $`\mathrm{\Omega }M`$ consists of a finite family of framed oriented embedded arcs, circles and coupons (the strata of $`\mathrm{\Omega }`$) such that
(i) the strata of $`\mathrm{\Omega }`$ are disjoint except that some endpoints of the arcs lie on the bases of the coupons;
(ii) all other endpoints of the arcs of $`\mathrm{\Omega }`$ lie on $`M`$ and form the (finite) set $`\mathrm{\Omega }M`$; this set does not contain the base points of $`M`$;
(iii) the framings of the strata form a framing of $`\mathrm{\Omega }`$, i.e., a continuous non-singular vector field on $`\mathrm{\Omega }`$ transversal to $`\mathrm{\Omega }`$; at $`\mathrm{\Omega }M`$ the framing is tangent to $`M`$;
(iv) on each coupon of $`\mathrm{\Omega }`$ the framing is transversal to the coupon and yields together with the orientation of the coupon the given orientation of $`M`$.
Slightly pushing $`\mathrm{\Omega }`$ along its framing we obtain a disjoint copy $`\stackrel{~}{\mathrm{\Omega }}`$ of $`\mathrm{\Omega }`$. Pushing a stratum $`t`$ of $`\mathrm{\Omega }`$ along the framing we obtain a stratum $`\stackrel{~}{t}`$ of $`\stackrel{~}{\mathrm{\Omega }}`$.
A $`\pi `$-graph in $`M`$ is a ribbon graph $`\mathrm{\Omega }M`$ endowed with a map $`g:M\backslash \mathrm{\Omega }X`$ which sends the base points of all the components of $`M`$ into $`xX`$. We consider $`g`$ up to homotopy constant on the base points of $`M`$.
Let $`(\mathrm{\Omega },g)`$ be a $`\pi `$-graph in $`M`$. A coloring of $`\mathrm{\Omega }`$ with respect to a base point $`z`$ of a component of $`M`$ comprises two functions $`u`$ and $`v`$. The function $`u`$ assigns to every arc or circle $`t`$ of $`\mathrm{\Omega }`$ and to every path $`\gamma :[0,1]M\backslash \mathrm{\Omega }`$ connecting $`z`$ to $`\gamma (1)\stackrel{~}{t}\stackrel{~}{\mathrm{\Omega }}`$ a certain object $`u_\gamma 𝒞_{g_\mathrm{\#}(\mu _\gamma )}`$ where $`\mu _\gamma \pi _1(M\backslash \mathrm{\Omega },z)`$ is the meridian of $`t`$ determined by $`\gamma `$ and $`g_\mathrm{\#}:\pi _1(M\backslash \mathrm{\Omega },z)\pi `$ is the homomorphism induced by $`g`$. The function $`v`$ assigns to every coupon $`Q`$ of $`\mathrm{\Omega }`$ and to every path $`\gamma `$ in $`M\backslash \mathrm{\Omega }`$ connecting $`z`$ to $`\gamma (1)\stackrel{~}{Q}\stackrel{~}{\mathrm{\Omega }}`$ a certain morphism $`v_\gamma `$ in $`𝒞_{g_\mathrm{\#}(\mu _\gamma )}`$. The functions $`u,v`$ should satisfy the same conditions as in Section 4.1 (where $`g`$ should be replaced with $`g_\mathrm{\#}`$).
A coloring $`(u,v)`$ of $`\mathrm{\Omega }`$ with respect to the base point $`z`$ of a component of $`M`$ can be canonically transported to a coloring of $`\mathrm{\Omega }`$ with respect to the base point, $`z^{}`$, of any other component of $`M`$. (Here it is essential that $`M`$ is connected). Namely, choose a path $`\rho :[0,1]M\backslash \mathrm{\Omega }`$ with $`\rho (0)=z,\rho (1)=z^{}`$. For an arc or circle $`t`$ of $`\mathrm{\Omega }`$ and a path $`\gamma :[0,1]M\backslash \mathrm{\Omega }`$ connecting $`z^{}`$ to $`\gamma (1)\stackrel{~}{t}`$ set
$$u_\gamma =(\phi _{[g\rho ]})^1(u_{\rho \gamma }).$$
Here the path $`g\rho `$ is a loop in $`(X,x)`$ so that we can consider its homotopy class $`[g\rho ]\pi `$. The path $`\rho \gamma `$ connects $`z`$ to $`\gamma (1)\stackrel{~}{t}`$ so that $`u_{\rho \gamma }`$ is an object of $`𝒞_{g_\mathrm{\#}(\mu _{\rho \gamma })}`$. We have $`\mu _{\rho \gamma }=\rho \mu _\gamma \rho ^1`$ and so $`g_\mathrm{\#}(\mu _{\rho \gamma })=[g\rho ]g_\mathrm{\#}(\mu _\gamma )[g\rho ]^1`$. Hence the functor $`(\phi _{[g\rho ]})^1`$ maps $`𝒞_{g_\mathrm{\#}(\mu _{\rho \gamma })}`$ into $`𝒞_{g_\mathrm{\#}(\mu _\gamma )}`$ and $`u_\gamma `$ is an object of $`𝒞_{g_\mathrm{\#}(\mu _\gamma )}`$. Similarly, for a coupon $`Q`$ of $`\mathrm{\Omega }`$ and a path $`\gamma :[0,1]M\backslash \mathrm{\Omega }`$ connecting $`z^{}`$ to $`\gamma (1)\stackrel{~}{Q}`$ set
$$v_\gamma =(\phi _{[g\rho ]})^1(v_{\rho \gamma }).$$
It is easy to check that these formulas define a coloring of $`\mathrm{\Omega }`$ with respect to $`z^{}`$. It follows from definitions that this coloring does not depend on the choice of $`\rho `$. Thus, a coloring of $`\mathrm{\Omega }`$ with respect to the base point of one component of $`M`$ canonically extends to a system of colorings of $`\mathrm{\Omega }`$ with respect to the base points of all components of $`M`$. This system is called a coloring of $`\mathrm{\Omega }`$. To specify a coloring of $`\mathrm{\Omega }`$ it is enough to specify a coloring of $`\mathrm{\Omega }`$ with respect to one base point, the colorings of $`\mathrm{\Omega }`$ with respect to all the other base points are obtained by the transport as above.
A tuple $`(M,\mathrm{\Omega },g,u,v,\lambda )`$ consisting of a colored $`\pi `$-graph $`(\mathrm{\Omega },g,u,v)`$ in $`M`$ and a Lagrangian space $`\lambda H_1(M;)`$ is called a connected extended $`\pi `$-manifold with boundary.
Given a connected extended $`\pi `$-manifold $`(M,\mathrm{\Omega },g,u,v,\lambda )`$, the surface $`M`$ becomes an extended $`\pi `$-surface as follows. By definition, $`M`$ is pointed. We provide $`M`$ with orientation induced by the one in $`M`$ (see Section 7.2 for our orientation convention). Set $`P=\mathrm{\Omega }M`$. Clearly, $`P`$ is a finite subset of $`M`$ disjoint from the base points. A point $`pP`$ is provided with sign $``$ if the adjacent arc of $`\mathrm{\Omega }`$ is oriented towards $`p`$ and with sign $`+`$ otherwise. We provide $`p`$ with tangent direction generated by the framing vector at $`p`$ (by (iii) this vector is tangent to $`M`$). We can assume that $`\stackrel{~}{P}=\stackrel{~}{\mathrm{\Omega }}M`$. Denote by $`g`$ the restriction of $`g:M\backslash \mathrm{\Omega }X`$ to $`M\backslash P`$. The pair $`(P,g)`$ is a $`\pi `$-marking on $`M`$. We define its coloring $`u`$ as follows. Every path $`\gamma :[0,1]M\backslash P`$ leading from a base point to $`\stackrel{~}{P}`$ is a path in $`M\backslash \mathrm{\Omega }`$ leading to $`\stackrel{~}{\mathrm{\Omega }}`$. Set $`(u)_\gamma =u_\gamma 𝒞`$. It is clear that the tuple $`(M,P,g,u,\lambda )`$ is an extended $`\pi `$-surface. By definition, the boundary of an extended $`\pi `$-manifold without boundary is an empty surface.
Let $`(M,\mathrm{\Omega },g,u,v,\lambda )`$ and $`(M^{},\mathrm{\Omega }^{},g^{},u^{},v^{},\lambda ^{})`$ be two connected extended $`\pi `$-manifolds with boundary. A weak $`e`$-homeomorphism
$$(M,\mathrm{\Omega },g,u,v,\lambda )(M^{},\mathrm{\Omega }^{},g^{},u^{},v^{},\lambda ^{})$$
is a degree $`+1`$ pointed homeomorphism of pairs $`f:(M,\mathrm{\Omega })(M^{},\mathrm{\Omega }^{})`$ preserving the framing, the orientation and the splitting of $`\mathrm{\Omega },\mathrm{\Omega }^{}`$ into strata and such that $`g^{}f=g:M\backslash \mathrm{\Omega }X`$ and for any path $`\gamma `$ in $`M\backslash \mathrm{\Omega }`$ leading from a base point of $`M`$ to an arc or a circle of $`\stackrel{~}{\mathrm{\Omega }}`$ (resp. a coupon of $`\stackrel{~}{\mathrm{\Omega }}`$) we have $`u_{f\gamma }^{}=u_\gamma `$ (resp. $`v_{f\gamma }^{}=v_\gamma `$). A weak $`e`$-homeomorphism is an $`e`$-homeomorphism if the induced isomorphism $`H_1(M;)H_1(M^{};)`$ maps $`\lambda `$ onto $`\lambda ^{}`$. It is clear that a (weak) $`e`$-homeomorphism of extended $`\pi `$-manifolds induces a (weak) $`e`$-homeomorphism of their boundaries.
Taking disjoint unions of connected extended $`\pi `$-manifolds with or without boundary we obtain general extended $`\pi `$-manifolds. The notions of boundary, $`e`$-homeomorphism and weak $`e`$-homeomorphism generalize to them in the obvious way.
An example of an extended $`\pi `$-manifold is provided by the cylinder $`\mathrm{{\rm Y}}\times [0,1]`$ over an extended $`\pi `$-surface $`\mathrm{{\rm Y}}=(\mathrm{{\rm Y}},P\mathrm{{\rm Y}},g:\mathrm{{\rm Y}}\backslash PX,u,\lambda )`$. The cylinder has the following structure of an extended $`\pi `$-manifold. We provide $`\mathrm{{\rm Y}}\times [0,1]`$ with the product orientation where the interval $`[0,1]`$ is oriented from left to right. We provide $`(\mathrm{{\rm Y}}\times [0,1])=(\mathrm{{\rm Y}}\times 0)(\mathrm{{\rm Y}}\times 1)`$ with base points $`z\times 0,z\times 1`$ where $`z\mathrm{{\rm Y}}\backslash P`$ runs over the base points of the components of $`\mathrm{{\rm Y}}`$. Set $`\mathrm{\Omega }=_{pP}(p\times [0,1])`$ and provide each arc $`p\times [0,1]`$ with constant framing determined by a vector representing the given tangent direction at $`p`$. We orient the arc $`p\times [0,1]`$ towards $`p\times 1`$ if the sign of $`p`$ is $``$ and towards $`p\times 0`$ otherwise. Let pr be the projection $`(\mathrm{{\rm Y}}\times [0,1])\backslash \mathrm{\Omega }\mathrm{{\rm Y}}\backslash P`$. Then the map $`g\text{pr}`$ makes $`\mathrm{\Omega }`$ a $`\pi `$-graph and the formula $`\gamma u_{\text{pr}\gamma }`$ defines its coloring. Finally, we provide $`\mathrm{{\rm Y}}\times [0,1]`$ with Lagrangian space in real 1-homology of the boundary equal to the direct sum of copies of $`\lambda `$ in $`H_1(\mathrm{{\rm Y}}\times 0;)`$ and $`H_1(\mathrm{{\rm Y}}\times 1;)`$. This makes $`\mathrm{{\rm Y}}\times [0,1]`$ an extended $`\pi `$-manifold. Clearly, $`(\mathrm{{\rm Y}}\times [0,1])=((\mathrm{{\rm Y}})\times 0)(\mathrm{{\rm Y}}\times 1)=(\mathrm{{\rm Y}})\mathrm{{\rm Y}}`$ where $`=`$ means the equality of extended $`\pi `$-surfaces.
10.2. The HQFT $`(𝒯_𝒞,\tau _𝒞)`$. The HQFT derived from a modular crossed $`\pi `$-category $`𝒞`$ comprises the 2-dimensional homotopy modular functor $`𝒯=𝒯_𝒞`$ discussed in Section 8 and a function, $`\tau =\tau _𝒞`$, assigning to every extended 3-dimensional $`\pi `$-manifold $`M`$ a vector $`\tau (M)𝒯(M)`$. A construction of $`\tau `$ will be outlined in Section 10.4. We state here main properties of $`\tau `$:
(10.2.1) if $`f:MM^{}`$ is an $`e`$-homeomorphism of extended $`\pi `$-manifolds then $`(f|_M)_\mathrm{\#}(\tau (M))=\tau (M^{})`$ where $`(f|_M)_\mathrm{\#}:𝒯(M)𝒯(M^{})`$ is the isomorphism induced by the $`e`$-homeomorphism $`f|_M:MM^{}`$;
(10.2.2) for disjoint extended $`\pi `$-manifolds $`M,M^{}`$, we have
$$\tau (MM^{})=\tau (M)\tau (M^{})𝒯(M)_K𝒯(M^{})=𝒯(MM^{});$$
(10.2.3) for an extended $`\pi `$-manifold $`M`$ without boundary, the invariant $`\tau (M)𝒯(\mathrm{})=K`$ is the invariant introduced in Section 7.5;
(10.2.4) for any extended $`\pi `$-surface $`\mathrm{{\rm Y}}`$, the vector $`\tau (\mathrm{{\rm Y}}\times [0,1])𝒯(\mathrm{{\rm Y}})_K𝒯(\mathrm{{\rm Y}})`$ is the image of $`d_\mathrm{{\rm Y}}(𝒯(\mathrm{{\rm Y}})_K𝒯(\mathrm{{\rm Y}}))^{}`$ under the identifications
$$(𝒯(\mathrm{{\rm Y}})_K𝒯(\mathrm{{\rm Y}}))^{}=(𝒯(\mathrm{{\rm Y}}))^{}_K(𝒯(\mathrm{{\rm Y}}))^{}=𝒯(\mathrm{{\rm Y}})_K𝒯(\mathrm{{\rm Y}})$$
induced by $`d_\mathrm{{\rm Y}}`$.
A fundamental property of $`\tau `$ is a formula which computes $`\tau (M)`$ for an extended $`\pi `$-manifold $`M`$ obtained by gluing two extended $`\pi `$-manifolds $`M_1,M_2`$ along their boundaries or along several components of their boundaries. In other words, if a closed surface in $`M`$ splits $`M`$ into two pieces $`M_1,M_2`$ then the gluing formula computes $`\tau (M)`$ from $`\tau (M_1),\tau (M_2)`$. By (10.2.2), it is enough to state this formula for connected $`M`$.
We first consider the case $`M=\mathrm{}`$. Let $`M`$ be a closed connected oriented 3-manifold and $`\mathrm{\Omega }=(\mathrm{\Omega }M,zM\backslash \mathrm{\Omega },g:M\backslash \mathrm{\Omega }X,u,v)`$ be a colored $`\pi `$-graph in $`M`$. Let $`\mathrm{\Sigma }M`$ be a closed non-void surface in $`M`$ splitting $`M`$ into two compact submanifolds $`M_1,M_2`$ with $`M_1=M_2=\mathrm{\Sigma }`$. We provide each $`M_r`$ with orientation induced by the one in $`M`$. Assume (deforming if necessary $`\mathrm{\Sigma }`$ in $`M`$) that $`\mathrm{\Sigma }`$ meets $`\mathrm{\Omega }`$ transversally in a finite set of points lying on the arcs and circles of $`\mathrm{\Omega }`$. We fix a base point on each component of $`\mathrm{\Sigma }`$ so that these base points do not belong to $`\mathrm{\Sigma }\mathrm{\Omega }`$. It is clear that $`\mathrm{\Omega }_r=\mathrm{\Omega }M_r`$ is a ribbon graph in $`M_r`$ for $`r=1,2`$. Here the framing of $`\mathrm{\Omega }_r`$ is the restriction of the framing of $`\mathrm{\Omega }`$. We can thus assume that $`\stackrel{~}{\mathrm{\Omega }}_r=\stackrel{~}{\mathrm{\Omega }}M_r`$. We upgrade $`\mathrm{\Omega }_r`$ to a colored $`\pi `$-graph as follows. Deforming if necessary $`g:M\backslash \mathrm{\Omega }X`$ (keeping $`g(z)=x)`$ we can assume that $`g`$ sends all the base points of $`\mathrm{\Sigma }`$ into $`xX`$. Denote by $`g_r`$ the restriction of $`g`$ to $`M_r\backslash \mathrm{\Omega }_rM\backslash \mathrm{\Omega }`$. The pair $`(\mathrm{\Omega }_r,g_r)`$ is a $`\pi `$-graph in $`M_r`$. Transporting (as in Section 10.1) the given coloring $`(u,v)`$ of $`\mathrm{\Omega }`$ along arbitrary paths in $`M`$ connecting $`z`$ to the base points of $`\mathrm{\Sigma }`$ we obtain colorings of $`\mathrm{\Omega }`$ with respect to these points. (These colorings do not depend on the choice of the paths). Now restricting these colorings to the paths $`\gamma `$ in $`M_r\backslash \mathrm{\Omega }_r`$ connecting the base points of $`\mathrm{\Sigma }=M_r`$ to $`\stackrel{~}{\mathrm{\Omega }}_r`$ we obtain a coloring $`(u_r,v_r)`$ of $`(\mathrm{\Omega }_r,g_r)`$. Fix a Lagrangian space $`\lambda _0H_1(\mathrm{\Sigma };)`$. Then the tuple $`(M_r,\mathrm{\Omega }_r,g_r,u_r,v_r,\lambda _0)`$ is an extended $`\pi `$-manifold briefly denoted $`M_r`$. Its boundary $`M_r`$ is an extended $`\pi `$-surface with underlying surface $`\mathrm{\Sigma }`$. It follows from definitions that $`M_2=M_1`$ where minus denotes the negation of extended $`\pi `$-surfaces. We have the vectors $`\tau (M_1)𝒯(M_1)`$ and $`\tau (M_2)𝒯(M_2)=𝒯(M_1)`$. Then
$$\tau (M)=(𝒟\mathrm{\Delta }_{}^1)^nd_{M_1}(\tau (M_1)\tau (M_2))K$$
$`(10.2.a)`$
where $`M`$ denotes the extended $`\pi `$-manifold $`(M,\mathrm{\Omega },z,g,u,v)`$, $`d_{M_1}`$ is the pairing $`𝒯(M_1)_K𝒯(M_1)K`$ provided by (8.3.5) and $`n`$ is an integer computed as follows. Consider the inclusion homomorphism $`H_1(M_r;)H_1(M_r;)`$ and observe that its kernel, $`\lambda _r`$, is a Lagrangian subspace in $`H_1(M_1;)=H_1(M_2;)=H_1(\mathrm{\Sigma };)`$. Then $`n=\mu (\lambda _1,\lambda _0,\lambda _2)`$ is the Maslov index of the Lagrangian spaces $`\lambda _1,\lambda _0,\lambda _2`$ in $`H_1(M_1;)`$.
Let now $`M=(M,\mathrm{\Omega },g:M\backslash \mathrm{\Omega }X,u,v,\lambda )`$ be a connected extended $`\pi `$-manifold with boundary. Let $`\mathrm{\Sigma }M\backslash M`$ be a closed non-void surface splitting $`M`$ into two compact submanifolds $`M_1,M_2`$. For $`r=1,2`$, we have $`M_r=\mathrm{\Sigma }(MM_r)`$ where $`MM_r=MM_r`$ is a disjoint union (possibly void) of certain components of $`M`$ such that $`M=(MM_1)(MM_2)`$. Assume that $`\lambda =\lambda ^1\lambda ^2`$ where $`\lambda ^r`$ is a Lagrangian space in $`H_1(MM_r;)`$. Then the extended $`\pi `$-surface $`M`$ splits as the disjoint union of extended $`\pi `$-surfaces $`\mathrm{{\rm Y}}^1=(MM_1,\lambda ^1)`$ and $`\mathrm{{\rm Y}}^2=(MM_2,\lambda ^2)`$.
Assume that $`\mathrm{\Sigma }`$ meets $`\mathrm{\Omega }`$ transversally in a finite set of points lying on the arcs and circles of $`\mathrm{\Omega }`$. We fix a base point on each component of $`\mathrm{\Sigma }`$ so that these base points do not belong to $`\mathrm{\Sigma }\mathrm{\Omega }`$. Deforming if necessary $`g:M\backslash \mathrm{\Omega }X`$ (keeping $`g`$ on $`M`$) we can assume that $`g`$ sends all the base points of $`\mathrm{\Sigma }`$ into $`xX`$. As above, $`\mathrm{\Omega }_r=\mathrm{\Omega }M_r`$ is a ribbon graph in $`M_r`$ for $`r=1,2`$ where the orientation in $`M_r`$ is induced by the one in $`M`$ and the framing on $`\mathrm{\Omega }_r`$ is the restriction of the one on $`\mathrm{\Omega }`$. Denote by $`g_r`$ the restriction of $`g`$ to $`M_r\backslash \mathrm{\Omega }_rM\backslash \mathrm{\Omega }`$. The pair $`(\mathrm{\Omega }_r,g_r)`$ is a $`\pi `$-graph in $`M_r`$. Choose a component of $`M`$ and transport the given coloring $`(u,v)`$ of $`\mathrm{\Omega }`$ at its base point along arbitrary paths in $`M`$ to the base points of $`\mathrm{\Sigma }`$. This gives colorings of $`\mathrm{\Omega }`$ with respect to the base points of $`\mathrm{\Sigma }`$. These colorings depends neither on the choice of the initial component of $`M`$ nor on the choice of the paths in $`M`$. Now restricting the colorings of $`\mathrm{\Omega }`$ (with respect to the base points of $`\mathrm{\Sigma }`$ and $`MM_r`$) to the paths $`\gamma `$ in $`M_r\backslash \mathrm{\Omega }_r`$ we obtain a coloring $`(u_r,v_r)`$ of $`(\mathrm{\Omega }_r,g_r)`$. Fix a Lagrangian space $`\lambda _0H_1(\mathrm{\Sigma };)`$. Then the tuple $`(M_r,\mathrm{\Omega }_r,g_r,u_r,v_r,\lambda _0\lambda ^r)`$ is an extended $`\pi `$-manifold briefly denoted $`M_r`$. Its boundary $`M_r`$ is a disjoint union of $`\mathrm{{\rm Y}}^r`$ and an extended $`\pi `$-surface $`\mathrm{{\rm Y}}_r`$ with underlying surface $`\mathrm{\Sigma }`$. Clearly, $`\mathrm{{\rm Y}}_2=\mathrm{{\rm Y}}_1`$. For $`r=1,2`$, we have a vector
$$\tau (M_r)𝒯(M_r)=𝒯(\mathrm{{\rm Y}}^r)_K𝒯(\mathrm{{\rm Y}}_r).$$
Then
$$\tau (M)=(𝒟\mathrm{\Delta }_{}^1)^nd_{\mathrm{{\rm Y}}_1}(\tau (M_1)\tau (M_2))𝒯(\mathrm{{\rm Y}}^1)_K𝒯(\mathrm{{\rm Y}}^2)$$
$`(10.2.b)`$
where $`d_{\mathrm{{\rm Y}}_1}`$ is the pairing $`𝒯(\mathrm{{\rm Y}}_1)_K𝒯(\mathrm{{\rm Y}}_2)K`$ provided by (8.3.5) and $`n`$ is an integer computed as follows. Let $`\lambda _r`$ be the subset of $`H_1(\mathrm{{\rm Y}}_r;)=H_1(\mathrm{\Sigma };)`$ consisting of elements homological in $`M_r`$ to elements of $`\lambda ^rH_1(MM_r;)`$. One can check that $`\lambda _r`$ is a Lagrangian space. Then $`n=\mu (\lambda _1,\lambda _0,\lambda _2)`$ is the Maslov index of $`\lambda _1,\lambda _0,\lambda _2`$ in $`H_1(\mathrm{{\rm Y}}^1;)`$.
10.3. Extended $`\pi `$-cobordisms. As in the study of TQFT’s, we can reformulate the vector $`\tau _𝒞`$ as an operator invariant of $`\pi `$-cobordisms. An extended (3-dimensional) $`\pi `$-cobordism is a triple $`(M,_{}M,_+M)`$ where $`M`$ is a 3-dimensional extended $`\pi `$-manifold and $`_{}M,_+M`$ are extended $`\pi `$-surfaces such that $`M=(_{}M)_+M`$. We call $`_{}M,_+M`$ the bottom base and top base of the cobordism, respectively. Clearly,
$$𝒯(M)=𝒯(_{}M)_K𝒯(_+M)=(𝒯(_{}M))^{}_K𝒯(_+M)$$
$$=\text{Hom}_K(𝒯(_{}M),𝒯(_+M)).$$
By these identifications, the vector $`\tau (M)𝒯(M)`$ determines a homomorphism $`𝒯(_{}M)𝒯(_+M)`$ denoted $`\tau (M,_{}M,_+M)`$. Note that this definition implicitly involves the pairing $`d__{}M`$. By axioms (10.2.1) and (10.2.2), the homomorphism $`\tau (M,_{}M,_+M)`$ is natural with respect to $`e`$-homeomorphisms of cobordisms and $``$-multiplicative with respect to disjoint unions.
For example, if $`_{}M=\mathrm{}`$ then $`_+M=M`$ and $`\tau (M,_{}M,_+M):𝒯(\mathrm{})=K𝒯(M)`$ sends any $`kK`$ into $`k\tau (M)`$. If $`_+M=\mathrm{}`$ then $`_{}M=M`$ and $`\tau (M,_{}M,_+M):𝒯(M)𝒯(\mathrm{})=K`$ sends any $`h𝒯(M)`$ into $`d_M(\tau (M),h)`$. By (10.2.3), if $`M=\mathrm{}`$ then $`\tau (M,_{}M,_+M):KK`$ is multiplication by the invariant $`\tau (M)K`$ introduced in Section 7.5.
We can reformulate formulas (10.2.a) and (10.2.b) as
$$\tau (M,_{}M,_+M)=(𝒟\mathrm{\Delta }_{}^1)^n\tau (M_2,_{}(M_2),_+(M_2))\tau (M_1,_{}(M_1),_+(M_1))$$
where we view $`M,M_1,M_2`$ as extended $`\pi `$-cobordisms with bottom bases $`(MM_1),(MM_1),\mathrm{{\rm Y}}_2=\mathrm{{\rm Y}}_1`$ and top bases $`MM_2,\mathrm{{\rm Y}}_1,MM_2`$, respectively.
For an extended $`\pi `$-surface $`\mathrm{{\rm Y}}=(\mathrm{{\rm Y}},P\mathrm{{\rm Y}},g:\mathrm{{\rm Y}}\backslash PX,u,\lambda )`$, the triple $`(\mathrm{{\rm Y}}\times [0,1],\mathrm{{\rm Y}}\times 0,\mathrm{{\rm Y}}\times 1)`$ is an extended $`\pi `$-cobordism whose bases are copies of $`\mathrm{{\rm Y}}`$. By (10.2.4),
$$\tau (\mathrm{{\rm Y}}\times [0,1],\mathrm{{\rm Y}}\times 0,\mathrm{{\rm Y}}\times 1)=\text{id}_{𝒯(\mathrm{{\rm Y}})}:T(\mathrm{{\rm Y}})𝒯(\mathrm{{\rm Y}}).$$
For every $`\alpha \pi `$, we define a twisted cylinder $`(\mathrm{{\rm Y}}\times [0,1])^\alpha `$ as follows. It differs from $`\mathrm{{\rm Y}}\times [0,1]`$ only by the choice of the map $`(\mathrm{{\rm Y}}\times [0,1])\backslash \mathrm{\Omega }X`$ and the choice of the coloring of the ribbon graph $`\mathrm{\Omega }=P\times [0,1]`$. The map in question is chosen so that its restriction to the bottom base equals $`g`$ and its restrictions to arcs $`z\times [0,1]`$ are loops representing $`\alpha ^1`$ for all the base points $`z\mathrm{{\rm Y}}\backslash P`$ of the components of $`\mathrm{{\rm Y}}`$. The coloring of $`\mathrm{\Omega }`$ is chosen so that its restriction to the bottom base coincides with $`u`$. It is clear that $`(\mathrm{{\rm Y}}\times [0,1])^\alpha `$ is an extended $`\pi `$-cobordism with bottom base $`\mathrm{{\rm Y}}`$ and top base $`{}_{}{}^{\alpha }\mathrm{{\rm Y}}`$. The operator invariant of this cobordism defines the action of $`\alpha `$ discussed in Section 8.5:
$$\alpha _{}=\tau ((\mathrm{{\rm Y}}\times [0,1])^\alpha ,\mathrm{{\rm Y}},{}_{}{}^{\alpha }\mathrm{{\rm Y}}):𝒯(\mathrm{{\rm Y}})𝒯({}_{}{}^{\alpha }\mathrm{{\rm Y}}).$$
10.4. Construction of $`(𝒯_𝒞,\tau _𝒞)`$. The construction closely follows the known construction of 3-dimensional TQFT’s from modular categories, see \[Tu2\]. First, one defines the operator invariant for extended $`\pi `$-cobordisms whose boundary components are parametrized, i.e., identified with standard $`\pi `$-surfaces in $`S^3`$. (Only the geometric position of the surfaces in $`S^3`$ is standard, the map to $`X=K(\pi ,1)`$ is arbitrary). Then one uses these operators to define the action of weak $`e`$-homeomorphisms. Finally, one replaces the parametrizations with Lagrangian spaces in homology. We skip the details and give here only one of the key lemmas whose proof is somewhat different from the standard case.
10.5. Lemma. Let $`T^2\times [0,1]`$ be a tangle consisting of a vertical interval $`t`$ oriented downwards and its meridian $`m`$, both with zero framing. Sending all meridians of $`t`$ into $`1\pi `$ and all meridians of $`m`$ into $`\alpha \pi `$, we make $`T`$ a $`\pi `$-tangle, $`T_\alpha `$. Let $`V𝒞_1`$ be a simple object in the neutral component of $`𝒞`$. Let $`T_\alpha (V)`$ be $`T_\alpha `$ colored so that $`m`$ acquires the canonical color as in Section 7.2 and the target of $`T_\alpha `$ is the triple $`(+1,1\pi ,V)`$. Then the source of $`T_\alpha (V)`$ is the triple $`(+1,1\pi ,\phi _{\alpha ^1}(V))`$ and $`F(T_\alpha (V))\text{Hom}_𝒞(\phi _{\alpha ^1}(V),V)`$ is computed by
$$F(T_\alpha (V))=\{\begin{array}{cc}𝒟^2\text{id}_{\text{1}\text{ I}},if\text{V=}\text{1}\text{ I},\hfill & \\ 0,ifV\text{is not isomorphic to}\text{1}\text{ I}.\hfill & \end{array}$$
Proof. We begin with another useful identity. Fix $`\beta \pi `$ and an object $`W𝒞_\beta `$. Consider the $`\pi `$-tangle $`T_\beta `$ as in the statement of the lemma with $`\alpha `$ replaced by $`\beta `$. We present $`T_\beta `$ by a plane diagram with two crossings. We attach $`W`$ to the arc of the diagram representing $`m`$ and attach $`V`$ (resp. $`\phi _{\beta ^1}(V)`$) to the arc incident to the output (resp. intput). Denote the resulting colored $`\pi `$-tangle by $`T_W^V`$. Its target and source are the triples $`(+1,1\pi ,V)`$ and $`(+1,1\pi ,\phi _{\beta ^1}(V))`$, respectively. We claim that
$$F(T_W^V)F(T_\alpha (\phi _{\beta ^1}(V)))=\text{dim}(W)F(T_{\beta \alpha }(V)):\phi _{\alpha ^1\beta ^1}(V)V.$$
$`(10.5.a)`$
We first prove this equality and then deduce from it the claim of the lemma. Glueing $`T_W^V`$ on the top of $`T_\alpha (\phi _{\beta ^1}(V))`$ we obtain a colored $`\pi `$-tangle
$$\stackrel{~}{T}=T_W^VT_\alpha (\phi _{\beta ^1}(V)).$$
Geometrically, $`\stackrel{~}{T}`$ consists of a vertical interval with two meridians and all framings zero. Clearly,
$$F(T_W^V)F(T_\alpha (\phi _{\beta ^1}(V)))=F(\stackrel{~}{T}).$$
Denote by $`\mathrm{}_W^\beta `$ a colored $`\pi `$-knot (or rather unknot) in $`^2\times [0,1]`$ represented by a plane circle labeled with $`(\beta ,W)`$. Sliding the $`W`$-colored circle of $`\stackrel{~}{T}`$ along the $`can`$-colored circle we can transform $`\stackrel{~}{T}`$ into a disjoint union $`\mathrm{}_W^\beta T_{\beta \alpha }(V)`$. (An explicit splitting of this handle sliding into a composition of Kirby-Fenn-Rourke moves is given in \[Tu2, p. 93\]). As in the proof of Theorem 7.3, we obtain
$$F(\stackrel{~}{T})=F(\mathrm{}_W^\beta T_{\beta \alpha }(V))=F(\mathrm{}_W^\beta )F(T_{\beta \alpha }(V))=\text{dim}(W)F(T_{\beta \alpha }(V)).$$
This implies (10.5.a).
Now we can prove the claim of the lemma. If $`V=\text{1}\text{ I}`$ then formula (2.2.f) implies that we can push the circle stratum of $`T`$ across the arc stratum without changing the operator invariant. This yields our claim in this case. Assume that $`V`$ is not isomorphic to 1 I. Let $`I`$ be the set of isomorphism classes of simple objects in $`𝒞_1`$ and let $`\{V_i𝒞_1\}_{iI}`$ be representatives of these classes. We can assume that $`V=V_i`$ for a certain $`iI`$. For $`W=V_j`$ with $`jI`$ we can compute $`F(T_W^V)\text{End}(V)`$ explicitly. Since $`V`$ is simple, $`F(T_W^V)=k\text{id}_V`$ with $`kK`$. The closure of $`T_W^V`$ is the Hopf link whose components are colored with $`V=V_i,W=V_j`$. Therefore $`k\text{dim}(i)=S_{i,j}`$ where $`\text{dim}(i)=\text{dim}(V)=\text{dim}(V_i)`$. Hence $`k=(\text{dim}(i))^1S_{i,j}`$ where we use the invertibility of $`\text{dim}(i)`$ (Lemma 6.5). Substituting this in (10.5.a) (for $`\beta =1`$) we obtain
$$(\text{dim}(i))^1S_{i,j}F(T_\alpha (V))=\text{dim}(j)F(T_\alpha (V)).$$
By \[Tu2, formula (3.8.b)\], $`_{jI}\text{dim}(j)S_{i,j}=0`$. Hence
$$F(T_\alpha (V))=𝒟^2\underset{jI}{}\text{dim}(j)\text{dim}(j)F(T_\alpha (V))$$
$$=𝒟^2(\text{dim}(i))^1\underset{jI}{}\text{dim}(j)S_{i,j}F(T_\alpha (V))=0.$$
11. Hopf group-coalgebras
11.1. Hopf algebras. For convenience of the reader we recall the standard definitions of quasitriangular and ribbon Hopf algebras, see for instance \[KRT\], \[Tu2\]. A Hopf algebra over $`K`$ is a tuple $`(A,\mathrm{\Delta },\epsilon ,s)`$ where $`A`$ is an associative unital algebra over $`K`$, $`\mathrm{\Delta }:AA^2=AA`$ and $`\epsilon :AK`$ are algebra homomorphisms, $`s:AA`$ is an algebra anti-homomorphism such that $`\mathrm{\Delta }`$ is coassociative and
$$(\text{id}_A\epsilon )\mathrm{\Delta }=(\epsilon \text{id}_A)\mathrm{\Delta }=\text{id}_A:AA,$$
$$\mu (s\text{id}_A)\mathrm{\Delta }=\mu (\text{id}_As)\mathrm{\Delta }=1_A\epsilon :AA$$
where $`\mu `$ is multiplication in $`A`$.
Let $`A`$ be a Hopf algebra over $`K`$. Denote the flip (permutation) $`A^2A^2`$ by Perm. A pair $`(A,RA^2)`$ is called a quasitriangular Hopf algebra if $`R`$ is invertible in $`A^2`$ and satisfies the following identities:
$$\text{Perm}(\mathrm{\Delta }(a))=R\mathrm{\Delta }(a)R^1,$$
for all $`aA`$ and
$$(\text{id}_A\mathrm{\Delta })(R)=R_{13}R_{12},(\mathrm{\Delta }\text{id}_A)(R)=R_{13}R_{23}$$
where $`R_{12}=R1_AA^3`$, $`R_{23}=1_ARA^3`$ and $`R_{13}=(\text{id}_A\text{Perm})(R_{12})A^3`$. These identities imply the Yang-Baxter equality $`R_{12}R_{13}R_{23}=R_{23}R_{13}R_{12}`$.
Let $`(A,R)`$ be a quasitriangular Hopf algebra. A triple $`(A,R,vA)`$ is called a ribbon Hopf algebra if $`v`$ is an invertible element of the center of $`A`$ such that $`s(v)=v`$ and $`\mathrm{\Delta }(v)=(vv)\text{Perm}(R)R`$. If $`A=(A,R,v)`$ is a ribbon Hopf algebra then the tuple $`(A,(\text{Perm}(R))^1,v^1)`$ is also a ribbon Hopf algebra.
It is well-known that the category of representations of a quasitriangular (resp. ribbon) Hopf algebra is a braided (resp. ribbon) monoidal category.
11.2. Hopf $`\pi `$-coalgebras. Let $`\pi `$ be a group. The notion of a $`\pi `$-coalgebra is dual to the notion of a (unital) $`\pi `$-graded algebra. By a $`\pi `$-coalgebra over $`K`$, we mean the following data:
\- a family of $`K`$-modules $`\{A_\alpha \}_{\alpha \pi }`$;
\- a family of $`K`$-linear homomorphisms (called the comultiplication)
$$\mathrm{\Delta }=\{\mathrm{\Delta }_{\alpha ,\beta }:A_{\alpha \beta }A_\alpha A_\beta \}_{\alpha ,\beta \pi };$$
\- a $`K`$-homomorphism $`\epsilon _1:A_1K`$ (the counit).
This data should satisfy the following two axioms:
(11.2.1) $`\mathrm{\Delta }`$ is coassociative in the sense that for any $`\alpha ,\beta ,\gamma \pi `$,
$$(\mathrm{\Delta }_{\alpha ,\beta }\text{id}_{A_\gamma })\mathrm{\Delta }_{\alpha \beta ,\gamma }=(\text{id}_{A_\alpha }\mathrm{\Delta }_{\beta ,\gamma })\mathrm{\Delta }_{\alpha ,\beta \gamma }:A_{\alpha \beta \gamma }A_\alpha A_\beta A_\gamma ;$$
(11.2.2) for any $`\alpha \pi `$,
$$(\text{id}_{A_\alpha }\epsilon _1)\mathrm{\Delta }_{\alpha ,1}=(\epsilon _1\text{id}_{A_\alpha })\mathrm{\Delta }_{1,\alpha }=\text{id}_{A_\alpha }:A_\alpha A_\alpha $$
A Hopf $`\pi `$-coalgebra over $`K`$ is a $`\pi `$-coalgebra $`(A,\mathrm{\Delta },\epsilon )`$ where each $`A_\alpha `$ is an associative $`K`$-algebra with multiplication $`\mu _\alpha `$ and (left and right) unit $`1_\alpha `$ endowed with algebra anti-isomorphisms $`s=\{s_\alpha :A_\alpha A_{\alpha ^1}\}_{\alpha \pi }`$ (the antipode) such that
(11.2.3) for all $`\alpha ,\beta \pi `$ the comultiplication $`\mathrm{\Delta }_{\alpha ,\beta }`$ is an algebra homomorphism and $`\mathrm{\Delta }_{\alpha ,\beta }(1_{\alpha \beta })=1_\alpha 1_\beta `$;
(11.2.4) the counit $`\epsilon _1:A_1K`$ is an algebra homomorphism and $`\epsilon _1(1_1)=1_K`$;
(11.2.5) for any $`\alpha \pi `$,
$$\mu _\alpha (s_{\alpha ^1}\text{id}_{A_\alpha })\mathrm{\Delta }_{\alpha ^1,\alpha }=\mu _\alpha (\text{id}_{A_\alpha }s_{\alpha ^1})\mathrm{\Delta }_{\alpha ,\alpha ^1}=1_\alpha \epsilon _1:A_1A_\alpha .$$
It follows from the properties of algebra anti-isomorphisms that $`s_\alpha (1_\alpha )=1_{\alpha ^1}`$ for all $`\alpha \pi `$. Note also that the tuple $`(A_1,\mathrm{\Delta }_{1,1},\epsilon _1,s_1)`$ is a Hopf algebra in the usual sense of the word. We call it the neutral component of $`A`$.
Warning: the notion of a Hopf $`\pi `$-coalgebra is not self-dual. It is quite interesting to study the dual notion of a Hopf $`\pi `$-algebra but we shall not need it in this paper.
A crossed Hopf $`\pi `$-coalgebra over $`K`$ is a Hopf $`\pi `$-coalgebra $`(\{A_\alpha \}_{\alpha \pi },\mathrm{\Delta },\epsilon _1,s)`$ endowed with a family of algebra isomorphisms $`\phi =\{\phi _\alpha :A_\beta A_{\alpha \beta \alpha ^1}\}_{\alpha ,\beta \pi }`$ such that
(11.2.6) each $`\phi _\alpha `$ preserves the counit, the antipode, and the comultiplication, i.e., for any $`\alpha ,\beta ,\gamma \pi `$, we have
$$\epsilon _1\phi _\alpha |_{A_1}=\epsilon _1,$$
$$\phi _\alpha s_\beta =s_{\alpha \beta \alpha ^1}\phi _\alpha :A_\beta A_{\alpha \beta ^1\alpha ^1},$$
$$(\phi _\alpha \phi _\alpha )\mathrm{\Delta }_{\beta ,\gamma }=\mathrm{\Delta }_{\alpha \beta \alpha ^1,\alpha \gamma \alpha ^1}\phi _\alpha :A_{\beta \gamma }A_{\alpha \beta \alpha ^1}A_{\alpha \gamma \alpha ^1};$$
(11.2.7) $`\phi `$ is an action of $`\pi `$, i.e., $`\phi _{\alpha \alpha ^{}}=\phi _\alpha \phi _\alpha ^{}`$ for all $`\alpha ,\alpha ^{}\pi `$.
It is clear that $`\phi _\alpha (1_\beta )=1_{\alpha \beta \alpha ^1}`$ for all $`\alpha ,\beta \pi `$ and $`\phi _\alpha (A_1)=A_1`$. Restricting the action $`\phi `$ of $`\pi `$ to $`A_1`$ we obtain an action of $`\pi `$ on $`A_1`$ by Hopf algebra endomorphisms.
We end this subsection with two examples of crossed Hopf $`\pi `$-coalgebras. Both examples are derived from an action of $`\pi `$ on a Hopf algebra $`(A,\mathrm{\Delta },\epsilon ,s)`$ over $`K`$ by Hopf algebra endomorphisms. Set $`A^\pi =\{A_\alpha \}_{\alpha \pi }`$ where for each $`\alpha \pi `$, the algebra $`A_\alpha `$ is a copy of $`A`$. Fix an identification isomorphism of algebras $`i_\alpha :AA_\alpha `$. For $`\alpha ,\beta \pi `$, we define a comultiplication $`\mathrm{\Delta }_{\alpha ,\beta }:A_{\alpha \beta }A_\alpha A_\beta `$ by
$$\mathrm{\Delta }_{\alpha ,\beta }(i_{\alpha \beta }(a))=\underset{(a)}{}i_\alpha (a^{})i_\beta (a^{\prime \prime })$$
where $`aA`$ and $`\mathrm{\Delta }(a)=_{(a)}a^{}a^{\prime \prime }`$ is the given comultiplication in $`A`$ written in Sweedler’s sigma notation. The counit $`\epsilon _1:A_1K`$ is defined by $`\epsilon _1(i_1(a))=\epsilon (a)K`$ for $`aA`$. For $`\alpha \pi `$, the antipode $`s_\alpha :A_\alpha A_{\alpha ^1}`$ is given by
$$s_\alpha (i_\alpha (a))=i_{\alpha ^1}(s(a))$$
where $`aA`$. For $`\alpha ,\beta \pi `$, the homomorphism $`\phi _\alpha :A_\beta A_{\alpha \beta \alpha ^1}`$ is defined by $`\phi _\alpha (i_\beta (a))=i_{\alpha \beta \alpha ^1}(\alpha (a))`$. All the axioms of a crossed Hopf $`\pi `$-coalgebra for $`A^\pi `$ follow directly from definitions.
The second example differs only by the definition of the comultiplication and the antipode. Let $`\overline{A}^\pi `$ be the same family of algebras $`\{A_\alpha =A\}_{\alpha \pi }`$ with the same counit, the same action $`\phi `$ of $`\pi `$, the comultiplication $`\overline{\mathrm{\Delta }}_{\alpha ,\beta }:A_{\alpha \beta }A_\alpha A_\beta `$ and the antipode $`\overline{s}_\alpha :A_\alpha A_{\alpha ^1}`$ defined by
$$\overline{\mathrm{\Delta }}_{\alpha ,\beta }(i_{\alpha \beta }(a))=\underset{(a)}{}i_\alpha (\beta (a^{}))i_\beta (a^{\prime \prime }),$$
$$\overline{s}_\alpha (i_\alpha (a))=i_{\alpha ^1}(\alpha (s(a)))=i_{\alpha ^1}(s(\alpha (a)))$$
where $`aA`$. The axioms of a crossed Hopf $`\pi `$-coalgebra for $`\overline{A}^\pi `$ follow from definitions. Both $`A^\pi `$ and $`\overline{A}^\pi `$ are extensions of $`A`$ since $`A_1^\pi =\overline{A}_1^\pi =A_1`$ as Hopf algebras.
In particular, if $`G`$ is a Lie group with Lie algebra $`𝕘`$ then the universal enveloping algebra $`U(𝕘)`$ has a canonical structure of a Hopf algebra and $`G`$ acts on $`U(𝕘)`$ by Hopf algebra endomorphisms induced by the group conjugation. The constructions above give crossed Hopf $`G`$-algebras $`(U(𝕘))^G=\{U(𝕘)_\alpha \}_{\alpha G}`$ and $`\overline{(U(𝕘))}^G=\{U(𝕘)_\alpha \}_{\alpha G}`$ where each $`U(𝕘)_\alpha `$ is a copy of $`U(𝕘)`$ sitting at $`\alpha G`$.
11.3. Quasitriangular Hopf $`\pi `$-coalgebras. Let $`A=(\{A_\alpha \},\mathrm{\Delta },\epsilon _1,s,\phi )`$ be a crossed Hopf $`\pi `$-coalgebra. A universal $`R`$-matrix in $`A`$ is a family of invertible elements
$$R=\{R_{\alpha ,\beta }A_\alpha A_\beta \}_{\alpha ,\beta \pi }$$
$`(11.3.a)`$
satisfying the following conditions:
(11.3.1) for any $`\alpha ,\beta \pi ,aA_{\alpha \beta }`$,
$$R_{\alpha ,\beta }\mathrm{\Delta }_{\alpha ,\beta }(a)=\text{Perm}_{\beta ,\alpha }((\phi _{\alpha ^1}\text{id}_{A_\alpha })\mathrm{\Delta }_{\alpha \beta \alpha ^1,\alpha }(a))R_{\alpha ,\beta }$$
where $`\text{Perm}_{\beta ,\alpha }`$ is the flip $`A_\beta A_\alpha A_\alpha A_\beta `$;
(11.3.2) for any $`\alpha ,\beta ,\gamma \pi `$,
$$(\text{id}_{A_\alpha }\mathrm{\Delta }_{\beta ,\gamma })(R_{\alpha ,\beta \gamma })=(R_{\alpha ,\gamma })_{1\beta 3}(R_{\alpha ,\beta })_{12\gamma }$$
and
$$(\mathrm{\Delta }_{\alpha ,\beta }\text{id}_{A_\gamma })(R_{\alpha \beta ,\gamma })=((\phi _\beta \text{id}_{A_\gamma })(R_{\beta ^1\alpha \beta ,\gamma }))_{1\beta 3}(R_{\beta ,\gamma })_{\alpha 23}$$
where for $`K`$-modules $`P,Q`$ and $`r=_jp_jq_jPQ`$ we set
$$r_{12\gamma }=r1_\gamma PQA_\gamma ,r_{\alpha 23}=1_\alpha rA_\alpha PQ$$
and
$$r_{1\beta 3}=\underset{j}{}p_j1_\beta q_jPA_\beta Q;$$
(11.3.3) the family (11.3.a) is invariant under the endomorphisms $`\phi _\alpha `$, i.e.,
$$(\phi _\alpha \phi _\alpha )(R_{\beta ,\gamma })=R_{\alpha \beta \alpha ^1,\alpha \gamma \alpha ^1}.$$
A crossed Hopf $`\pi `$-coalgebra endowed with a universal $`R`$-matrix is said to be quasitriangular. It is easy to deduce from (11.3.1) - (11.3.3) the Yang-Baxter equality for $`R`$:
$$(R_{\alpha ,\beta })_{12\gamma }((\phi _\beta \text{id}_{A_\gamma })(R_{\beta ^1\alpha \beta ,\gamma }))_{1\beta 3}(R_{\beta ,\gamma })_{\alpha 23}=(R_{\beta ,\gamma })_{\alpha 23}(R_{\alpha ,\gamma })_{1\beta 3}(R_{\alpha ,\beta })_{12\gamma }.$$
11.4. Ribbon Hopf $`\pi `$-coalgebras. Let $`A`$ be a quasitriangular (crossed) Hopf $`\pi `$-coalgebra with universal $`R`$-matrix (11.3.a). A twist in $`A`$ is a collection of invertible elements $`\{\theta _\alpha A_\alpha \}_{\alpha \pi }`$ such that
(11.4.1) $`\phi _\alpha (a)=\theta _\alpha ^1a\theta _\alpha `$ for all $`\alpha \pi ,aA_\alpha `$;
(11.4.2) $`s_\alpha (\theta _\alpha )=\theta _{\alpha ^1}`$ for all $`\alpha \pi `$;
(11.4.3) for all $`\alpha ,\beta \pi `$,
$$\mathrm{\Delta }_{\alpha ,\beta }(\theta _{\alpha \beta })=(\theta _\alpha \theta _\beta )\text{Perm}_{\beta ,\alpha }((\text{id}_{A_\beta }\phi _\alpha )R_{\beta ,\alpha })R_{\alpha ,\beta };$$
(11.4.4) $`\phi _\alpha (\theta _\beta )=\theta _{\alpha \beta \alpha ^1}`$ for all $`\alpha ,\beta \pi `$.
A quasitriangular crossed Hopf $`\pi `$-coalgebra endowed with a twist is said to be ribbon.
It follows from definitions that the neutral component $`A_1`$ of $`A`$ endowed with $`R_{1,1}A_1A_1,\theta _1A_1`$ is a ribbon Hopf algebra in the sense of Section 11.1. In particular, the equality $`\phi _1=\text{id}`$ implies that $`\theta _1`$ lies in the center of $`A_1`$.
For $`\pi =1`$, the notions introduced in Sections 11.3 and 11.4 boil down to the standard notions of quasitriangular and ribbon Hopf algebras.
11.5. Examples. We give here two examples of ribbon Hopf $`\pi `$-coalgebras. Let $`(A,\mathrm{\Delta },\epsilon ,s,R,v)`$ be a ribbon Hopf algebra over $`K`$. An element $`\alpha A`$ is group-like if $`\mathrm{\Delta }(\alpha )=\alpha \alpha `$ and $`\epsilon (\alpha )=1K`$. Any group-like element $`\alpha `$ is invertible and $`s(\alpha )=\alpha ^1`$. The group-like elements of $`A`$ form a group, $`\pi =\pi (A)`$, under multiplication in $`A`$. For $`\alpha \pi `$, the formula $`a\alpha a\alpha ^1`$ with $`aA`$ defines a Hopf algebra endomorphism of $`A`$. This gives an action of $`\pi `$ on $`A`$ by Hopf algebra endomorphisms. Applying the constructions of Section 11.2 to this action we obtain crossed Hopf $`\pi `$-coalgebras $`A^\pi `$ and $`\overline{A}^\pi `$. We define a universal $`R`$-matrix and twist in $`A^\pi `$ by
$$R_{\alpha ,\beta }=(i_\alpha i_\beta )((1_A\alpha ^1)R)A_\alpha A_\beta \text{and}\theta _\alpha =i_\alpha (v\alpha ^1)A_\alpha $$
where $`\alpha ,\beta \pi `$. We define a universal $`R`$-matrix and twist in $`\overline{A}^\pi `$ by
$$\overline{R}_{\alpha ,\beta }=(i_\alpha i_\beta )(R(\beta ^11_A))A_\alpha A_\beta \text{and}\overline{\theta }_\alpha =i_\alpha (v\alpha ^1)A_\alpha $$
where $`\alpha ,\beta \pi `$. A direct computation shows that $`(A^\pi ,\{R_{\alpha ,\beta }\}_{\alpha ,\beta \pi },\{\theta _\alpha \}_{\alpha \pi })`$ and $`(\overline{A}^\pi ,\{\overline{R}_{\alpha ,\beta }\}_{\alpha ,\beta \pi },\{\overline{\theta }_\alpha \}_{\alpha \pi })`$ are ribbon Hopf $`\pi `$-coalgebras.
Group-like elements of a quantum universal enveloping algebra $`A=U_q(𝕘)`$ are well-known. For instance, if $`𝕘=sl(N+1)`$ and $`q`$ is generic then $`\pi (A)=^N`$ is a free abelian group of rank $`N`$ generated by the canonical group-like elements $`K_1,\mathrm{},K_NA`$. If $`q`$ is a primitive root of unity of order $`\mathrm{}`$, then one usually considers a version $`A^{\text{res}}`$ of $`A=U_q(sl(N+1))`$ with $`K_i^{\mathrm{}}=1`$ for all $`i=1,\mathrm{},N`$ (see \[KRT\]). Then $`\pi (A^{\text{res}})=(/\mathrm{})^N`$.
11.6. Operations on crossed Hopf group-coalgebras. In analogy with the pull-back of group-categories, we can pull back a Hopf $`\pi `$-coalgebra $`A`$ along a group homomorphism $`q:\pi ^{}\pi `$. This gives a Hopf $`\pi ^{}`$-algebra $`A^{}=q^{}(A)`$ defined by $`A_\alpha ^{}=A_{q(\alpha )}`$ for any $`\alpha \pi ^{}`$. If $`A`$ is crossed (resp. quasitriangular, ribbon), then $`A^{}`$ has the structure of a crossed (resp. quasitriangular, ribbon) Hopf $`\pi ^{}`$-algebra obtained by lifting the data from $`A`$ to $`A^{}`$ in the obvious way. Taking $`\pi =1`$ and choosing as $`A`$ any quantum group, we obtain an example of a ribbon Hopf $`\pi ^{}`$-algebra for any $`\pi ^{}`$.
The constructions of direct sum, tensor product, and transfer discussed in Sections 1.4, 2.5, 13 have their analogues for crossed, quasitriangular and ribbon Hopf group-coalgebras. We leave the details to the reader. (A transfer for $`\pi `$-algebras is discussed in a related setting in \[Tu3\]).
Given a crossed Hopf $`\pi `$-coalgebra $`A=(\{A_\alpha \}_{\alpha \pi },\mathrm{\Delta },\epsilon _1,s,\phi )`$ we define its mirror $`\overline{A}=(\{\overline{A}_\alpha \}_{\alpha \pi },\overline{\mathrm{\Delta }},\overline{\epsilon }_1,\overline{s},\overline{\phi })`$. For $`\alpha \pi `$, set $`\overline{A}_\alpha =A_{\alpha ^1}`$. For $`a\overline{A}_{\alpha \beta }`$, set
$$\overline{\mathrm{\Delta }}_{\alpha ,\beta }(a)=(\phi _\beta \text{id}_{A_{\beta ^1}})\mathrm{\Delta }_{\beta ^1\alpha ^1\beta ,\beta ^1}(a)$$
$$=(\text{id}_{A_{\alpha ^1}}\phi _{\beta ^1})\mathrm{\Delta }_{\alpha ^1,\beta ^1}\phi _\beta (a)A_{\alpha ^1}A_{\beta ^1}=\overline{A}_\alpha \overline{A}_\beta .$$
This defines a comultiplication $`\overline{\mathrm{\Delta }}_{\alpha ,\beta }:\overline{A}_{\alpha \beta }\overline{A}_\alpha \overline{A}_\beta `$. Set $`\overline{\epsilon }_1=\epsilon _1:\overline{A}_1=A_1K`$. For $`\alpha \pi `$, set
$$\overline{s}_\alpha =\phi _\alpha s_{\alpha ^1}:\overline{A}_\alpha =A_{\alpha ^1}A_\alpha =\overline{A}_{\alpha ^1}.$$
Finally set $`\overline{\phi }_\alpha =\phi _\alpha `$ for all $`\alpha \pi `$. A direct computation shows that $`\overline{A}=(\{\overline{A}_\alpha \}_{\alpha \pi },\overline{\mathrm{\Delta }},\overline{\epsilon }_1,\overline{s},\overline{\phi })`$ is a crossed Hopf $`\pi `$-coalgebra. Moreover, if $`R,\theta `$ are a universal $`R`$-matrix and a twist in $`A`$ then the formulas
$$\overline{R}_{\alpha ,\beta }=(\text{Perm}(R_{\beta ^1,\alpha ^1}))^1\overline{A}_\alpha \overline{A}_\beta ,\overline{\theta }_\alpha =(\theta _{\alpha ^1})^1\overline{A}_\alpha $$
define a universal $`R`$-matrix and a twist in $`\overline{A}`$. It is easy to see that $`\overline{\overline{A}}=A`$.
For example, the crossed Hopf $`\pi `$-coalgebras $`A^\pi `$ and $`\overline{A}^\pi `$ defined in Section 11.2 are mirrors of each other. The ribbon Hopf $`\pi `$-coalgebras $`A^\pi `$ and $`\overline{A}^\pi `$ constructed in Section 11.5 are related as follows: $`\overline{A}^\pi `$ is the mirror of $`B^\pi `$ where $`B=(A,(\text{Perm}(R))^1,v^1)`$ and $`\pi =\pi (A)=\pi (B)`$.
11.7. Categories of representations. We shall associate with every Hopf $`\pi `$-coalgebra $`A=(\{A_\alpha \},\mathrm{\Delta },\epsilon _1,s,\phi )`$ a category of representations $`\text{Rep}(A)`$ which has a natural structure of a $`\pi `$-category. Moreover, if $`A`$ is crossed (resp. quasitriangular, ribbon) then $`\text{Rep}(A)`$ is crossed (resp. braided, ribbon).
By an $`A_\alpha `$-module we mean a left $`A_\alpha `$-module whose underlying $`K`$-module is projective of finite type. (The unit of $`A_\alpha `$ is supposed to act as the identity.) The category $`\text{Rep}(A)`$ is the disjoint union of the categories $`\{\text{Rep}(A_\alpha )\}_{\alpha \pi }`$ where $`\text{Rep}(A_\alpha )`$ is the category of $`A_\alpha `$-modules and $`A_\alpha `$-linear homomorphisms. The tensor product and the unit object in $`\text{Rep}(A)`$ are defined in the usual way using the comultiplication $`\mathrm{\Delta }`$ and the counit $`\epsilon _1`$. The associativity morphisms are the standard identification isomorphisms for modules $`(UV)W=U(VW)`$; they will be supressed from the notation. The same applies to the morphisms $`l,r`$ which are the standard identifications $`UK=U=KU`$.
For $`U\text{Rep}(A_\alpha )`$, we have $`U^{}=\text{Hom}_K(U,K)\text{Rep}(A_{\alpha ^1})`$ where $`aA_{\alpha ^1}`$ acts as the transpose of $`xs_{\alpha ^1}(a)(x):UU`$. The duality morphism $`d_U:U^{}U\text{1}\text{ I}=K`$ is the evaluation pairing; it gives rise to $`b_U`$ in the usual way, cf. \[Tu2, Chapter XI\]. The second line of equalities in (11.2.2) implies that $`d_U,b_U`$ are $`A_1`$-linear.
The automorphism $`\phi _\alpha `$ of $`A`$ defines an automorphism, $`\mathrm{\Phi }_\alpha `$, of $`\text{Rep}(A)`$. If $`U\text{Rep}(A_\beta )`$ then $`\mathrm{\Phi }_\alpha (U)`$ has the same underlying $`K`$-module as $`U`$ and each $`aA_{\alpha \beta \alpha ^1}`$ acts as multiplication by $`\phi _\alpha ^1(a)A_\beta `$. Every $`A_\beta `$-homomorphism $`UU^{}`$ is mapped to itself considered as a $`A_{\alpha \beta \alpha ^1}`$-homomorphism. It is easy to check that $`\text{Rep}(A)`$ is a crossed $`\pi `$-category.
A universal $`R`$-matrix (11.3.a) in $`A`$ induces a braiding in $`\text{Rep}(A)`$ as follows. For $`U\text{Rep}(A_\alpha ),V\text{Rep}(A_\beta )`$, the braiding $`c_{V,W}:VW{}_{}{}^{V}WV`$ is the composition of multiplication by $`R_{\alpha ,\beta }`$, permutation $`VWWV`$ and the $`K`$-isomorphism $`WV={}_{}{}^{V}WV`$ which comes from the fact that $`W={}_{}{}^{V}W`$ as $`K`$-modules. The conditions defining a universal $`R`$-matrix ensure that $`\{c_{V,W}\}_{V,W}`$ is a braiding.
A twist $`\theta `$ in $`A`$ induces a twist in $`\text{Rep}(A)`$: for any $`A_\alpha `$-module $`V`$, the morphism $`\theta _V:V{}_{}{}^{V}V`$ is the composition of multiplication by $`\theta _\alpha A_\alpha `$ and the $`K`$-isomorphism $`V{}_{}{}^{V}V`$ which comes from the fact that $`{}_{}{}^{V}V=V`$ as $`K`$-modules. Conditions (11.4.1), (11.4.4) imply that the resulting homomorphism $`V{}_{}{}^{V}V`$ is $`A_\alpha `$-linear. Condition (2.3.1) follows from definitions, conditions (2.3.2) - (2.3.4) follow from (11.4.2) - (11.4.4), respectively. Thus, $`\text{Rep}(A)`$ is a ribbon crossed $`\pi `$-category.
As an exercise the reader may check that $`\overline{\text{Rep}(A)}=\text{Rep}(\overline{A})`$.
11.8. Remarks. 1. The idea of a Hopf group-coalgebra comes from the following observation. Consider a topological group $`G`$. For $`\alpha G`$, denote by $`C_\alpha =C_\alpha (G)`$ the algebra of germs of continuous functions in $`\alpha G`$. The group multiplication $`G\times GG`$ induces an algebra homomorphism $`C_1C_1\widehat{}C_1`$ which makes $`C_1`$ a topological Hopf algebra. Similarly, the group multiplication in $`G`$ induces an algebra homomorphism $`\mathrm{\Delta }_{\alpha ,\beta }:C_{\alpha \beta }C_\alpha \widehat{}C_\beta `$ for any $`\alpha ,\beta G`$. This makes the system of algebras $`\{C_\alpha \}_{\alpha G}`$ a (topological) Hopf $`G`$-coalgebra.
In this example we can compute $`\mathrm{\Delta }_{\alpha ,\beta }`$ via $`\mathrm{\Delta }_{1,1}`$ and the adjoint action of $`G`$ on $`C_1`$ as follows. Observe first that left multiplication by $`\alpha `$ induces an algebra isomorphism, $`i_\alpha :C_1C_\alpha `$. Let $`x,y`$ be two elements of $`G`$ close to $`1G`$. Then $`\alpha x,\beta y`$ are close to $`\alpha ,\beta `$, respectively. For $`fC_1`$,
$$(i_\alpha ^1i_\beta ^1)\mathrm{\Delta }_{\alpha ,\beta }(i_{\alpha \beta }(f))(x,y)=\mathrm{\Delta }_{\alpha ,\beta }(i_{\alpha \beta }(f))(\alpha x,\beta y)$$
$$=i_{\alpha \beta }(f)(\alpha x\beta y)=f((\alpha \beta )^1\alpha x\beta y)=f(\beta ^1x\beta y)=\mathrm{\Delta }_{1,1}(f)(\beta ^1x\beta ,y).$$
This computation leads to the second example of Hopf group-coalgebras in Section 11.2.
2. Many aspects of the theory of Hopf algebras generalize to crossed Hopf $`\pi `$-coalgebras. The notion of a modular Hopf algebra \[RT\] can be extended to this setting which gives rise to modular crossed $`\pi `$-categories. The Drinfeld’s double construction and the Drinfeld’s theory of quasi-Hopf algebras also generalize to this setting. In particular, the double of a crossed Hopf $`\pi `$-coalgebra is a quasitriangular crossed Hopf $`\pi `$-coalgebra, see \[Zu\].
3. By now there is quite a number of constructions of topological invariants of 3-manifolds from Hopf algebras. Note in particular the constructions introduced by Turaev-Viro \[TV\], Hennings \[He\], and Kuperberg \[Ku\]. All of them can be generalized to our setting. This allows to derive invariants of 3-dimensional $`\pi `$-manifolds from certain crossed Hopf $`\pi `$-coalgebras, see Appendix 2 and \[Vi\].
4. There is a special case of the definitions above where all the crossed isomorphisms $`\{\phi _\alpha \}_{\alpha \pi }`$ are the identity maps. Namely, assume that $`A`$ is a Hopf $`\pi `$-coalgebra over an abelian group $`\pi `$. Then the trivial homomorphism $`\phi =1:\pi \text{Aut}(A)`$ makes $`A`$ a crossed Hopf $`\pi `$-coalgebra. In this case the definitions of the universal $`R`$-matrix and the twist in $`A`$ considerably simplify. Such quasitriangular crossed Hopf $`\pi `$-coalgebras were first considered by T. Ohtsuki \[Oh1\], \[Oh2\]. He calls them colored Hopf algebras and derives examples of such algebras from $`U_q(sl_2)`$.
12. Canonical extensions
We show in this section that a monoidal (resp. braided, ribbon) category canonically extends to a certain crossed (resp. braided, ribbon) group-category.
12.1. Extensions of a monoidal category. Let $`𝒞`$ be a $`K`$-additive monoidal category with left duality. Observe that $`𝒞`$ is a $`\{1\}`$-category where $`\{1\}`$ is a trivial group. For any group $`\pi `$, the category $`𝒞`$ gives rise to a $`\pi `$-category $`𝒞^\pi `$ obtained by pulling back $`𝒞`$ along the trivial homomorphism $`\pi 1`$. By definition, $`𝒞^\pi =_{\alpha \pi }𝒞_\alpha ^\pi `$ where the objects of $`𝒞_\alpha ^\pi `$ are pairs $`(U𝒞,\alpha )`$. A morphism $`(U,\alpha )(V,\beta )`$ with $`\alpha ,\beta \pi `$ is $`0\text{Hom}_𝒞(U,V)`$ if $`\alpha \beta `$ and any element of $`\text{Hom}_𝒞(U,V)`$ if $`\alpha =\beta `$. The operations on objects and the unit object are defined by
$$(U,\alpha )(V,\beta )=(UV,\alpha \beta ),(U,\alpha )^{}=(U^{},\alpha ^1),\text{1}\text{ I}_{𝒞^\pi }=(\text{1}\text{ I}_𝒞,1).$$
The structural morphisms (1.1.a,b,e) and composition and tensor product of morphisms are induced by the corresponding operations in $`𝒞`$ in the obvious way.
Assume now that $`\pi `$ acts on $`𝒞`$ by automorphisms. Such an action is determined by a group homomorphism $`\pi \text{Aut}(𝒞)`$ where $`\text{Aut}(𝒞)`$ is the group of automorphisms of $`𝒞`$ defined in Section 2.1. Then $`𝒞^\pi `$ acquires the structure of a crossed $`\pi `$-category as follows. For $`\alpha \pi `$ and $`(V,\beta )𝒞_\beta ^\pi `$, set
$$\phi _\alpha (V,\beta )=(\alpha (V),\alpha \beta \alpha ^1)𝒞_{\alpha \beta \alpha ^1}^\pi .$$
For a morphism $`f:(V,\beta )(W,\gamma )`$ in $`𝒞^\pi `$ with $`f\text{Hom}_𝒞(V,W)`$ set
$$\phi _\alpha (f)=\alpha (f)\text{Hom}_𝒞(\alpha (V),\alpha (W))=\text{Hom}_{𝒞^\pi }(\phi _\alpha (V,\beta ),\phi _\alpha (W,\gamma )).$$
All the axioms of a crossed $`\pi `$-category follow from definitions. Applying this construction to $`\pi =\text{Aut}(𝒞)`$, we obtain a canonical extension of $`𝒞`$ to a crossed $`\text{Aut}(𝒞)`$-category.
12.2. The group $`\text{Aut}_0(𝒞)`$. Let $`𝒞`$ be a $`K`$-additive monoidal category with left duality. Denote by $`\text{id}_𝒞`$ the identity functor $`𝒞𝒞`$ sending each object and each morphism of $`𝒞`$ into itself. We introduce a group $`\text{Aut}_0(𝒞)`$, formed by monoidal equivalences of $`\text{id}_𝒞`$ to automorphisms of $`𝒞`$. More precisely, an element of $`\text{Aut}_0(𝒞)`$ is a pair ($`\alpha \text{Aut}(𝒞)`$, an invertible monoidal morphism of functors $`F:\text{id}_𝒞\alpha `$). The latter means that for each object $`U𝒞`$ we have an invertible morphism $`F_U:U\alpha (U)`$ in $`𝒞`$ such that
(i) for any morphism $`f:UV`$ in $`𝒞`$ the following diagram is commutative:
$$\begin{array}{ccc}U& \stackrel{f}{}& V\\ F_U& & F_V& & \\ \alpha (U)& \stackrel{\alpha (f)}{}& \alpha (V);\end{array}$$
(ii) $`F_{\text{1}\text{ I}}=\text{id}_{\text{1}\text{ I}}`$ and for any $`U,V𝒞`$, we have $`F_{UV}=F_UF_V`$.
The product of two pairs $`(\alpha ,F),(\alpha ^{},F^{})\text{Aut}_0(𝒞)`$ is the pair $`(\alpha \alpha ^{},FF^{})`$ where for each $`U𝒞`$ we have $`(FF^{})_U=F_{\alpha ^{}(U)}F_U^{}:U(\alpha \alpha ^{})(U)`$. It is clear that $`\text{Aut}_0(𝒞)`$ is a group with respect to this multiplication. The pair $`(\alpha =\text{id}_𝒞,F=\{\text{id}_U\}_{U𝒞})`$ is the unit object of $`\text{Aut}_0(𝒞)`$.
Forgetting $`F`$, we obtain a group homomorphism $`i_0:\text{Aut}_0(𝒞)\text{Aut}(𝒞)`$ whose image consists of the automorphisms of $`𝒞`$ monoidally equivalent to $`\text{id}_𝒞`$. The key property of the elements of the image is given by the following lemma.
12.2.1. Lemma. If $`𝒞`$ is braided then all elements of $`i_0(\text{Aut}_0(𝒞))\text{Aut}(𝒞)`$ preserve the braiding. If $`𝒞`$ is ribbon then all elements of $`i_0(\text{Aut}_0(𝒞))`$ preserve both the braiding and the twist.
Proof. Let $`\{c_{U,V}:UVVU\}_{U,V𝒞}`$ be a braiding in $`𝒞`$. Let $`(\alpha ,F)\text{Aut}_0(𝒞)`$. Then for any $`U,V𝒞`$ we have a commutative diagram
$$\begin{array}{ccc}UV& \stackrel{c_{U,V}}{}& VU\\ F_{UV}& & F_{VU}& & \\ \alpha (U)\alpha (V)& \stackrel{\alpha (c_{U,V})}{}& \alpha (V)\alpha (U).\end{array}$$
$`(12.2.a)`$
On the other hand, by the naturality of the braiding we have
$$F_{VU}c_{U,V}=(F_VF_U)c_{U,V}=c_{\alpha (U),\alpha (V)}(F_UF_V)=c_{\alpha (U),\alpha (V)}F_{UV}.$$
Thus, if we replace in (12.2.a) the bottom arrow by $`c_{\alpha (U),\alpha (V)}`$ we obtain a commutative diagram. Since all the arrows in (12.2.a) are invertible morphisms, $`c_{\alpha (U),\alpha (V)}=\alpha (c_{U,V})`$. Similarly, if $`\{\theta _U:UU\}_{U𝒞}`$ is a twist in $`𝒞`$ then for any $`U𝒞`$ we have a commutative diagram
$$\begin{array}{ccc}U& \stackrel{\theta _U}{}& U\\ F_U& & F_U& & \\ \alpha (U)& \stackrel{\alpha (\theta _U)}{}& \alpha (U).\end{array}$$
By the naturality of the twist we have $`F_U\theta _U=\theta _{\alpha (U)}F_U`$ and therefore $`\alpha (\theta _U)=\theta _{\alpha (U)}`$.
12.3. The canonical extension of a ribbon category. Let $`𝒞`$ be a braided $`K`$-additive monoidal category with left duality. Set $`\pi =\text{Aut}_0(𝒞)`$. We define here a canonical extension of $`𝒞`$ to a braided crossed $`\pi `$-category. Applying the constructions of Section 12.1 to the group homomorphism $`i_0:\pi =\text{Aut}_0(𝒞)\text{Aut}(𝒞)`$ we obtain a crossed $`\pi `$-category $`𝒞^\pi `$. The braiding $`\{c_{U,V}:UVVU\}_{U,V𝒞}`$ in $`𝒞`$ induces a braiding in $`𝒞^\pi `$ as follows. Let $`u=(U,(\alpha ,F)),v=(V,(\beta ,G))`$ be objects of $`𝒞^\pi `$ where $`U,V𝒞`$ and $`(\alpha ,F),(\beta ,G)\pi `$. Observe that
$$uv=(UV,(\alpha ,F)(\beta ,G)),\phi _{(\alpha ,F)}(v)=(\alpha (V),(\alpha ,F)(\beta ,G)(\alpha ,F)^1),$$
$${}_{}{}^{u}vu=\phi _{(\alpha ,F)}(v)u=(\alpha (V)U,(\alpha ,F)(\beta ,G)).$$
The invertible morphism $`(F_V\text{id}_U)c_{U,V}:UV\alpha (V)U`$ in $`𝒞`$ defines an invertible morphism in $`𝒞^\pi `$
$$c_{u,v}:uv{}_{}{}^{u}vu.$$
$`(12.3.a)`$
If $`𝒞`$ is a ribbon category, then the twist $`\{\theta _U:UU\}_{U𝒞}`$ in $`𝒞`$ induces a twist in $`𝒞^\pi `$ as follows. Let $`u=(U,(\alpha ,F))𝒞^\pi `$. Then the invertible morphism $`F_U\theta _U:U\alpha (U)`$ in $`𝒞`$ defines an invertible morphism in $`𝒞^\pi `$
$$\theta _u:u{}_{}{}^{u}u=\phi _{(\alpha ,F)}(u)=(\alpha (U),(\alpha ,F)).$$
$`(12.3.b)`$
12.3.1. Theorem. The morphisms (12.3.a) with $`u,v𝒞^\pi `$ form a braiding in $`𝒞^\pi `$. If $`𝒞`$ is ribbon, then the morphisms (12.3.b) with $`u𝒞^\pi `$ form a twist in $`𝒞^\pi `$. If $`𝒞`$ is modular, then so is $`𝒞^\pi `$.
The proof goes by a routine verification of the axioms, we leave it to the reader. Note that the neutral component of $`𝒞^\pi `$ is $`𝒞`$ with its original braiding and twist.
12.4. Remarks and examples. 1. Condition (12.2.i) on a pair $`(\alpha ,F)\text{Aut}_0(𝒞)`$ shows that the action of $`\alpha `$ on morphisms is completely determined by $`F`$. Setting $`F(U)=\alpha (U)`$ for $`U𝒞`$, we can reformulate the definition of $`\text{Aut}_0(𝒞)`$ entirely in terms of $`F`$. An element of $`\text{Aut}_0(𝒞)`$ is thus described as a pair (a bijection $`F`$ from the set of objects of $`𝒞`$ into itself, a system of invertible morphisms $`\{F_U:UF(U)\}_{U𝒞}`$) such that
(i) $`F(\text{1}\text{ I})=\text{1}\text{ I}`$ and $`F_{\text{1}\text{ I}}=\text{id}_{\text{1}\text{ I}}`$;
(ii) for any $`U,V𝒞`$, we have $`F(UV)=F(U)F(V)`$ and $`F_{UV}=F_UF_V`$,
(iii) for any $`U𝒞`$, we have $`F(U^{})=(F(U))^{}`$ and $`F_U^{}=((F_U)^{})^1`$ where $`(F_U)^{}:(F(U))^{}U^{}`$ is the transpose of $`F_U`$.
2. Assume that in Example 2.6, the group $`\pi `$ is abelian. Then the corresponding category $`𝒞`$ is a ribbon monoidal category in the usual sense of the word. Any automorphism of $`𝒞`$ monoidally equivalent to $`\text{id}_𝒞`$ is equal to $`\text{id}_𝒞`$ since all non-zero morphisms in $`𝒞`$ are proportional to the identity morphisms of objects. An element $`(\alpha ,F)\text{Aut}_0(𝒞)`$ is therefore completely determined by the map $`UF_U\text{Hom}_𝒞(U,U)=K`$ where $`U`$ runs over the elements of $`\pi `$. The inclusion $`(\alpha ,F)\text{Aut}_0(𝒞)`$ is equivalent to the condition that this map is a group homomorphism $`\pi K^{}K`$. Hence, $`\text{Aut}_0(𝒞)=\text{Hom}(\pi ,K^{})=\pi ^{}`$. By Theorem 12.3.1, $`𝒞`$ gives rise to a ribbon crossed $`\pi ^{}`$-category $`𝒞^\pi ^{}`$.
3. Let $`G`$ be a topological group and $`𝒞=\text{Rep}(G)`$ be the category of finite dimensional linear representations of $`G`$ and $`G`$-linear homomorphisms. It is clear that $`𝒞`$ is a $`K`$-additive monoidal category with left duality. Moreover, this category is braided (in fact symmetric) with braiding given by the flips (permutations) $`UVVU`$. The category $`𝒞`$ is ribbon with twist $`\theta _U=\text{id}_U:UU`$ for all $`U𝒞`$. There is a group homomorphism $`g(\alpha ^g,F^g):G\text{Aut}_0(𝒞)`$ defined as follows. Given $`gG`$, the functor $`\alpha ^g:𝒞𝒞`$ sends a $`G`$-module $`(U,\rho _U:G\text{End}(U))`$ into the same $`K`$-module $`U`$ where each $`hG`$ acts as $`\rho _U(g^1hg)`$. The functor $`\alpha ^g`$ sends any $`G`$-linear homomorphism $`f:UV`$ into itself. It is clear that $`\alpha ^g\text{Aut}(𝒞)`$. The morphism $`F_U^g:U\alpha ^g(U)`$ in $`𝒞`$ is defined by $`u\rho _U(g^1)(u):UU`$ where $`u`$ runs over $`U`$. The family $`\{F_U^g\}_{U𝒞}`$ satisfies the conditions of Section 12.2. Thus, $`(\alpha ^g,F^g)\text{Aut}_0(𝒞)`$. Pulling back the canonical extension $`𝒞^{\text{Aut}_0(𝒞)}`$ along the group homomorphism $`g(\alpha ^g,F^g):G\text{Aut}_0(𝒞)`$ we obtain a ribbon crossed $`G`$-category $`(\text{Rep}(G))^G`$.
If $`G`$ is a semisimple complex connected Lie group with Lie algebra $`𝕘`$, then the category $`\text{Rep}(G)`$ coincides with the category $`U(𝕘)`$-mod of finite dimensional $`U(𝕘)`$-modules. It would be most interesting to find a quantum deformation of the ribbon crossed $`G`$-category $`(\text{Rep}(G))^G=(U(𝕘)\text{mod})^G`$ generalizing the deformation $`U_q(𝕘)`$-mod of $`U(𝕘)`$-mod arising in the theory of quantum groups.
4. One can generalize the constructions of this section to the case where the initial category $`𝒞`$ is a (braided or ribbon) crossed group-category; one should then consider only those automorphisms of $`𝒞`$ which are compatible with the given group action. We shall not pursue this line here.
13. Transfer of categories
Throughout this section $`G\pi `$ is a subgroup of a group $`\pi `$. We shall show that each $`G`$-category $`𝒞`$ gives rise to a $`\pi `$-category $`\stackrel{~}{𝒞}`$ via a natural transfer. If $`𝒞`$ is crossed (resp. braided, ribbon) then $`\stackrel{~}{𝒞}`$ is crossed (resp. braided, ribbon).
13.1. Transfer for group-categories. Fix a $`G`$-category $`𝒞`$. We shall construct a $`\pi `$-category $`\stackrel{~}{𝒞}`$ called the transfer of $`𝒞`$.
Fix a representative $`\omega _i\pi `$ for each right coset class $`iG\backslash \pi `$ so that $`i=G\omega _i\pi `$. For $`\alpha \pi `$, set
$$N(\alpha )=\{iG\backslash \pi |\omega _i\alpha \omega _i^1G\}G\backslash \pi .$$
An object $`U`$ of $`\stackrel{~}{𝒞}`$ is a triple $`(\alpha \pi `$, a subset $`A`$ of $`N(\alpha )`$, a family $`\{U_i𝒞_{\omega _i\alpha \omega _i^1}\}_{iA}`$). The set $`A`$ (which may be void) will be denoted by $`|U|`$. The morphisms in $`\stackrel{~}{𝒞}`$ are defined by
$$\text{Hom}_{\stackrel{~}{𝒞}}(U,U^{})=\underset{i|U||U^{}|}{}\text{Hom}_𝒞(U_i,U_i^{}).$$
Thus a morphism $`f:UU^{}`$ in $`\stackrel{~}{𝒞}`$ is a family $`\{f_i:U_iU_i^{}\}_{i|U||U^{}|}`$ of morphisms in $`𝒞`$. We view each $`f_i`$ as the $`i`$-th coordinate of $`f`$. The $`K`$-linear structure in $`\text{Hom}_{\stackrel{~}{𝒞}}(U,U^{})`$ is coordinate-wise. The composition of two morphisms $`f:UU^{}`$ and $`f^{}:U^{}U^{\prime \prime }`$ is defined in coordinates by
$$(f^{}f)_i=\{\begin{array}{cc}f_i^{}f_i:U_iU_i^{\prime \prime },ifi|U||U^{}||U^{\prime \prime }|,\hfill & \\ 0:U_iU_i^{\prime \prime },ifi(|U||U^{\prime \prime }|)\backslash |U^{}|.\hfill & \end{array}$$
This is an associative composition in $`\stackrel{~}{𝒞}`$ because the composition of morphisms in $`𝒞`$ is associative and the composition of a zero morphism in $`𝒞`$ with any other morphism is again a zero morphism. This defines $`\stackrel{~}{𝒞}`$ as a $`K`$-additive category. We have a splitting $`\stackrel{~}{𝒞}=_{\alpha \pi }\stackrel{~}{𝒞}_\alpha `$ where $`\stackrel{~}{𝒞}_\alpha `$ is the full subcategory of $`\stackrel{~}{𝒞}`$ with objects $`(\alpha ,A,\{U_i\}_{iA})`$.
The unit object of $`\stackrel{~}{𝒞}`$ is the triple $`(1\pi ,A=G\backslash \pi ,\{U_i=\text{1}\text{ I}_𝒞\}_{iA})`$. The duality and tensor product for objects of $`\stackrel{~}{𝒞}`$ are defined by
$$(\alpha ,A,\{U_i\}_{iA})^{}=(\alpha ^1,A,\{U_i^{}\}_{iA}),$$
$$(\alpha ,A,\{U_i\}_{iA})(\beta ,B,\{V_j\}_{jB})=(\alpha \beta ,AB,\{U_iV_i\}_{iAB}).$$
Note that $`N(\alpha )=N(\alpha ^1)`$ and that the inclusions $`AN(\alpha ),BN(\beta )`$ imply that $`ABN(\alpha \beta )`$. Observe the identities $`|U^{}|=|U|,|UV|=|U||V|`$.
The tensor product of morphisms $`f:UU^{}`$ and $`g:VV^{}`$ is defined by
$$(fg)_i=f_ig_i:U_iV_iU_i^{}V_i^{}$$
for all $`i|U||V||U^{}||V^{}|`$. It is a simple exercise to check the identity $`(f^{}g^{})(fg)=f^{}fg^{}g`$.
The structural morphisms $`a,l,r,b,d`$ in $`\stackrel{~}{𝒞}`$ are all defined coordinate-wisely and their coordinates are the corresponding morphisms in $`𝒞`$. In particular, for every $`U\stackrel{~}{𝒞}`$ we define $`b_U:\text{1}\text{ I}UU^{}`$ by $`(b_U)_i=b_{U_i}:\text{1}\text{ I}U_i(U_i)^{}`$ where $`i|U|`$. Similarly, $`(d_U)_i=d_{U_i},(l_U)_i=l_{U_i},(r_U)_i=r_{U_i}`$ where $`i|U|`$. The associativity morphisms are defined by $`(a_{U,V,W})_i=a_{U_i,V_i,W_i}`$ for all $`i|U||V||W|`$. The naturality of $`a,l,r`$ and the identities (1.1.c,d,f,g) follow from the corresponding properties of $`𝒞`$. In general, the isomorphism class of $`\stackrel{~}{𝒞}`$ depends on the choice of $`\omega _ii`$.
13.2. Transfer for crossed group-categories. For each crossed $`G`$-category $`(𝒞,\phi :G\text{Aut}(𝒞))`$, its transfer $`\stackrel{~}{𝒞}`$ is a crossed $`\pi `$-category. We need only to define the action of $`\pi `$ on $`\stackrel{~}{𝒞}`$. Fix the representatives $`\omega _ii`$ for $`iG\backslash \pi `$ needed in the construction of $`\stackrel{~}{𝒞}`$. Consider the left action of $`\pi `$ on $`G\backslash \pi `$ defined by $`\alpha (i)=i\alpha ^1`$ for $`\alpha \pi ,iG\backslash \pi `$. We have $`G\omega _{\alpha (i)}=G\omega _i\alpha ^1`$ so that $`\alpha _i=\omega _{\alpha (i)}\alpha \omega _i^1G`$ for all $`\alpha \pi ,iG\backslash \pi `$.
For $`\beta \pi `$, the map $`j\alpha (j)`$ sends bijectively $`N(\beta )`$ onto $`N(\alpha \beta \alpha ^1)`$. For every $`jN(\beta )`$, we have the functor
$$\phi _{\alpha _j}:𝒞_{\omega _j\beta \omega _j^1}𝒞_{\omega _{\alpha (j)}\alpha \beta \alpha ^1(\omega _{\alpha (j)})^1}.$$
Given an object $`V=(\beta ,B,\{V_j\}_{jB})`$ of $`\stackrel{~}{𝒞}_\beta `$, we apply $`\{\phi _{\alpha _j}\}_{jB}`$ coordinate-wisely to obtain an object $`\stackrel{~}{\phi }_\alpha (V)=(\alpha \beta \alpha ^1,\alpha (B),\{\phi _{\alpha _j}(V_j)\}_{jB})`$ of $`\stackrel{~}{𝒞}_{\alpha \beta \alpha ^1}`$. More precisely,
$$\stackrel{~}{\phi }_\alpha (V)=(\alpha \beta \alpha ^1,\alpha (B),\{\phi _{\alpha _{\alpha ^1(i)}}(V_{\alpha ^1(i)})\}_{i\alpha (B)}).$$
Clearly, $`|\stackrel{~}{\phi }_\alpha (V)|=\alpha (|V|)`$. The action of $`\stackrel{~}{\phi }_\alpha `$ on a morphism $`f:VV^{}`$ is given by
$$\stackrel{~}{\phi }_\alpha (f)=\{\phi _{\alpha _{\alpha ^1(i)}}(f_{\alpha ^1(i)}:V_{\alpha ^1(i)}V_{\alpha ^1(i)}^{})\}_{i\alpha (|V|)\alpha (|V^{}|)}.$$
The functor $`\stackrel{~}{\phi }_\alpha `$ preserves the structural morphisms in $`\stackrel{~}{𝒞}`$ because by assumptions the functors $`\{\phi _{\alpha _j}\}_j`$ preserve the structural morphisms in $`𝒞`$. The category $`\stackrel{~}{𝒞}`$ with action $`\stackrel{~}{\phi }`$ of $`\pi `$ satisfies all the axioms of a crossed $`\pi `$-category. The isomorphism class of $`\stackrel{~}{𝒞}`$ does not depend on the choice of $`\omega _ii`$.
13.3. Transfer for braided and ribbon group-categories. If $`𝒞`$ is a braided crossed $`G`$-category then its transfer $`\stackrel{~}{𝒞}`$ acquires the structure of a braided $`\pi `$-category as follows. Let $`U=(\alpha ,A,\{U_i\}_{iA})`$ and $`V=(\beta ,B,\{V_j\}_{jB})`$ be objects of $`\stackrel{~}{𝒞}`$. By definition,
$$UV=(\alpha \beta ,AB,\{U_iV_i\}_{iAB}),$$
$${}_{}{}^{U}V=\stackrel{~}{\phi }_\alpha (V)=(\alpha \beta \alpha ^1,\alpha (B),\{\phi _{\alpha _{\alpha ^1(i)}}(V_{\alpha ^1(i)})\}_{i\alpha (B)}),$$
$${}_{}{}^{U}VU=(\alpha \beta ,\alpha (B)A,\{\phi _{\alpha _{\alpha ^1(i)}}(V_{\alpha ^1(i)})U_i\}_{i\alpha (B)A}).$$
The latter expression simplifies if we observe that for any $`iN(\alpha )`$, we have $`G\omega _i=G\omega _i\alpha ^1`$ and therefore $`\alpha (i)=i`$. This implies that the map $`i\alpha (i):G\backslash \pi G\backslash \pi `$ is the identity on $`A`$. Hence, $`\alpha (B)A=BA=AB`$ and
$${}_{}{}^{U}VU=(\alpha \beta ,AB,\{\phi _{\alpha _i}(V_i)U_i\}_{iAB}).$$
Now, we define a morphism $`c_{U,V}:UV{}_{}{}^{U}VU`$ by its coordinates $`(c_{U,V})_i=c_{U_i,V_i}:U_iV_i\phi _{\alpha _i}(V_i)U_i`$ where $`i`$ runs over $`AB`$ and $`c_{U_i,V_i}`$ is the given braiding in $`𝒞`$. Note that $`\phi _{\alpha _i}(V_i)=\phi _{\omega _i\alpha \omega _i^1}(V_i)={}_{}{}^{U_i}(V_i)`$. The morphisms $`\{c_{U,V}\}_{U,V}`$ satisfy all the axioms of a braiding in the crossed $`\pi `$-category $`\stackrel{~}{𝒞}`$; this is easily verified coordinate-wisely.
Similarly, if $`𝒞`$ is a ribbon crossed $`G`$-category then its transfer $`\stackrel{~}{𝒞}`$ has the structure of a ribbon $`\pi `$-category. For an object $`U=(\alpha ,A,\{U_i\}_{iA})`$ of $`\stackrel{~}{𝒞}`$ we have $`{}_{}{}^{U}U=(\alpha ,A,\{\phi _{\alpha _i}(U_i)\}_{iA})`$. Here $`\alpha _i=\omega _i\alpha \omega _i^1`$ so that $`\phi _{\alpha _i}(U_i)={}_{}{}^{U_i}(U_i)`$. The twist $`\theta _U:U{}_{}{}^{U}U`$ is defined by its coordinates $`(\theta _U)_i=\theta _{U_i}:U_i{}_{}{}^{U_i}(U_i)`$ where $`i`$ runs over $`A`$. The morphisms $`\{\theta _U\}_U`$ satisfy all the axioms of a twist in the braided $`\pi `$-category $`\stackrel{~}{𝒞}`$.
If $`G\pi `$ then the transfer $`\stackrel{~}{𝒞}`$ of a modular $`G`$-category $`𝒞`$ is not modular: the algebra of endomorphisms of the unit object of $`\stackrel{~}{𝒞}`$ equals $`(\text{End}_𝒞(\text{1}\text{ I}_𝒞))^{[\pi :G]}=K^{[\pi :G]}K`$.
Appendix 1. Quasi-abelian cohomology of groups
Let $`a=\{a_{\alpha ,\beta ,\gamma }K^{}\}_{\alpha ,\beta ,\gamma \pi }`$ be a 3-cocycle of a group $`\pi `$ with values in $`K^{}`$ and $`c=\{c_{\alpha ,\beta }K^{}\}_{\alpha ,\beta \pi }`$ be a family of elements of $`K^{}`$. Equations (2.6.a,c,e,f) on the pair $`(a,c)`$ are equivalent to the equations introduced by C. Ospel \[Os\] from a different viewpoint (in his notation $`c_{\alpha ,\beta }=\mathrm{\Omega }_{\beta ,\alpha }`$ and $`a_{\alpha ,\beta ,\gamma }=f(\alpha ,\beta ,\gamma )`$). Following Ospel, we call such pairs $`(a,c)`$ satisfying (2.6.a,c,e,f) quasi-abelian 3-cocycles on $`\pi `$. Examples of such cocycles are provided by pairs $`(a,c)`$ where $`a=1`$ and $`c`$ is a bilinear form $`(\alpha ,\beta )c_{\alpha ,\beta }:H_1(\pi )\times H_1(\pi )K^{}`$.
Quasi-abelian 3-cocycles $`(a,c)`$ on $`\pi `$ form a commutative group under pointwise multiplication with unit $`a=1,c=1`$. This group contains the coboundaries of the conjugation invariant 2-cochains. A conjugation invariant 2-cochain is a map $`(\alpha ,\beta )\eta _{\alpha ,\beta }:\pi \times \pi K^{}`$ such that $`\eta _{\delta \alpha \delta ^1,\delta \beta \delta ^1}=\eta _{\alpha ,\beta }`$ for any $`\alpha ,\beta ,\delta \pi `$. Its coboundary is defined by
$$a_{\alpha ,\beta ,\gamma }=\eta _{\alpha ,\beta }\eta _{\alpha \beta ,\gamma }(\eta _{\alpha ,\beta \gamma })^1(\eta _{\beta ,\gamma })^1,c_{\alpha ,\beta }=\eta _{\alpha ,\beta }(\eta _{\beta ,\alpha })^1.$$
A direct computation shows that this is a quasi-abelian 3-cocycle. The quotient of the group of quasi-abelian 3-cocycles by the subgroup of coboundaries is the group of quasi-abelian cohomology of $`\pi `$ denoted $`H_{qa}^3(\pi ;K^{})`$. Involution (2.6.i) transforms the subgroup of coboundaries into itself and defines an involution on $`H_{qa}^3(\pi ;K^{})`$. Forgetting $`c`$ we obtain a homomorphism to the usual cohomology $`H_{qa}^3(\pi ;K^{})H^3(\pi ;K^{})`$.
The constructions of Section 2.6 associate with any quasi-abelian 3-cocycle of $`\pi `$ and a conjugation invariant family $`\{b_\alpha K^{}\}_{\alpha \pi }`$ a certain braided crossed $`\pi `$-category. A twist in this category is determined by a family $`\{\theta _\alpha K^{}\}_{\alpha \pi }`$ satisfying equations (2.6.g,h). The next lemma explicitly describes all the solutions to (2.6.g,h).
Lemma. Let $`(a,c)`$ be a quasi-abelian 3-cocycle of $`\pi `$. Then $`\{\theta _\alpha =c_{\alpha ,\alpha }\}_{\alpha \pi }`$ satisfies (2.6.g,h). A general solution to (2.6.g,h) is the product of this solution with a group homomorphism $`\pi \{kK^{}|k^2=1\}`$.
Proof. It is obvious that any two solutions to (2.6.g) (with given $`c`$) are obtained from each other by multiplication by a group homomorphism $`\pi K^{}`$. The condition $`\theta _{\alpha ^1}=\theta _\alpha `$ implies that this homomorphism takes values in the group $`\{kK|k^2=1\}K^{}`$.
Let us prove the first claim. A direct computation using first (2.6.e) and then (2.6.f), (2.6.a) yields
$$c_{\alpha \beta ,\alpha \beta }=c_{\alpha ,\alpha }c_{\beta ,\beta }c_{\alpha ,\beta }c_{\beta ,\alpha }\times $$
$$\times [(a_{\alpha \beta ,\alpha ,\beta }a_{\alpha ,\beta ,\alpha \beta })^1a_{\alpha \beta \alpha \beta ^1\alpha ^1,\alpha \beta ,\beta }a_{\beta \alpha \beta ^1,\beta ,\alpha }a_{\alpha \beta \alpha ^1,\alpha ,\beta }].$$
Formula (2.6.g) follows from the fact that the expression in the square brackets is equal to 1. This can be obtained from (1.3.a) by applying substitution $`\alpha \beta \alpha \beta ^1,\gamma \alpha ,\delta \alpha ^1\beta \alpha `$ and using (2.6.a).
It remains to check the identity $`c_{\alpha ^1,\alpha ^1}=c_{\alpha ,\alpha }`$. Applying (2.6.e) to $`\beta =\gamma =1`$ we obtain $`c_{\alpha ,1}=a_{\alpha ,1,1}a_{1,1,\alpha }`$. Applying (2.6.e) to $`\beta =\alpha ,\gamma =\alpha ^1`$ we obtain
$$c_{\alpha ,\alpha ^1}=(c_{\alpha ,\alpha })^1c_{\alpha ,1}a_{\alpha ,\alpha ^1,\alpha }=(c_{\alpha ,\alpha })^1a_{\alpha ,1,1}a_{1,1,\alpha }a_{\alpha ,\alpha ^1,\alpha }.$$
Similarly, substituting $`\beta =\delta =1,\gamma =\alpha `$ and $`\beta =\alpha ,\gamma =\alpha ^1,\delta =\alpha ^1`$ in (2.6.f) we obtain $`c_{1,\alpha }=(a_{\alpha ,1,1}a_{1,1,\alpha })^1`$ and
$$c_{\alpha ^1,\alpha ^1}=c_{1,\alpha ^1}(c_{\alpha ,\alpha ^1}a_{\alpha ^1,\alpha ,\alpha ^1})^1=(c_{\alpha ,\alpha ^1}a_{\alpha ^1,1,1}a_{1,1,\alpha ^1}a_{\alpha ^1,\alpha ,\alpha ^1})^1.$$
Substituting here the expressions for $`c_{\alpha ,\alpha ^1}`$ obtained above and using the fact that the righ-hand sides of (1.3.b) and (1.3.c) are equal we obtain $`c_{\alpha ^1,\alpha ^1}=c_{\alpha ,\alpha }`$.
Appendix 2. State sum invariants of 3-dimensional $`\pi `$-manifolds
1. Spherical categories. We begin with a definition of a pivotal monoidal category, cf. \[FY\], \[BW2\]. Let $`𝒞`$ be a $`K`$-additive monoidal category with (left) duality $`b,d`$ and structural morphisms $`a,l,r`$, as in Section 1. Using these morphisms one can define for any objects $`V,W𝒞`$ a canonical isomorphism $`\gamma _{V,W}:V^{}W^{}(WV)^{}`$ (see, for instance, \[Tu2, p.31\]). In particular, if $`𝒞`$ is strict then
$$\gamma _{V,W}=(d_V\text{id}_{(WV)^{}})(\text{id}_V^{}d_W\text{id}_V\text{id}_{(WV)^{}})(\text{id}_V^{}\text{id}_W^{}b_{WV}),$$
$$\gamma _{V,W}^1=(d_{WV}\text{id}_V^{}\text{id}_W^{})(\text{id}_{(WV)^{}}\text{id}_Wb_V\text{id}_W^{})(\text{id}_{(WV)^{}}b_W).$$
Recall also that the duality defines a contravariant functor $`:𝒞𝒞`$ (see Section 2.4).
The data needed to make $`𝒞`$ pivotal is a system of invertible morphisms $`\{\tau _V:VV^{}\}_{V𝒞}`$ satisfying the following four conditions:
(1) $`\tau `$ is a natural transformation $`\text{id}_𝒞`$, i.e., for any morphism $`f:VW`$ in $`𝒞`$, the following diagram is commutative:
$$\begin{array}{ccc}V& \stackrel{f}{}& W\\ \tau _V& & \tau _W& & \\ V^{}& \stackrel{f^{}}{}& W^{};\end{array}$$
(2) for any objects $`V,W𝒞`$, the following diagram is commutative:
$$\begin{array}{ccc}VW& \stackrel{\tau _{VW}}{}& (VW)^{}\\ \tau _V\tau _W& & \gamma _{W,V}^{}& & \\ V^{}W^{}& \stackrel{\gamma _{V^{},W^{}}}{}& (W^{}V^{})^{};\end{array}$$
(3) for any object $`V𝒞`$, we have $`\tau _V^{}=(\tau _V^1)^{}:V^{}V^{}`$;
(4) the morphism $`\nu =(r_{\text{1}\text{ I}^{}})^1b_{\text{1}\text{ I}}:\text{1}\text{ I}\text{1}\text{ I}^{}`$ is invertible and $`\tau _{\text{1}\text{ I}}=(\nu ^1)^{}\nu `$.
Let $`𝒞`$ be a pivotal category. For a morphism $`f:VV`$ in $`𝒞`$, the trace $`\text{tr}(f)\text{End}_𝒞(\text{1}\text{ I})`$ is defined by
$$\text{tr}(f)=d_V^{}(\tau _Vf\text{id}_V^{})b_V:\text{1}\text{ I}\text{1}\text{ I}.$$
The trace verifies $`\text{tr}(fg)=\text{tr}(gf)`$ for any morphisms $`f:VW,g:WV`$.
Following \[BW2\], we call a pivotal category $`𝒞`$ spherical if $`\text{tr}(f)=\text{tr}(f^{})`$ for any endomorphism $`f`$ of an object in $`𝒞`$. In a spherical category we always have $`\text{tr}(fg)=\text{tr}(f)\text{tr}(g)`$ where on the right-hand side we use multiplication in $`\text{End}_𝒞(\text{1}\text{ I})`$ defined by composition. This multiplication makes $`\text{End}_𝒞(\text{1}\text{ I})`$ a commutative $`K`$-algebra with unit $`\text{id}_{\text{1}\text{ I}}`$. In a spherical category, $`\text{tr}(\text{id}_{\text{1}\text{ I}})=\text{id}_{\text{1}\text{ I}}`$.
Barrett and Westbury \[BW1\] (see also \[GK\]) extended the state sum approach to 3-manifold invariants introduced in \[TV\] to spherical categories satisfying a few extra conditions. Namely, they showed that the state sum approach works for any finite semisimple spherical category $`𝒞`$ over a field $`K`$ such that $`\text{End}(\text{1}\text{ I}_𝒞)=K`$. Thus, such a category gives rise to a $`K`$-valued topological invariant of closed oriented 3-manifolds.
2. Remarks. 1. A pivotal category $`𝒞`$ has a right duality $`\stackrel{~}{b},\stackrel{~}{d}`$ defined by
$$\stackrel{~}{b}_V=(\text{id}_V^{}\tau _V^1)b_V^{}:\text{1}\text{ I}V^{}V,\stackrel{~}{d}_V=d_V^{}(\tau _V\text{id}_V^{}):VV^{}\text{1}\text{ I}.$$
One can check that the contravariant functor defined as in Section 2.4 but using $`\stackrel{~}{b},\stackrel{~}{d}`$ coincides with the duality functor defined in Section 2.4 using $`b,d`$. Similarly, the natural isomorphisms $`V^{}W^{}(WV)^{}`$ defined using the right (resp. left) duality coincide. Note also the equality $`\nu =r_{\text{1}\text{ I}^{}}^1b_{\text{1}\text{ I}}=l_{\text{1}\text{ I}^{}}^1\stackrel{~}{b}_{\text{1}\text{ I}}:\text{1}\text{ I}\text{1}\text{ I}^{}`$.
2. Our definition of a pivotal category differs from but is essentially equivalent to the one in \[BW2\]. Barrett and Westbury define a pivotal category as a monoidal category $`𝒞`$ equipped with a contravariant functor $`:𝒞𝒞`$, an isomorphism $`\nu :\text{1}\text{ I}\text{1}\text{ I}^{}`$, morphisms $`\{b_V:\text{1}\text{ I}VV^{},\tau _V:VV^{}\}_{V𝒞}`$, and invertible morphisms $`\{\gamma _{V,W}:V^{}W^{}(WV)^{}\}_{V,W𝒞}`$ satisfying certain conditions. Among their conditions are axioms (1) - (3) above and the equality $`\tau _{\text{1}\text{ I}}=(\nu ^1)^{}\nu `$ replacing axiom (4) above. Barrett and Westbury do not suppose the existence of morphisms $`d`$ forming together with $`b`$ a duality in $`𝒞`$. However, this can be deduced from their axioms. We prefer to assume from the very beginning that $`𝒞`$ has a duality and extract from it the duality functor $``$ and the isomorphisms $`\gamma `$ and $`\nu `$. Conversely, our axioms imply all the Barrett-Westbury axioms. Only Condition (4) in \[BW2, p. 362\] is somewhat involved; the proof uses the right duality $`\stackrel{~}{b},\stackrel{~}{d}`$ defined above.
3. Spherical crossed $`\pi `$-categories. We say that a crossed $`\pi `$-category $`𝒞`$ is spherical if it is spherical in the sense of the definition above and the given system of morphisms $`\{\tau _V:VV^{}\}_{V𝒞}`$ is invariant under the action of $`\pi `$ on $`𝒞`$. A crossed $`\pi `$-category $`𝒞`$ is finite semisimple if it satisfies axioms (6.1.1) - (6.1.4). The methods of \[TV\], \[BW1\] allow to derive from a finite semisimple spherical crossed $`\pi `$-category $`𝒞`$ a topological invariant $`|M|_𝒞K`$ of a closed oriented 3-dimensional $`\pi `$-manifold $`M`$. The construction goes as follows. A local order on a triangulation $`T`$ of $`M`$ is a compatible choice for each simplex $`\mathrm{\Delta }T`$ of a total order on the set of vertices of $`\mathrm{\Delta }`$. By “compatible” we mean that the order on the vertices of any subsimplex of any $`\mathrm{\Delta }T`$ is induced by the order on the vertices of $`\mathrm{\Delta }`$. For instance, a total order on the set of all vertices of $`T`$ induces a local order on $`T`$. Note that a local order on $`T`$ determines an orientation of all edges of $`T`$. Fix now a triangulation $`T`$ of $`M`$ and endow it with a local order. Choose in the given homotopy class of maps $`MK(\pi ,1)`$ a representative, $`g`$, sending all the vertices to the base point of $`K(\pi ,1)`$. We assign to each edge $`e`$ of $`T`$ the element $`g_e\pi `$ represented by the loop $`g|_e`$. In terminology of \[Tu3, Sect. 7.2\], the function $`eg_e,e^1g_e^1`$ is a $`\pi `$-system on $`T`$ representing the homotopy class of $`g`$. For $`\alpha \pi `$, denote by $`I_\alpha `$ the set of the isomorphism classes of simple objects of $`𝒞_\alpha `$. We define $`|T|_𝒞`$ as in \[BW1\] with the difference that we involve only the labelling assigning to each edge $`e`$ of $`T`$ an element of $`I_{g(e)}`$. The rest of the construction is similar to the one in \[BW1\].
We claim that $`|M|_𝒞=|T|_𝒞`$ is a well-defined topological invariant of $`M`$, i.e., it does not depend on the choice of $`T`$, the choice of the local order on $`T`$, and the choice of $`g`$ in the given homotopy class. We outline the proof. Note first that a Pachner move on $`T`$ extends in a unique way to the $`\pi `$-systems. The extension is uniquely determined by the condition that the values of the $`\pi `$-system on all edges preserved by the move should be preserved. The same arguments as in \[BW1\] show that $`|T|_𝒞`$ does not depend of the choice of the local order on $`T`$ and is invariant under the Pachner moves. It remains to prove that $`|T|_𝒞`$ does not depend on the choice of $`g`$ in its homotopy class. Any two $`\pi `$-systems on $`T`$ representing homotopic maps $`MK(\pi ,1)`$ can be related by homotopy moves at the vertices of $`T`$, cf. \[Tu3, Sect. 7.2\]. It would be nice to give a direct proof of the invariance of $`|T|_𝒞`$ under these moves. The proof outlined below is based on the theory of skeletons of $`M`$. A skeleton of $`M`$ is a finite simple 2-polyhedron in $`M`$ whose complement is a disjoint union of open 3-balls. For instance, the closed 2-cells in $`M`$ dual to the edges of $`T`$ form a skeleton $`T^{}M`$. Using this observation, we can dualize the notion of a $`\pi `$-system from triangulations to skeletons of $`M`$. (A $`\pi `$-system on a skeleton assigns elements of $`\pi `$ to oriented 2-faces). If $`T`$ is locally ordered then we can provide the 2-cell dual to an edge $`e`$ with an orientation which determines together with the orientation of $`e`$ the given orientation of $`M`$. This makes $`T^{}`$ an oriented branched 2-polyhedron in the sense of \[BP\]. Now we observe that the state sum invariant can be defined for a $`\pi `$-system on an oriented branched skeleton of $`M`$. This allows us to switch from the language of state sums on triangulations to the language of state sums on oriented branched skeletons. In the latter language, a homotopy move on $`\pi `$-systems can be decomposed into a composition of more elementary Matveev-Piergallini moves and bubblings, cf. \[TV\] and \[Tu3, Sect. 8\]. The invariance of the state sum under these moves is verified as in \[BW1\].
4. Spherical algebras. The notion of a spherical category has its counterpart in the theory of Hopf algebras. Following \[BW2\] we call a spherical Hopf algebra over $`K`$ any pair $`(A,w)`$ where $`A`$ is a Hopf algebra over $`K`$ and $`wA`$ is a group-like element such that the square of the antipode in $`A`$ equals the conjugation by $`w`$ and for any $`A`$-linear endomorphism $`f`$ of an $`A`$-module (in the sense of Section 11.7) we have $`\text{Tr}(fw)=\text{Tr}(fw^1)`$ where Tr denotes the trace of a $`K`$-endomorphism of a projective $`K`$-module. We can extend this definition to crossed Hopf $`\pi `$-coalgebras. A spherical crossed Hopf $`\pi `$-coalgebra is a pair consisting of a crossed Hopf $`\pi `$-coalgebra $`(\{A_\alpha \}_{\alpha \pi },\mathrm{\Delta },\epsilon _1,s,\phi )`$ and invertible elements $`\{w_\alpha A_\alpha \}_{\alpha \pi }`$ satisfying the following conditions:
$$s_{\alpha ^1}s_\alpha (a)=w_\alpha aw_\alpha ^1,\mathrm{\Delta }_{\alpha ,\beta }(w_{\alpha \beta })=w_\alpha w_\beta ,s_\alpha (w_\alpha )=w_{\alpha ^1}^1,$$
$$\epsilon _1(w_1)=1,\phi _\alpha (w_\beta )=w_{\alpha \beta \alpha ^1},\text{Tr}(fw_\alpha )=\text{Tr}(fw_\alpha ^1)$$
for any $`\alpha ,\beta \pi ,aA_\alpha `$ and any $`A_\alpha `$-linear endomorphism $`f`$ of an $`A_\alpha `$-module (in the sense of Section 11.7).
It is clear that an involutory crossed Hopf $`\pi `$-coalgebra (so that $`s_{\alpha ^1}s_\alpha =\text{id}`$ for all $`\alpha \pi `$) is spherical with $`w_\alpha =1_\alpha A_\alpha `$ for all $`\alpha \pi `$. A ribbon crossed Hopf $`\pi `$-coalgebra $`(A,\theta )`$ is sperical with $`w_\alpha =\theta _\alpha u_\alpha `$ where $`u_\alpha A_\alpha `$ is the (generalized) Drinfeld element of $`R`$ (see \[Vi\]).
It is proven in \[BW2\] that for a spherical Hopf algebra $`(A,w)`$, the monoidal category $`\text{Rep}(A)`$ is spherical. For $`V\text{Rep}(A)`$, the morphism $`\tau _V:VV^{}`$ is defined as the standard identification $`V=V^{}`$ followed by multiplication by $`w^1`$. Similarly, for a spherical crossed Hopf $`\pi `$-coalgebra $`(A,w)`$, the category $`\text{Rep}(A)`$ is spherical: for $`V\text{Rep}(A_\alpha )`$, the morphism $`\tau _V:VV^{}`$ is defined as the standard identification $`V=V^{}`$ followed by multiplication by $`w_\alpha ^1`$.
Appendix 3. Open problems
1. Classification of crossed $`\pi `$-algebras. Classify crossed $`\pi `$-algebras for sufficiently simple groups $`\pi `$, say, cyclic, abelian, finite, etc. Find interesting examples of non-semisimple crossed $`\pi `$-algebras.
2. Relations between various approaches. In the theory of quantum invariants of 3-dimensional manifolds, there is a narrow relationship between the surgery approach and the state sum approach, see \[Tu2\], \[Ro\]. It would be interesting to generalize it to HQFT’s. A related question: generalize the state sums considered in Appendix 2 to shadows, cf. \[Tu2\].
3. Modular $`\pi `$-categories. Can one systematically produce interesting modular $`\pi `$-categories ? By “interesting” I mean those yielding non-trivial topological invariants allowing to distinguish 3-dimensional $`\pi `$-manifolds. In particular, if $`\pi `$ is a subgroup of a semisimple Lie group $`G`$ with Lie algebra $`𝕘`$ and $`q`$ is a complex root of unity, can one use the quantum group $`U_q(𝕘)`$ to construct modular crossed Hopf $`\pi `$-algebras ?
4. Invariants of spin-structures. Instead of maps from a manifold $`M`$ to a fixed target space one can consider maps whose source is the oriented frame bundle of $`M`$ (the principal bundle of positively oriented bases in the tangent spaces of points). When the target space is $`K(\pi ,1)`$, this should lead to new algebraic notions generalizing $`\pi `$-algebras and crossed $`\pi `$-categories. This theory should include the quantum invariants of spin-structures on closed oriented 3-manifolds, see \[Bl\], \[KM\], \[Tu1\].
5. Factors and subfactors. It was established by A. Ocneanu that the subfactors of type $`II_1`$ give rise to 3-manifold invariants via the state sum approach (see for instance \[EK\]). What is the counterpart of the theory introduced above in the setting of subfactors ?
6. Perturbative aspects. What are the perturbative aspects of HQFT’s ? Are there perturbative invariants of 3-dimensional $`\pi `$-manifolds generalizing the Le-Murakami-Ohtsuki invariant \[LMO\] ?
7. Higher-dimensional generalizations. The quantum invariants of 3-manifolds have 4-dimensional counterparts, see \[CKY\], \[CKS\], \[CKJLS\], \[Mac\] and references therein. The role of categories in these constructions is typically played by 2-categories. The homotopy quantum field theory should have similar 4-dimensional and high-dimensional versions and give rise in particular to the notion of a crossed 2-category.
In this paper we considered mainly target spaces of type $`K(\pi ,1)`$. In dimension 4 it can be easier and more appropriate to consider target spaces of type $`K(H,2)`$ where $`H`$ is an abelian group. Note that 2-dimensional HQFT’s with such target space were studied in \[BT\].
It is most interesting to analyze the algebraic data yielding invariants of $`\text{Spin}^c`$-structures on 4-manifolds. Here as the sources of maps one takes the oriented frame bundles of 4-manifolds and as the target space one takes $`K(H,2)`$. Can one include the Seiberg-Witten invariants in this framework ?
8. Miscellaneous questions. Study the representations of subgroups of the mapping class groups resulting from HQFT’s. Find an interpretation of HQFT’s in terms of algebraic geometry, i.e., in terms of sections of bundles over moduli spaces. Study the 2-dimensional homotopy modular functors and their relations to quantum computations (cf. \[FKW\]). Study relations with number theory, cf. \[LZ\].
References
\[BW1\] Barrett, J., Westbury, B., Invariants of piecewise-linear 3-manifolds. Trans. Amer. Math. Soc. 348 (1996), 3997–4022.
\[BW2\] Barrett, J., Westbury, B., Spherical categories. Adv. Math. 143 (1999), 357–375.
\[BP\] Benedetti, R., Petronio, C., Branched Standard Spines of $`3`$-manifolds. Lect. Notes in Math., 1653. Springer-Verlag, Berlin, 1997.
\[Bl\] Blanchet, C., Invariants of three-manifolds with spin structure. Comm. Math. Helv. 67 (1992), 406–427.
\[Br\] Brieskorn, E., Automorphic sets and singularities. Contemp. Math. 78 (1988), 45–115.
\[BT\] Brightwell, M. Turner, P., Representations of the homotopy surface category of a simply connected space. Preprint math. AT/9910026.
\[Bru\] Bruguières, A., Catégories prémodulaires, modularisations et invariants des variétés de dimension 3. Math. Ann. 316 (2000), 215–236.
\[CKS\] Carter, J., Kauffman, L., Saito, M., Structures and Diagrammatics of Four Dimensional Topological Lattice Field Theories. Adv. Math. 146 (1999), no. 1, 39–100.
\[CKJLS\] Carter, J., Kamada, S., Jelsovsky, D., Langford, L., Saito, M., State-sum invariants of knotted curves and surfaces from quandle cohomology. Electron. Res. Announc. Amer. Math. Soc. 5 (1999), 146–156 (electronic).
\[CKY\] Crane, L., Kauffman, L., Yetter, D. State-sum invariants of 4-manifolds. J. Knot Theory Ramifications 6 (1997), 177–234.
\[EK\] \] Evans, D., Kawahigashi, Y., Quantum symmetries on operator algebras. Oxford Mathematical Monographs. Oxford Science Publications. The Clarendon Press, Oxford University Press, New York, 1998.
\[FR1\] Fenn, R., Rourke, C., On Kirby’s calculus of links. Topology 18 (1979), 1–15.
\[FR2\] Fenn, R., Rourke, C., Racks and Links in Codimension Two. J. of Knot Theory and Ramifications 1 (1992), 343–406.
\[FKW\] Freedman, M., Kitaev, A., Wang, Z., Simulation of topological field theories by quantum computers. Preprint quant-ph/0001071.
\[FY\] Freyd, P., Yetter, D., Braided compact closed categories with applications to low-dimensional topology. Adv. Math. 77 (1989), 156–182.
\[GK\] Gelfand, S., Kazhdan, D. Invariants of 3-dimensional manifolds. Geom. Funct. Anal. 6 (1996), 268–300.
\[He\] Hennings, M., Invariants of links and $`3`$-manifolds obtained from Hopf algebras. J. London Math. Soc. (2) 54 (1996), 594–624.
\[KRT\] Kassel, C., Rosso, M. , Turaev, V., Quantum Groups and Knot Invariants, Panoramas et Synthèses 5, Société Mathématique de France, 115 p., 1997.
\[Ki\] Kirby, R., A calculus of framed links in $`S^3`$. Invent. Math. 45 (1978), 35–56.
\[KM\] Kirby, R., Melvin, P., On the 3-manifold invariants of Witten and Reshetikhin-Turaev for $`sl_2()`$. Invent. Math. 105 (1991), 473–545.
\[Ku\] Kuperberg, G., Involutory Hopf algebras and $`3`$-manifold invariants. Internat. J. Math. 2 (1991), 41–66.
\[LZ\] Lawrence, R., Zagier, D., Modular forms and quantum invariants of $`3`$-manifolds. Asian J. Math. 3 (1999), 93–107.
\[LMO\] Le, T., Murakami, J., Ohtsuki, T., On a universal perturbative invariant of $`3`$-manifolds. Topology 37 (1998), 539–574.
\[LT\] Le, T., Turaev, V., Quantum groups and abelian HQFT’s, in preparation.
\[Mac\] Mackaay, M., Spherical $`2`$-categories and $`4`$-manifold invariants. Adv. Math. 143 (1999), 288–348.
\[Ma\] MacLane, S., Categories for the working mathematician. Graduate Texts in Math. 5. Springer-Verlag, New York-Heidelberg-Berlin 1971.
\[Oh1\] Ohtsuki, T., Colored ribbon Hopf algebras and universal invariants of framed links. J. of Knot Theory and Ramifications 2 (1993), 211–232.
\[Oh2\] Ohtsuki, T., Invariants of $`3`$-manifolds derived from universal invariants of framed links. Math. Proc. Cambridge Philos. Soc. 117 (1995), 259–273.
\[Os\] Ospel, C., Tressages et théories cohomologiques pour les algèbres de Hopf. Application aux invariants des 3-variététs. Ph. D. thesis (Strasbourg, 1999).
\[RT\] Reshetikhin, N., Turaev, V., Invariants of 3-manifolds via link polynomials and quantum groups. Invent. Math. 103 (1991), 547–598.
\[Ro\] Roberts, J., Skein theory and Turaev-Viro invariants. Topology 34 (1995), 771–787.
\[Th\] Thys, H,. Description topologique des représentations de $`U_q(sl_2)`$, Preprint math. QA/9810088, to appear in Ann. Fac. Sci. Toulouse.
\[Tu1\] Turaev, V., State sum models in low-dimensional topology. Proc. ICM, Kyoto, 1990 vol. 1 (1991), 689–698.
\[Tu2\] Turaev, V., Quantum invariants of knots and 3-manifolds. Studies in Mathematics 18, Walter de Gruyter, 588 p., 1994.
\[Tu3\] Turaev, V., Homotopy field theory in dimension 2 and group-algebras. Preprint QA/9910010.
\[TV\] Turaev, V., Viro, O., State sum invariants of 3-manifolds and quantum $`6j`$-symbols. Topology 31 (1992), 865–902.
\[Vi\] Virelizier, A., Crossed Hopf $`\pi `$-coalgebras and invariants of links and 3-manifolds, in preparation.
\[Zu\] Zunino, M. Center of $`\pi `$-categories and double of crossed Hopf $`\pi `$-coalgebras, in preparation.
Institut de Recherche Mathématique Avancée, Université Louis Pasteur - CNRS, 7 rue René Descartes, 67084 Strasbourg Cedex, France
turaevmath.u-strasbg.fr
|
warning/0005/quant-ph0005093.html
|
ar5iv
|
text
|
# Anisotropic Vacuum Induced Interference in Decay Channels
## Abstract
We demonstrate how the anisotropy of the vacuum of the electromagnetic field can lead to quantum interferences among the decay channels of close lying states. Our key result is that interferences are given by the scalar formed from the antinormally ordered electric field correlation tensor for the anisotropic vacuum and the dipole matrix elements for the two transitions. We present results for emission between two conducting plates as well as for a two photon process involving fluorescence produced under coherent cw excitation.
PACS. No.: 42.50 Ct., 42.50 Gy., 32.50 +d.
Recently considerable effort has been devoted to the question of coherence and interference effects arising from the decay of close lying energy levels . Such interference effects lead to many remarkable phenomena such as population trapping , spectral narrowing , gain without inversion , phase dependent line shapes and quantum beats . However the very existence of the interference effect depends on the validity of a very stringent condition viz. the dipole matrix moments for two close lying states decaying to a common final state should be non-orthogonal . This last condition is really the bottleneck in the observation of the predicted new effects. Some progress however was made by the use of static and electromagnetic fields to mix the levels so that the relevant dipole moments become non-orthogonal. In this letter, we propose a totally different mechanism to overcome the problem of the orthogonality of the dipole moments. We suggest working in such situations where the vacuum of the electromagnetic field is anisotropic, so that the interference among decay channels can occur even if the corresponding dipole moments are orthogonal. This provides a possible solution to the long standing problem in the subject of interference among decay channels. Our key result is that interferences are given by the scalar formed from the antinormally ordered electric field correlation tensor for the anisotropic vacuum and the dipole matrix elements for the two transitions. This is in contrast to the usual result that interferences are given in terms of the scalar formed out of the dipole matrix elements. This opens up the possibility of studying quantum interferences in a variety of new class of systems. At the outset, we mention that one can consider many situations where vacuum will be anisotropic, for example (a) doped active centers in anisotropic glasses , (b) emission of active atoms in a waveguide , (c) spontaneous emission from atoms adsorbed on metallic or dielectric surfaces , (d) emission in a spatially dispersive medium - which allows the possibility of longitudinal electromagnetic fields, (e) emission between two conducting plates , which is a problem of great interest since the early work of Casimir. Our results suggest the study of quantum interferences in a totally new class of situations involving atoms and molecules adsorbed on surfaces. Explicit results in some of these situations would be given later.
Interferences in Fluorescence under Coherent Excitation -Two Photon Processes
Before we present detailed dynamical equations, we consider a simple situation which enables us to bring out the essential physics of the interferences associated with anisotropic vacuum fields. We basically examine the nature of interferences in spontaneous emission. However here the way the system is excited is important. A practical way will be to excite by a coherent cw radiation. In technical terms this is a two-photon or a second order process. This is in contrast to those studies which ignore how the system was excited. Note that in the experiment of Xia et al. the fluorescence produced by two photon excitation was studied. Let us consider the process\[Fig. 1\] in which the atom in the state $`|g`$ absorbs a photon of frequency $`\omega _l`$ and wavevector $`\stackrel{}{k_l}`$ and emits a photon to end up in the state $`|f`$ which is distinct from the ground state $`|g`$. In the process of absorption and emission, the atom goes through a number of virtual intermediate states. For the purpose of the argument, we retain only two intermediate states $`|j(j=1,2)`$. The transition probability for this process can be calculated using the second order perturbation theory (cf. ref. ). The initial state of the field is vacuum $`|v`$. For simplicity we assume that the absorption is from a coherent field $`_l`$. Let $`H_I(t)`$ be the interaction Hamiltonian in the interaction picture. As usual we assume that the perturbation is switched on slowly. Then the transition probability of the above second order process can be written as
$$T_{gf}=\underset{t\mathrm{}}{lim}\frac{d}{dt}\underset{F}{}\left|\frac{1}{\mathrm{}^2}_{\mathrm{}}^t𝑑t_1_{\mathrm{}}^{t_1}𝑑t_2f,F\left|H_I(t_1)H_I(t_2)e^{ϵ(t_1+t_2)}\right|g,v\right|^2.$$
(1)
We sum over all final states $`|F`$ of the field i.e. we assume that no spectral measurement of the emitted field is done. The interaction Hamiltonian can be written as
$`H_I(t)`$ $`=`$ $`\stackrel{}{d}(t).\stackrel{}{}_le^{i\omega _lt+i\stackrel{}{k}_l.\stackrel{}{r}}`$ (2)
$``$ $`\stackrel{}{d}(t).\stackrel{}{E}_v(t)+\mathrm{H}.\mathrm{C}.,`$ (3)
where $`\stackrel{}{E}_v`$ is the electric field operator for the vacuum and $`\stackrel{}{d}`$ is the dipole moment operator for the atom. On substituting (2) in (1) and on carrying out all the simplifications and on making rotating wave approximation we find the expression for the transition probability
$$T_{gf}=\frac{1}{\mathrm{}^2}\underset{i,j}{}\frac{g_ig_j^{}\stackrel{}{d}_{fj}^{}.\stackrel{}{C}(\omega _l\omega _{fg}).\stackrel{}{d}_{fi}}{(\omega _{ig}\omega _l)(\omega _{jg}\omega _l)},g_i=\frac{\stackrel{}{d}_{ig}.\stackrel{}{}_l}{\mathrm{}}.$$
(4)
Here we have introduced the correlation function tensor $`\stackrel{}{C}`$ for the anti-normally ordered correlation function for the vacuum field
$$\stackrel{}{C}(\omega )=_{\mathrm{}}^+\mathrm{}𝑑te^{i\omega t}\stackrel{}{E}_v^{(+)}(t)\stackrel{}{E}_v^{()}(0),\stackrel{}{E}_v(t)=\stackrel{}{E}_v^{(+)}(t)+\stackrel{}{E}_v^{()}(t),$$
(5)
where $`\stackrel{}{E}_v^{(+)}`$ and $`\stackrel{}{E}_v^{()}`$ are respectively the positive and negative frequency parts of the field operator representing anisotropic vacuum. The two field operators in (4) are to be evaluated at the position of the atom. The expression (3) displays explicitly the atomic and the vacuum field characteristics. The anisotropic vacuum enters through the correlation tensor $`\stackrel{}{C}.`$ The terms $`ij`$ in (3) correspond to the interferences between the decay channels $`|i|f`$ and $`|j|f`$. The quantum interferences will be non-vanishing only if
$$\stackrel{}{d}_{fj}^{}.\stackrel{}{C}(\omega _l\omega _{fg}).\stackrel{}{d}_{fi}0.$$
(6)
This is one of the key results of this paper. For isotropic vacuum the correlation tensor $`\stackrel{}{C}`$ is proportional to the unit tensor: $`\stackrel{}{C}=\stackrel{}{I}C`$ and hence (5) reduces to
$$\stackrel{}{d}_{fj}^{}.\stackrel{}{d}_{fi}0.$$
(7)
Clearly the interferences will survive even if the corresponding dipole matrix elements are orthogonal provided that the vacuum field is anisotropic. Note further that with a proper tuning of the field the amplitude $`T_{gf}`$ can, in principle, become zero. It may be noted that the correlation functions $`\stackrel{}{C}`$ are known in the literature for a variety of situations including the ones mentioned in the introduction.
Anisotropy Induced Interference in Dynamical Evolution
Let us next consider the dynamical evolution of the atomic density matrix so that we can study various line shapes and other dynamical aspects of emission. For simplicity we consider a $`j=1`$ to $`j=0`$ transition. Let a static magnetic field be applied along $`y`$ direction. This defines the quantization axis. The magnetic sub-level $`|1|j=1,m=1`$ with energy $`\mathrm{}\omega _1`$ ($`|2|j=1,m=1`$ with energy $`\mathrm{}\omega _2`$) decays to the state $`|3=|j=0,m=0`$ (energy = 0) with the emission of a right (left) circularly polarized photon. We drop the level $`j=1,m=0`$ from our consideration as it does not participate in interferences. The Hamiltonian for interaction of the atom with the vacuum is
$`H_1=`$ $`+`$ $`d|13|\widehat{ϵ}_{}.\stackrel{}{E}_v(t)`$ (8)
$``$ $`d|23|\widehat{ϵ}_+.\stackrel{}{E}_v(t)+\mathrm{H}.\mathrm{C}.,`$ (9)
where $`\widehat{ϵ}_\pm =(\widehat{z}\pm i\widehat{x})/\sqrt{2}`$ and $`d`$ is the reduced dipole matrix element. In order to describe the dynamics of the atom, we use the master equation framework. We use the Born and Markov approximations to derive the master equation for the atomic density matrix $`\rho `$. In rotating wave approximation, our calculations lead to the equation
$`{\displaystyle \frac{\rho }{t}}=`$ $``$ $`i(\omega _1|11|+\omega _2|22|)\rho `$ (10)
$``$ $`\gamma _1(|11|\rho \rho _{11}|33|)`$ (11)
$``$ $`\gamma _2(|22|\rho \rho _{22}|33|)`$ (12)
$`+`$ $`\kappa _2(|12|\rho \rho _{21}|33|)`$ (13)
$`+`$ $`\kappa _1(|21|\rho \rho _{12}|33|)+\mathrm{H}.\mathrm{C}..`$ (14)
Here the coefficients $`\gamma ^{}s`$ and $`\kappa ^{}s`$ are related to the antinormally ordered correlation functions of the vacuum field
$$\stackrel{(+)}{\stackrel{}{C}}(\omega )=_0^{\mathrm{}}\stackrel{}{E}_v^{(+)}(\tau )\stackrel{}{E}_v^{()}(0)e^{i\omega \tau }𝑑\tau ,$$
(15)
$$\gamma _1=\frac{d^2}{\mathrm{}^2}\widehat{\epsilon }_{}.\stackrel{}{C}(\omega _1).\widehat{\epsilon }_+;\gamma _2=\frac{d^2}{\mathrm{}^2}\widehat{\epsilon }_+.\stackrel{}{C}(\omega _2).\widehat{\epsilon }_{},$$
(16)
$$\kappa _2=\frac{d^2}{\mathrm{}^2}\widehat{\epsilon }_{}.\stackrel{}{C}(\omega _2).\widehat{\epsilon }_{};\kappa _1=\frac{d^2}{\mathrm{}^2}\widehat{\epsilon }_+.\stackrel{}{C}(\omega _1).\widehat{\epsilon }_+.$$
(17)
Note that the terms involving $`\kappa _1`$ and $`\kappa _2`$ are responsible for interferences between the two decay channels $`|1|3`$ and $`|2|3`$. For the case of free space, vacuum is isotropic
$$C_{zz}^{(+)}(\omega )=C_{xx}^{(+)}(\omega ),C_{zx}^{(+)}(\omega )=C_{xz}^{(+)}(\omega )=0,$$
(18)
and hence
$$\kappa _1=\kappa _2=0,$$
(19)
leading to no interferences in the decay channels. Clearly, for decay in free space the interferences could be possible only if the dipole matrix elements were non-orthogonal: $`\stackrel{}{d}_{13}.\stackrel{}{d}_{23}^{}0`$. Our development of the master equation shows how the interferences in the decay channels are possible even if the dipole matrix elements were orthogonal. We need the anisotropy of the vacuum. The anisotropy leads to the non-vanishing of the coefficients $`\kappa _1`$ and $`\kappa _2`$. The interferences are particularly prominent when $`\kappa `$’s become comparable to $`\gamma `$’s. Thus for our situation we will recover all previous results on line shapes and trapping.
We could now consider explicitly situations of the type mentioned in the introduction. The correlation functions $`\stackrel{}{C}`$ can be computed for example in situations corresponding to the emission from an atom in a metallic waveguide or an atom between the plates of a perfect conductor. On using the relation
$$\frac{1}{\epsilon +i(\omega _0\omega )}=P\frac{1}{i(\omega _0\omega )}+\pi \delta (\omega _0\omega ),$$
(20)
and on ignoring the principal value terms in Eq.(9), we can approximate $`\stackrel{(+)}{\stackrel{}{C}}`$ as
$$\stackrel{(+)}{\stackrel{}{C}}(\omega )\frac{1}{2}\stackrel{}{C}(\omega ).$$
(21)
The correlation function for the vacuum field can be calculated by quantizing the field and by using the properties of annihilation and creation operators. However in certain situations the explicit quantization of the field is complicated and hence we follow a different method. Using the linear response theory the correlation function $`\stackrel{}{C}(\omega )`$ can be related to the solution $`\stackrel{}{E}(\stackrel{}{r},t)=\stackrel{}{}(\overline{r},\omega )e^{i\omega t}+\mathrm{c}.\mathrm{c}.`$ of Maxwell’s equations with a source polarization $`\stackrel{}{P}\stackrel{}{p}\delta (\stackrel{}{r}\stackrel{}{r_0})e^{i\omega t}+\mathrm{c}.\mathrm{c}.`$
$$C_{\alpha \beta }(\omega )2\mathrm{}\mathrm{Im}(_\alpha (r_0,\omega )/p_\beta );\omega >0.$$
(22)
Note that the dynamical equation (8) can be used to calculate all the line shapes (both absorption and emission) for an anisotropic vacuum by using Eqs. (10), (11), (15) and (16). Note that the quantity in the bracket in Eq.(16) is the Green’s tensor for the Maxwell equations.Thus the procedure for a given geometrical arrangement will consist of evaluation of the Green’s tensor and then the application of (16) to obtain the correlation tensor.
Interferences in Emission between two Conducting Plates.
Let us now consider an important problem in cavity QED viz the emission from an atom located between two conducting plates \[Fig.2\] at $`z=0`$ and $`z=d`$. The atom is located at $`z=b`$. The C’s as defined by (4) can be calculated using (16). These calculations are extremely long. We will only quote the final result. For this geometry $`C_{zx},C_{xz}=0`$. Furthermore the parameters $`\gamma ^{}s`$ and $`\kappa ^{}s`$ entering the master equation (8) can be shown to be
$$\gamma _i=\gamma _i^{(0)}\left[\mathrm{\Gamma }_{}(\omega _i)+\mathrm{\Gamma }_{||}(\omega _i)\right]/2,$$
(23)
$$\kappa _i=\gamma _i^{(0)}\left[\mathrm{\Gamma }_{}(\omega _i)\mathrm{\Gamma }_{||}(\omega _i)\right]/2,\gamma _i^{(0)}\frac{2\omega _i^3d^2}{3c^3\mathrm{}},$$
(24)
where
$$\mathrm{\Gamma }_{}(\omega )=\frac{3\pi }{2kd}+\frac{3\pi }{kd}\underset{n=1}{\overset{𝒩}{}}\left(1\frac{\pi ^2n^2}{k^2d^2}\right)\mathrm{cos}^2\left(\frac{b\pi n}{d}\right),$$
(25)
$$\mathrm{\Gamma }_{||}(\omega )=\frac{3\pi }{2kd}\underset{n=1}{\overset{𝒩}{}}\left(1+\frac{\pi ^2n^2}{k^2d^2}\right)\mathrm{sin}^2\left(\frac{b\pi n}{d}\right),k=\frac{\omega }{c},$$
(26)
and where $`𝒩`$ is the largest integer smaller than $`kd/\pi `$. The explicit results (17) and (18) show the expected interferences between the decay channels. Clearly the interferences will be sensitive to the magnitude of the magnetic field which enters through $`\omega `$. The case of small magnetic fields is especially interesting. We further note that below the cut off frequency $`kd/\pi <1,\mathrm{\Gamma }_{||}(\omega )0,\mathrm{\Gamma }_{}(\omega )3\pi /2kd`$ and $`\kappa _i=\gamma _i`$. In this limit quantum interferences become especially prominent. We further note that (a) in the limit $`d\mathrm{}`$, we get results for emission in the presence of a single conducting plate, (b) the quantities $`\mathrm{\Gamma }_{||}`$ and $`\mathrm{\Gamma }_{}`$ are related to the emission from a single two level atom between the two conducting plates.
In conclusion we have demonstrated how the anisotropy of the vacuum field can lead to new types of interference effects between the decay of close lying states. We have related the interference terms to the antinormally ordered correlation tensor of the vacuum. The anisotropy related interferences are especially significant for emission from atoms, molecules adsorbed on surfaces and thus our study opens up the possibility of studying quantum interferences in a totally new class of systems. We have given explicit example of decay between two conducting plates. We have also shown the role of interference effects in two photon processes where fluorescence is detected following excitation by a coherent cw field. Clearly the anisotropy related interference effects would also be important in considerations of higher order radiative processes which could be studied in the same manner as two photon processes.
|
warning/0005/math0005093.html
|
ar5iv
|
text
|
# Large characteristic subgroups of surface groups not containing any simple loops
## 1 Introduction
For any compact orientable surface $`S`$ we determine the smallest characteristic non-geometric quotient of $`\pi _1(S)`$. Non-geometric means that no non-trivial element that can be represented by a simple closed curve is mapped to the identity and characteristic means that its kernel is kept fixed under all automorphisms. We write $`\pi `$, $`H`$ and $`g`$ for $`\pi _1(S)`$, $`H_1(S,𝐙)`$ and the genus of $`S`$ respectively. We assume that $`g`$ is at least 2. Consider the following characteristic subgroups of $`\pi `$ : $`\pi ^{[1]}:=\pi `$ and inductively $`\pi ^{[k+1]}:=[\pi ,\pi ^{[k]}]`$. We have a well known isomorphism $`\pi ^{[2]}/\pi ^{[3]}^2H/\omega `$, $`[x,y]xy`$, where $`\omega `$ is the intersection form on $`H`$. We have an intersection product $`^2H𝐙`$. Let $`K`$ be the kernel of the composition of the map $`\pi ^{[2]}^2H/\omega `$ with the intersection product $`^2H/\omega 𝐙/g𝐙`$. Our result is
###### Proposition 1.1
If $`g`$ is odd, then the largest (i.e. smallest index) characteristic non-geometric subgroup of $`\pi `$ is given by all $`g`$th powers and $`K`$. If $`g`$ is even, then the largest characteristic non-geometric subgroup of $`\pi `$ is given by all $`2g`$th powers and $`K`$. The indices of these are $`g^{2g+1}`$ and $`(2g)^{2g}g`$ respectively.
Recently, Livingston proved that for $`g=2`$ the smallest non-geometric quotient of $`\pi `$ is a group of order $`2^5`$, and he raised the question whether an easy generalisation of his result (which yields a group of order $`g^{2g+1}`$), holds true for any genus. The theorem above shows that his generalisation for odd genus is in any case the smallest non-geometric characteristic quotient.
## 2 Notation and preliminary computations
Before we start, we pick once and for all a set of generators for $`\pi `$: $`x_1,\mathrm{},x_{2g}`$ with defining relation $`\mathrm{\Pi }_{i=1}^g[x_{2i1},x_{2i}]`$. We call a pair of integers $`(i,j)`$ related if and only if there exists an integer $`h`$ with $`1hg1`$ such that $`(i,j)=(2h1,2h)`$. We write $`[a,b]`$ for $`a^1b^1ab`$, such that $`ab=ba[a,b]`$. Furthermore, if $`\alpha `$ and $`\beta `$ are elements of $`\pi `$, then $`\alpha \beta `$ means: perform $`\alpha `$ first, then $`\beta `$.
Consider the following normal subgroups of $`\pi `$:
$$\begin{array}{ccc}\hfill \pi ^{[1]}& :=& \pi ;\hfill \\ \hfill \pi ^{[k+1]}& :=& [\pi ,\pi ^{[k]}];\hfill \\ \hfill K& :=& Ker(\pi ^{[2]}𝐙/g𝐙);\hfill \\ \hfill \pi ^n& :=& <x^n|x\pi >.\hfill \end{array}$$
Notice that $`K`$ contains $`\pi ^{[3]}`$. All these are clearly characteristic subgroups. For $`K`$ this holds since $`\pi ^{[2]}`$ is characteristic and all involved maps are natural. The quotient we are interested in, is $`\pi /\pi ^gK`$ respectively $`\pi /\pi ^{2g}K`$. In order to prove the proposition, we first describe $`\pi /\pi ^{[3]}`$.
###### Lemma 2.1
$`\pi /\pi ^{[3]}\{\mathrm{\Pi }_{i=1}^{2g}x_i^{n_i}\mathrm{\Pi }_{1i<j2g}^{}[x_i,x_j]^{m_{i,j}}|n_i,m_{i,j}𝐙,meaning:(i,j)(2g1,2g)\},`$
where multiplication on the right hand side is defined as follows:
$$\begin{array}{c}(\mathrm{\Pi }_{i=1}^{2g}x_i^{n_i}\mathrm{\Pi }_{1i<j2g}^{}[x_i,x_j]^{m_{i,j}})(\mathrm{\Pi }_{i=1}^{2g}x_i^{k_i}\mathrm{\Pi }_{1i<j2g}^{}[x_i,x_j]^{l_{i,j}})\\ \\ =\mathrm{\Pi }_{i=1}^{2g}x_i^{n_i+k_i}\mathrm{\Pi }_{1i<j2g}^{}[x_i,x_j]^{m_{i,j}+l_{i,j}k_in_j+\stackrel{~}{\delta }_{i+1,j}k_{2g1}n_{2g}}.\end{array}$$
Here $`\stackrel{~}{\delta }_{a,b}=1`$ if $`a=b`$ and $`a`$ is even, else $`\stackrel{~}{\delta }_{a,b}=0`$.
###### Proof 2.2
We have the exact sequence
$$1\pi ^{[2]}/\pi ^{[3]}\pi /\pi ^{[3]}\pi /\pi ^{[2]}1,$$
where the outer factors are finitely generated free abelian groups (\[La, Th.5.12\]) of rank $`(_2^{2g})1`$ and $`2g`$ respectively. Explicitly, we have:
$$\begin{array}{c}\pi /\pi ^{[2]}=x_i_{ab},i=1,\mathrm{},2g,\hfill \\ \pi ^{[2]}/\pi ^{[3]}=[x_i,x_j]_{ab},1i<j2g,(i,j)(2g1,2g).\hfill \end{array}$$
Here the subscript ab means the free abelian group generated by these elements, Now recall that if $`a\pi ^{[k]}`$ and $`b\pi ^{[l]}`$ then $`ab=ba[a,b]`$ with $`[a,b]\pi ^{(k+l)}`$, so that
$`abba\text{ mod }\pi ^{[k+l]}.`$ (1)
Furthermore we have the following identities, modulo $`\pi ^{[3]}`$: (\[MKS, Th.5.1\])
$`[a,b]`$ $`=`$ $`[b,a]^1,`$ (2)
$`[a,bc]`$ $`=`$ $`[a,c][a,b],`$ (3)
$`[ab,c]`$ $`=`$ $`[a,c][b,c].`$ (4)
It is clear now that any element of $`\pi /\pi ^{[3]}`$ can be written uniquely in the form $`\mathrm{\Pi }_{i=1}^{2g}x_i^{n_i}\mathrm{\Pi }_{1i<j2g}^{}[x_i,x_j]^{m_{i,j}}`$ where $`n_i,m_{i,j}𝐙`$ and the meaning of the * is: $`(i,j)(2g1,2g)`$. Note that the last $`(_2^{2g})1`$ factors commute by (1). Before we start the computation, notice that modulo $`\pi ^{[3]}`$ we have
$`[a^i,b^j][a,b]^{ij},`$ (5)
as one proves easily by induction. The following identities hold in the group $`\pi /\pi ^{[3]}`$:
$$\begin{array}{cc}& (x_1^{n_1}\mathrm{}x_{2g}^{n_{2g}})(x_1^{k_1}\mathrm{}x_{2g}^{k_{2g}})\hfill \\ =& x_1^{n_1+k_1}\mathrm{}x_{2g}^{n_{2g}}x_2^{k_2}\mathrm{}x_{2g}^{k_{2g}}[x_1,x_2]^{k_1n_2}\mathrm{}[x_1,x_{2g}]^{k_1n_{2g}}\hfill \\ =& \mathrm{}\hfill \\ =& \mathrm{\Pi }x_i^{n_i+k_i}\mathrm{\Pi }_{1i<j2g}[x_i,x_j]^{k_in_j}\hfill \\ =& \mathrm{\Pi }x_i^{n_i+k_i}\mathrm{\Pi }_{1i<j2g}^{}[x_i,x_j]^{k_in_j+\stackrel{~}{\delta }_{i+1,j}k_{2g1}n_{2g}}\hfill \end{array}$$
with $`\stackrel{~}{\delta }_{i+1,j}k_{2g1}n_{2g}`$ as above. Combining these proves the lemma. (Cf.\[PdJ, Lemma 6.1\])
Here the term with $`\stackrel{~}{\delta }_{i+1,j}k_{2g1}n_{2g}`$ stems from the defining relation for the group $`\pi :\mathrm{\Pi }_{i=1}^{2g}[x_{2i1},x_{2i}]=1`$, which we used to get rid of $`[x_{2g1},x_{2g}]`$ as a generator for $`\pi ^{[2]}/\pi ^{[3]}`$.
Clearly the subgroup $`\pi ^{[2]}/\pi ^{[3]}`$ is given in terms of these generators by all expressions of the form $`\mathrm{\Pi }_{1i<j2g}^{}[x_i,x_j]^{m_{i,j}}`$.
###### Lemma 2.3
$`\pi /K\{\mathrm{\Pi }_{i=1}^{2g}x_i^{n_i}[x_1,x_2]^m|n_i𝐙,m𝐙/g𝐙\},`$ where multiplication on the right hand side is defined as follows:
$$\begin{array}{c}(\mathrm{\Pi }_{i=1}^{2g}x_i^{n_i}[x_1,x_2]^m)(\mathrm{\Pi }_{i=1}^{2g}x_i^{k_i}[x_1,x_2]^l)\\ \\ =\mathrm{\Pi }_{i=1}^{2g}x_i^{n_i+k_i}[x_1,x_2]^{m+l\mathrm{\Sigma }_{i=1}^{g1}k_{2i1}n_{2i}+(g1)k_{2g1}n_{2g}}.\end{array}$$
###### Proof 2.4
Clearly, $`K/\pi ^{[3]}`$ is generated by the elements $`[x_i,x_j]`$ for $`i`$ and $`j`$ not related, by the elements $`[x_{2k1},x_{2k}][x_{2l1},x_{2l}]^1`$ and by $`\mathrm{\Pi }_{i=1}^g[x_{2i1},x_{2i}]`$.
###### Proposition 2.5
If $`g`$ is odd, then
$$\pi /\pi ^gK\{\mathrm{\Pi }_{i=1}^{2g}x_i^{n_i}[x_1,x_2]^m|n_i,m𝐙/g𝐙,\},$$
and if $`g`$ is even, say $`g=2h`$, then
$$\pi /\pi ^gK\{\mathrm{\Pi }_{i=1}^{2g}x_i^{n_i}[x_1,x_2]^m|n_i𝐙/g𝐙,m𝐙/h𝐙\},$$
where multiplication on the right hand sides is defined as above.
###### Proof 2.6
Consider the short exact sequence
$$1\pi ^{[2]}/(\pi ^{[2]}\pi ^gK)\pi /\pi ^gK\pi /\pi ^g\pi ^{[2]}1.$$
Clearly, $`\pi /\pi ^g\pi ^{[2]}H_1(S,𝐙/g𝐙)`$. Furthermore, if $`g`$ is odd, $`\pi ^{[2]}/(\pi ^{[2]}\pi ^gK)(𝐙/g𝐙)`$, whereas in case $`g`$ is even, say $`g=2h`$, we have $`\pi ^{[2]}/(\pi ^{[2]}\pi ^gK)(𝐙/h𝐙)`$ . This follows directly from \[PdJ, Lemma 6.3\].
###### Corollary 2.7
The subgroup $`\pi ^g`$ is generated by all $`g`$th powers of geometric (i.e. representable by a simple closed curve) elements modulo $`K`$.
###### Proof 2.8
Clearly the elements $`x_i^g`$ and $`[x_1,x_2]^g`$ are $`g`$th powers of geometric elements, settling the statement for odd $`g`$. For even $`g`$, say $`g=2h`$, we have $`(x_1x_2)^{2h}x_1^{2h}x_2^{2h}[x_1,x_2]^{(2h1)h}`$ modulo $`\pi ^{[3]}`$, as one proves easily by induction (cf. \[PdJ, Lemma 6.3\]). Therefore, $`[x_1,x_2]^hx_2^{2h}x_1^{2h}(x_1x_2)^{2h}[x_1,x_2]^{2h}`$, proving the corollary.
We define the following simple loops on $`S`$: $`\gamma _1=x_1`$, $`\gamma _{2h}=x_{2h}`$, for $`h=1,\mathrm{},g`$, $`\gamma _{2h1}=x_{2h1}[x_{2h3},x_{2h2}]^1x_{2h3}^1`$, for $`h=2,\mathrm{},g`$,and finally $`\gamma _{2g+1}`$ $`=[x_{2g1},x_{2g}]^1x_{2g1}^1.`$
We write $`\tau _i`$ respectively $`\sigma _i`$ for (the right handed) Dehn twist around $`\gamma _i`$ and $`x_{2i1}`$ respectively. For later convenience we list the action of these Dehn twists on the generators of $`\pi `$ and the action modulo $`\pi ^{[3]}`$ as above (if the action of some Dehn twist on a generator is not given, it is the trivial action):
$$\begin{array}{ccc}\hfill \tau _1(x_2)& =& x_1^1x_2,\hfill \\ \hfill \tau _{2h}(x_{2h1})& =& x_{2h}x_{2h1}x_{2h1}x_{2h}[x_{2h1},x_{2h}]^1,\hfill \\ \hfill \tau _{2h1}(x_{2h2})& =& x_{2h2}\gamma _{2h1}=x_{2h2}x_{2h1}[x_{2h3},x_{2h2}]^1x_{2h3}^1\hfill \\ & & x_{2h3}^1x_{2h2}x_{2h1}[x_{2h3},x_{2h2}]^2[x_{2h2},x_{2h1}]^1\hfill \\ \hfill \tau _{2h1}(x_{2h1})& =& \gamma _{2h1}^1x_{2h1}\gamma _{2h1}=x_{2h1}[x_{2h1},\gamma _{2h1}]\hfill \\ & =& x_{2h1}[x_{2h1},x_{2h1}[x_{2h3},x_{2h2}]^1x_{2h3}^1]\hfill \\ & & x_{2h1}[x_{2h3},x_{2h1}]\hfill \\ \hfill \tau _{2h1}(x_{2h})& =& \gamma _{2h1}^1x_{2h}\hfill \\ & =& x_{2h3}[x_{2h3},x_{2h2}]x_{2h1}^1x_{2h}\hfill \\ & & x_{2h3}x_{2h1}^1x_{2h}[x_{2h3},x_{2h2}]\hfill \\ \hfill \tau _{2g+1}(x_{2g})& =& x_{2g}\gamma _{2g+1}=x_{2g}[x_{2g1},x_{2g}]^1x_{2g1}^1\hfill \\ & =& x_{2g1}^1x_{2g}\hfill \\ \hfill \sigma _i(x_{2i})& =& x_{2i1}^1x_{2i}\hfill \end{array}$$
## 3 The proofs
###### Proposition 3.1
The quotient $`\pi /\pi ^gK`$ is characteristic and it is non-geometric if and only if $`g`$ is odd. If $`g`$ is even, the quotient $`\pi /\pi ^{2g}K`$ is characteristic and non-geometric.
###### Proof 3.2
$`\pi ^gK`$ is characteristic, while $`\pi ^g`$ and $`K`$ are. For $`g`$ odd, we have that $`\pi ^gK`$ is non-geometric since it is exactly the group described by Livingston in \[Li, Section 5\]. For $`g`$ even, say $`g=2h`$, it follows from Proposition 2.5 that $`\mathrm{\Pi }_{i=1}^h[x_{2i1},x_i]`$ is contained in $`\pi ^gK`$, so $`\pi ^gK`$ is geometric. On the other hand, for $`g`$ even, Livingston’s group is a quotient of $`\pi ^{2g}K`$, namely the one generated by the elements $`x_i^g`$ for $`i=1,\mathrm{},2g`$. Thus $`\pi ^{2g}K`$ is non-geometric.
###### Theorem 3.3
If $`g`$ is odd, the largest characteristic non-geometric quotient of $`\pi `$ is $`\pi /\pi ^gK`$ and if $`g`$ is even, the largest characteristic non-geometric quotient of $`\pi `$ is $`\pi /\pi ^{2g}K`$. The indices of these groups are $`g^{2g+1}`$ respectively $`(2g)^{2g}g=(2g)^{2g+1}/2`$.
###### Proof 3.4
The indices follow directly from Proposition 2.5. Let $`M`$ be any characteristic finite-index non-geometric subgroup of $`\pi `$. Let $`k`$ be the smallest positive integer such that for some simple closed not separating curve $`\delta `$ we have $`\delta ^kM`$. Since $`M`$ is characteristic and all simple closed not separating curves can be mapped one onto the other, we have that $`k`$ is the smallest positive integer with $`\delta ^kM`$ for all simple closed not separating curves $`\delta `$. Notice that $`k3`$. Namely, if $`k=2`$, then $`[x_1,x_2]=x_1^2(x_1x_2)^2x_2^2M`$, so $`M`$ is geometric, contradiction. We have $`\pi /M\pi ^{[2]}(𝐙/k𝐙)^{2g}`$ for some positive integer $`k`$.
Let $`P`$ be the subgroup of $`\pi ^{[2]}`$ generated by all $`[x_i,x_j]`$ for $`(i,j)`$ not related (recall that $`i<j`$). By an analogous argument we have that $`(\pi ^{[2]}M)/(\pi ^{[3]}M)P`$ is a quotient of $`(𝐙/m𝐙)^{g1}`$, generated by the elements $`[x_{2i1},x_{2i}]`$ for $`i=1,\mathrm{},g`$. Again by the classification of surfaces and by the fact that $`M`$ is characteristic, we get a uniform power $`l`$ such that $`[x_{2i1},x_{2i}]^tM`$ if and only if $`l`$ divides $`t`$. We have that
$$[x_{2i1}^k,x_{2i}][x_{2i1},x_{2i}]^k\text{ modulo }\pi ^{[3]}$$
and thus $`l`$ divides $`k`$ if $`k`$ odd and $`2l`$ divides $`k`$ if $`k`$ even, by Proposition 2.5. Similarly, we obtain a uniform power $`m`$ such that for $`(i,j)`$ not related, the elements $`[x_i,x_j]^t`$ are in $`M`$ if and only if $`m`$ divides $`t`$. Furthermore we have that $`m`$ divides $`l`$, since $`\tau _3([x_1,x_2]^l)[x_1,x_2]^l[x_1,x_3]^l`$ modulo $`\pi ^{[3]}`$.
Now suppose $`[x_1,x_2][x_3,x_4]`$ modulo $`M`$. Since $`M`$ is characteristic and all classes $`[x_{2i1},x_{2i}]`$ can be transformed one into the other by an automorphism of $`\pi `$, it follows that they all have different classes modulo $`M`$, thus $`[x_{2i1},x_{2i}][x_{2j1},x_{2j}]^1`$ is not contained in $`M`$. Consider the short exact sequence
$$1M\pi ^{[3]}/\pi ^{[3]}\pi /\pi ^{[3]}\pi /M\pi ^{[3]}1.$$
We claim that $`M\pi ^{[3]}/\pi ^{[3]}`$ does not contain any element of the form $`[x_i,x_j]`$ for $`(i,j)`$ not related (equivalently, $`m>1`$). Namely, suppose that there is a pair $`(i,j)`$, with $`(i,j)`$ not related, such that $`[x_i,x_j]M`$. Then $`[x_i,x_j]`$ is in $`M`$ for all $`(i,j)`$ not related, again by the classification of surfaces and the fact that $`M`$ is characteristic. We compute $`\tau _{2h1}([x_{2h2},x_{2h}])`$. We have the following identities modulo $`\pi ^{[3]}`$, where $`2hg`$:
$$\begin{array}{cc}& \tau _{2h1}([x_{2h2},x_{2h}])\hfill \\ & [x_{2h3}^1x_{2h2}x_{2h1},x_{2h3}x_{2h1}^1x_{2h}]\hfill \\ & [x_{2h3},x_{2h1}][x_{2h3},x_{2h}]^1[x_{2h2},x_{2h1}]^1[x_{2h2},x_{2h}][x_{2h3},x_{2h1}]^1\hfill \\ & [x_{2h3},x_{2h2}]^1[x_{2h1},x_{2h}]\hfill \\ & [x_{2h3},x_{2h}]^1[x_{2h2},x_{2h1}]^1[x_{2h2},x_{2h}][x_{2h3},x_{2h2}]^1[x_{2h1},x_{2h}].\hfill \end{array}$$
The product $`[x_{2h3},x_{2h2}]^1[x_{2h1},x_{2h}]`$ is contained in $`M`$ since the first three commutators are in $`M`$. This leads to a contradiction.
Now we claim that the abelian $`m`$-torsion subgroup of $`\pi /M\pi ^{[3]}`$ generated by all $`[x_i,x_j]`$ for $`(i,j)`$ not related, has rank $`(_2^{2g})g=2g^22g`$. Namely, suppose there is an element $`z=_{(i,j)notrel.}n_{i,j}[x_i,x_j]`$ contained in $`M`$ with all $`n_{i,j}\{0,\mathrm{},m1\}`$ (in additive notation). We compute a number of elements of the form $`\tau (z)z`$ to show that all these $`n_{i,j}`$ are actually zero. We use repeatly that $`n_{i,j}[x_i,x_j]`$ is in $`M`$ if and only if $`m`$ divides $`n_{i,j}`$
$$\begin{array}{ccccc}\hfill z_{2g}:=& \tau _2(z)z\hfill & =_{3j2g}n_{1,j}[x_1,x_j]\hfill & & \\ \hfill z_{2g1}:=& \sigma _g(z_{2g})z_{2g}\hfill & =n_{1,2g}[x_1,x_{2g1}]\hfill & \hfill & n_{1,2g}=0\hfill \\ \hfill z_{2g2}:=& \tau _{2g}(z_{2g1})z_{2g1}\hfill & =n_{1,2g1}[x_1,x_{2g}]\hfill & \hfill & n_{1,2g1}=0\hfill \\ \hfill z_{2g3}:=& \sigma _{g1}(z_{2g2})z_{2g2}\hfill & =n_{1,2g2}[x_1,x_{2g3}]\hfill & \hfill & n_{1,2g2}=0\hfill \\ \hfill z_{2g4}:=& \tau _{2g2}(z_{2g3})z_{2g3}\hfill & =n_{1,2g3}[x_1,x_{2g2}]\hfill & \hfill & n_{1,2g3}=0\hfill \\ & & \mathrm{}\hfill & & \\ & \tau _4(z_{..})z_{..}\hfill & =n_{1,3}[x_1,x_3]\hfill & \hfill & n_{1,3}=0\hfill \\ & \text{we get }z\hfill & =_{2j2g}n_{1,j}[x_1,x_j]\hfill & & \\ \hfill z_{2g}:=& \tau _1(z)z\hfill & =_{3j2g}n_{2,j}[x_2,x_j]\hfill & & \end{array}$$
It is clear that continuing in this way we we show that all coefficients $`n_{i,j}`$ are zero. Since there are $`(_2^{2g})2g`$ elements of this form, this proves the claim.
It remains to show that the index of this quotient is larger than the index of $`\pi ^gK`$ respectively $`\pi ^{2g}K`$. We have that the index of $`M`$ is at least $`\mathrm{\#}(\pi /M\pi ^{[2]})\mathrm{\#}(M\pi ^{[2]})/(M\pi ^{[3]})`$, which in turn is at least $`k^{2g}m^{2g^22g}`$. If $`m`$ is even, this leaves $`m=2`$ as smallest possibility. Since $`m`$ divides $`l`$, $`l`$ is also even and therefore $`2l`$ divides $`k`$. So the smallest possibility we get is $`4^{2g}2^{2g^22g}=(2^g)^{2g+2}`$. This is larger than $`(2g)^{2g+1}/2`$ for all $`g`$.
If $`m`$ is odd, the smallest possibility becomes $`3^{2g}3^{2g^22g}=(3^g)^{2g}`$, which is again larger than $`g^{2g+1}`$ for all $`g`$.
On the other hand, suppose $`[x_1,x_2][x_3,x_4]`$ modulo $`M`$. By the same argument, we get that all the classes $`[x_{2i1},x_{2i}]`$ are equal modulo $`M`$. Since $`\mathrm{\Pi }_{i=1}^h[x_{2i1},x_{2i}]M`$ for all $`h=1,\mathrm{},g1`$, the smallest possibility for $`(\pi ^{[2]}M)/(\pi ^{[3]}M)P`$ is $`(𝐙/g𝐙)`$. This implies that $`g`$ divides $`k`$ and $`2g`$ divides $`k`$ if $`g`$ even. Thus, $`M`$ is contained in $`\pi ^gK`$ or $`\pi ^{2g}K`$ if $`g`$ is even.
|
warning/0005/astro-ph0005274.html
|
ar5iv
|
text
|
# Wide-field imaging at mid-infrared wavelengths : reconstruction of chopped and nodded data
## 1 Introduction
Ground-based astronomical observations at infrared (IR) wavelengths can only be performed through a limited number of atmospheric windows (Thomas & Duncan (1993)). Under the best observing conditions, the atmospheric transmission closely approaches $`\tau 1`$ $`(100\%)`$ within a few narrow spectral intervals, but typical broad-band values are significantly smaller. Even at excellent sites like Mauna Kea, the median transmission in the 8–13 $`\mu `$window (N band) is $`\tau 0.85`$ (see e.g. the UKIRT web page).
Besides reducing the number of source photons reaching the detector, the atmosphere absorbs and thermally re-radiates isotropically a corresponding fraction of energy coming from the space and especially from the ground. In a first approximation, the atmosphere can be conveniently regarded as a gray-body radiator with emissivity $`ϵ=1\tau `$ (Kirchoff’s law) and temperature $`230250`$K. The corresponding photon flux, peaking at 10 $`\mu `$m, is huge compared to the celestial background, e.g. $`10^6`$ times higher than the zodiacal background at 10$`\mu `$m. The telescope itself also adds an important contribution to the background flux. Opaque surfaces within the optical beam (secondary mirror spiders, primary mirror cell and central obstruction) are near-blackbody radiators, whereas the mirrors themselves and the other warm optics, partially reflecting or transmitting, are gray-body emitters. Proper opto-mechanical design and regular cleaning of the optical surfaces can reduce the emissivity budget of a telescope to a few percents. Still, the total background flux at 10$`\mu `$within a 1$`\mu `$bandpass remains of the order of $`10^9`$ photons s<sup>-1</sup> m<sup>-2</sup> arcsec<sup>-2</sup>, roughly corresponding to an astronomical source of magnitude $`N3.0`$.
Today, it is possible to routinely detect at 10$`\mu `$point sources some 12 magnitudes fainter than the background per square arcsecond in a few minutes with 3–4 m telescopes and, in particular, UKIRT (Robberto & Herbst (1998)). To attain these performances special observing techniques must be adopted, but they are not without drawbacks.
If $`x,y`$ are angular coordinates in the sky, the signal $`s_P`$ coming from the direction $`\{x,y\}`$ at time $`t`$ and detected on the corresponding pixel $`P`$ of the detector can be expressed as:
$$s_P=T_P[f(x,y)+a(x,y,t)]$$
(1)
where $`f`$ is the unknown brightness distribution of the celestial source and $`a`$ is the large and time-variable thermal background flux. The transfer function of the detection system $`T_P`$ includes the collecting area of the telescope, the field of view of each individual pixel and the overall optical transmission. Under the conditions described by equation (1) it is clear that a small error in the estimate of $`a`$ will dramatically affect the extraction of the signal $`f`$.
The background $`a`$ can be obtained in principle by pointing the telescope to a sky area close to the region of interest at a time $`t^{}`$ close to $`t`$. Assuming that this area corresponds to a shift $`\mathrm{\Delta }`$ in the $`y`$ coordinate, then the new signal $`s_P^{}`$ detected at the pixel $`P`$ is
$$s_P^{}=T_P[f(x,y+\mathrm{\Delta })+a(x,y+\mathrm{\Delta },t^{})].$$
(2)
If the sky area close to that of interest is empty, i.e. $`f(x,y+\mathrm{\Delta })=0`$, and close enough in space and time that the background is approximately the same, i.e. $`a(x,y+\mathrm{\Delta },t^{})a(x,y,t)`$, then by subtracting equation (2) from equation (1) one can obtain $`T_Pf(x,y)`$. In this way the signal can be known within the accuracy of the system response.
In practice, the telescope must sample the two areas fast enough that the temporal fluctuations of $`a`$ are small. The actual frequency depends on various factors: observing wavelength, weather conditions, telescope location etc., but is typically faster than a few Hz (Kaeufl et al. (1991)). Whereas it is, in general, impossible to repeatedly move a telescope at these frequencies, a single optical element can be rapidly “chopped” between two slightly different positions, allowing the detector to see two nearby sky areas. To minimize pupil motion at the cold stop, the secondary mirror of the telescope is often undersized (so it becomes the exit pupil of the tescope) and modulated. This classic technique is called secondary chopping and the quantity $`\mathrm{\Delta }`$ is the chopping throw or chopping amplitude. Given a certain amplitude, the maximum chopping frequency is constrained by the settling time of the secondary mirror structure, typically of the order of 20-50 milliseconds.
Even neglecting the efficiency loss due to the time spent observing the sky and waiting for the secondary mirror to settle, this method presents important limitations. First, by moving an optical element of the system, the detectors look through different parts of the optical system (including high emissivity surfaces like central obstruction, mirror edges, spiders, etc.). Therefore, the resulting differential image turns out to be affected by a residual background variation due to the thermal differences between the two beams. In other words, chopping is equivalent to rapid switching between two different telescopes, one $`(A)`$ for the source and another $`(B)`$ for the sky. We will denote by $`\mathrm{\Delta }a_{AB}`$ the residual difference between the corresponding background patterns. Second, for optical and mechanical reasons the typical chopping throws are usually less than 60 arcseconds, possibly much less for 8-m class telescopes. If the source is extended enough, and/or if the telescope is sensitive enough, it can be $`f(x,y+\mathrm{\Delta })0`$. Therefore, the subtraction of equation (2) from equation (1) gives:
$$\mathrm{\Delta }s_A=s_Ps_P^{}=T_P[f(x,y)f(x,y+\mathrm{\Delta })+\mathrm{\Delta }a_{AB}]$$
(3)
where with $`\mathrm{\Delta }s_A`$ we indicate that the source has been observed with the $`(A)`$ beam.
To remove the term $`\mathrm{\Delta }a_{AB}`$, the so-called beam-switching or nodding technique is applied: the telescope is pointed to a different point on the sky, so that the source will be observed with the $`(B)`$ beam (this maximizes the signal-to-noise ratio). In our notation, the telescope is shifted by $`\mathrm{\Delta }`$ arcseconds in the $`y`$ coordinate. In this way, at the pixel $`P`$ the signal $`s_P^{\prime \prime }`$ is obtained and, repeating the entire sequence, the result is:
$$\mathrm{\Delta }s_B=s_P^{\prime \prime }s_P=T_P[f(x,y\mathrm{\Delta })f(x,y)+\mathrm{\Delta }a_{AB}].$$
(4)
Subtracting equation (4) from equation (3), one gets the so-called chopped and nodded image:
$$g_\mathrm{\Delta }(x,y)=\mathrm{\Delta }s_A\mathrm{\Delta }s_B=T_P[f(x,y\mathrm{\Delta })+2f(x,y)f(x,y+\mathrm{\Delta })],$$
(5)
i.e. an image which is independent of the atmospheric background and telescope thermal pattern. If the source brightness distribution is sufficiently compact, i.e. if $`f(x,y\mathrm{\Delta })=f(x,y+\mathrm{\Delta })=0`$, then the problem of extracting $`f(x,y)`$ is solved. Otherwise a method for recovering $`f(x,y)`$ from the image $`g(x,y)`$ is required.
As already pointed out (Beckers (1994), Kaeufl (1995)), there is little doubt that such a method is becoming highly needed. The continuous technological developments both in the field of infrared array detectors, which allow one to reach the natural background-noise limited performances for broad-band imaging, and of large telescope engineering, with approximately a dozen of 8-meter class telescopes in operation or in an advanced construction phase, are greatly improving the sensitivity of thermal infrared instrumentation. Because giant telescopes have smaller pixel scales and have limited chopping amplitudes, the case $`f(x,y\mathrm{\Delta })=f(x,y+\mathrm{\Delta })=0`$ cannot be regarded any more as the typical one so that the $`f(x,y\pm \mathrm{\Delta })`$ components of the scene appear as negative signals in the final image (see equation (5)). In such cases a restoration method is required to obtain, in the general case of a structured astronomical object, an image free from negative values and reproducing, as far possible, the correct intensity distribution within the source.
In (Bertero et al. (1998), hereafter BBR98) a preliminary investigation of this problem has been performed and it has been found that an iterative regularization method, the so-called projected Landweber method, can provide a viable solution. The method provides a positive restoration of the chopped and nodded image by removing completely the negative counterparts of compact and extended sources. The mathematical properties of the problem, as well as the validation of the method by means of simulated images, have been presented in (Bertero et al. (1999), hereafter BBDR99).
In Section 2 of this paper we outline the mathematical model we use for describing the chopping and nodding technique, we discuss the various kinds of instrumental errors which are not included in the mathematical model and we describe the iterative restoration method we propose. In addition, we discuss the criteria for stopping the iterations, the so-called stopping rules, which are an important issue for the correct use of the method. Finally, we describe three types of artifacts and show how they are related to the mathematical properties of the problem.
In Section 3 we apply the method to real astronomical images taken at the UKIRT telescope. We show that not only the effects of the negative counterparts of the sources are successfully corrected, but also the noise is reduced, if the number of iterations is properly chosen. We indicate the stopping rule which is the most convenient at the present stage of our analysis. We also show that an enlargement of the field is possible, even if the restoration outside the field is, in general, less accurate than within. Moreover our restored images provide several examples of the artifacts described at the end of Section 2.
In Section 4 we propose three computational and observational procedures for the reduction of the artifacts, and we show that by applying these methods it is possible to produce mid-IR images of extended sources with high accuracy and sensitivity. Finally, in Section 5 we summarize our results and provide a few practical hints on the instrument alignment at the telescope that should permit the algorithm to produce the optimal results.
## 2 An outline of the restoration method
We take, for simplicity, $`T_P=1`$ in equation (5). Then by computing the Fourier transform of both sides we get
$$\widehat{g}_\mathrm{\Delta }(\omega _x,\omega _y)=4sin^2(\frac{1}{2}\mathrm{\Delta }\omega _y)\widehat{f}(\omega _x,\omega _y)$$
(6)
where $`\omega _x,\omega _y`$ are the spatial frequencies associated with the variables $`x,y`$ respectively. As already observed by Beckers (Beckers (1994)), equation (6) shows that the chopped and nodded image does not contain information about $`\widehat{f}(\omega _x,\omega _y)`$ at the frequencies $`\omega _{y,k}=2\pi k/\mathrm{\Delta }`$ $`(k=0,\pm 1,\pm 2,\mathrm{})`$, so that the chopping and nodding technique looks equivalent to the application of a Fourier grating to the original object. However, the Fourier transform of the measured data is not zero at these frequencies because they are contaminated by noise (Kaeufl (1995)). As a consequence, the restoration of $`\widehat{f}(\omega _x,\omega _y)`$ cannot be obtained by dividing the Fourier transform of the chopped and nodded image by the factor $`sin^2(\mathrm{\Delta }\omega _y/2)`$.
In general Fourier based methods cannot be used for this problem because the functions $`f`$ and $`g`$ are not defined on the same domain. If the region of interest corresponds to the interval $`[0,Y]`$ in the y-variable, then the image $`g_\mathrm{\Delta }(x,y)`$, defined on this interval, receives contributions from the values of $`f(x,y)`$ on the broader interval $`[\mathrm{\Delta },Y+\mathrm{\Delta }]`$. A method, able to restore $`f(x,y)`$ on this interval, will provide an enlargement of the field.
Let us assume that the detector plane is partitioned into $`N\times N`$ pixels, with size $`\delta \times \delta `$, labeled with an index $`j`$ corresponding to the columns and an index $`m`$ corresponding to the rows of the array. Since the chopping is in the $`y`$-direction, it is parrallel to the columns of the array. Further we assume that the chopping amplitude $`\mathrm{\Delta }`$ is a multiple of the sampling distance $`\delta `$, i.e. $`\mathrm{\Delta }=K\delta `$ with $`K`$ integer and smaller than $`N`$. Most of the images we consider are $`128\times 128`$ and $`K`$ is typically (but not necessarily) betwen 30 and 50.
Let $`g_{j,m}`$ and $`f_{j,m}`$ be the samples of $`g(x,y)`$ and $`f(x,y)`$ respectively. For each $`j`$ the values $`g_{j,m}`$, with $`m`$ running from $`1`$ to $`N`$, form a vector of length $`N`$ which will be denoted by $`𝐠_j`$. It receives contributions from $`N+2K`$ values $`f_{j,m}`$, with $`m`$ running from $`1`$ to $`N+2K`$, which form a vector of length $`N+2K`$, denoted by $`𝐟_j`$ . The components of $`𝐟_j`$ with $`m`$ running from $`K+1`$ up to $`K+N`$ correspond to the sampling points in the region of interest, which will be called the observation region.
Using these notations, equation (5) with $`T_P=1`$ is replaced by the following discrete relationship
$$g_{j,m}=f_{j,m}+2f_{j,m+K}f_{j,m+2K}$$
(7)
which, by introducing the matrix $`[A]`$, defined by
$$[A]_{m,n}=\delta _{m,n}+2\delta _{m+K,n}\delta _{m+2K,n},m=1,2,\mathrm{},N;n=1,2,\mathrm{},N+2K$$
(8)
and called in the following the imaging matrix, can be written in the synthetic form
$$𝐠_j=[A]𝐟_j.$$
(9)
The image restoration problem consists in estimating $`𝐟_j`$ (a vector of length $`N+2K`$), given $`𝐠_j`$ (a vector of length N). Since the imaging matrix does not depend on the index $`j`$, one has to solve the same restoration problem for all columns of the image.
There are various kinds of errors that may arise using the model described by equation (9). The first is due to the noise, which mostly consists of read-out and background Poisson noise. Since in the case of large background the latter dominates and can be approximated by white Gaussian noise, the true image will be given by
$$𝐠_j=[A]𝐟_j+𝐰_j$$
(10)
where $`𝐰_𝐣`$ is a random vector generated by a white Gaussian process. When we approximate equation (10) with equation (9) we commit an error which can be amplified by the restoration method and therefore must be properly controlled.
A second kind of error occurs when the chopping amplitude $`\mathrm{\Delta }`$ is not an integer multiple of the pixel size $`\delta `$. In such a case the simple expression we use for the matrix $`[A]`$ must be replaced by a more complicated one. However, we performed numerical simulations showing that this effect is not very important when the ratio $`\mathrm{\Delta }/\delta `$ is greater than 10, as it is in most practical situations: in practice the same results can be obtained by using the exact model or the simpler one described above, with a value of $`K`$ given by the integer closest to $`\mathrm{\Delta }/\delta `$.
Finally, a third kind of error can be generated by a misalignment of the detector array with the chopping and nodding direction. This error is the most serious one, because it violates the assumption that image restoration can be performed column by column. If this occurs, it must be corrected by an appropriate rotation of the image.
The imaging matrix $`[A]`$ is rectangular, with $`N`$ rows and $`N+2K`$ columns. A complete analysis of the mathematical properties of this matrix is given in (BBDR99), where it is shown that a unique solution of the restoration problem can be obtained if one looks for a positive solution with minimal root mean square value. However, since the matrix $`[A]`$ is ill-conditioned, this solution can be corrupted by an amplified propagation of the data noise, so that regularization methods must be used for controlling this noise propagation (for an introduction to regularization methods in image restoration see, for instance, Bertero & Boccacci (1998)).
Taking into account that restored images must be positive and not corrupted by noise amplification, we have implemented a particular version of the so-called projected Landweber method (Eicke (1992)), proposing the following iterative method (BBR98):
$`𝐟_j^{(0)}`$ $`=`$ $`0`$ (11)
$`𝐟_j^{(k+1)}`$ $`=`$ $`P_+\left[𝐟_j^{(k)}+\tau \left([A]^Tg[A]^T[A]𝐟_j^{(k)}\right)\right]`$
where:
* $`[A]^T`$ denotes the transposed of the matrix $`[A]`$;
* $`P_+`$ is the convex projection onto the closed and convex set of the non-negative vectors, defined by
$`(P_+𝐟)_n`$ $`=`$ $`f_n,\mathrm{if}f_n>0`$
$`=`$ $`0,\mathrm{if}f_n0;`$
* $`\tau `$ is a positive parameter, known as relaxation parameter, which, for our problem, must be smaller than $`0.125`$ (in our code we take $`\tau =0.1125`$).
The implementation of the algorithm defined by equation (11) is discussed in (BBDR99). Here we focus on the fact that a basic property of the method is that it has a regularizing effect, usually called semiconvergence: the iterates $`𝐟_j^{(k)}`$ first approach the unknown brightness distribution and then move away, because the iterations amplify noise propagation (Bertero & Boccacci (1998)). For this reason, it is very important to have a criterion for selecting the optimal number of iterations in order to obtain the best possible approximation of the unknown brightness distribution. Below we discuss the criteria which can be used to stop the iterations.
We propose two stopping rules based on the so-called discrepancy principle (Bertero & Boccacci (1998)). The first works column by column. We define the discrepancy $`\epsilon _j^{(k)}`$ between the j-th column of the measured data and the j-th column of the data computed by means of the k-th iterate as the root means square (r.m.s.) value of the vector $`[A]𝐟_j^{(k)}𝐠_j`$:
$$\epsilon _j^{(k)}=[A]𝐟_j^{(k)}𝐠_j=\left(\frac{1}{N}\underset{n=1}{\overset{N}{}}([A]𝐟_j^{(k)})_ng_{j,n}^2\right)^{1/2}.$$
(13)
From general results on the projected Landweber method (Eicke (1992)), it is know that $`\epsilon _j^{(k)}`$ is a decreasing function of $`k`$, tending to zero when $`k\mathrm{}`$. Then, according to the discrepancy principle, the iterations can be stopped when $`\epsilon _j^{(k)}`$ becomes smaller than some estimate $`\epsilon _j`$ of the r.m.s. error $`𝐰_𝐣`$. In the case of white noise with variance $`\sigma ^2`$, a quite natural estimate is $`𝐰_𝐣\sigma `$, so that the iterations can be stopped when $`\epsilon _j^{(k)}\sigma `$. This criterion essentially means that one does not look for a very accurate fit of the data because, in such a case, one would be looking for a solution fitting not only the signal but also the noise.
Even if the same value of $`\epsilon _j`$ is used for all columns, the number of iterations in general is changing from column to column: the number of iterations is small if the column is characterized by a low value of the signal-to-noise ratio S/N and is larger if S/N is higher. If one does not expect the ratio S/N to change dramatically from column to column it may be more convenient to use a second stopping rule which provides the same number of iterations for all columns. To this purpose we define the average relative discrepancy as follows
$$\epsilon ^{(k)}=\left(\frac{\underset{j=1}{\overset{N^{}}{}}[A]𝐟_j^{(k)}𝐠_j^2}{_{j=1}^N^{}𝐠_j^2}\right)^{1/2}$$
(14)
where $`N^{}`$ is the number of columns to be restored. The iterations can be stopped when $`\epsilon ^{(k)}`$ is smaller than some estimated value $`\epsilon `$ of the relative r.m.s. error affecting the image. Typically we use a value of $`\epsilon `$ corresponding to a few percent error. If an estimate of $`\sigma `$ is available, then an estimate of $`\epsilon `$ can be obtained by replacing each term in the numerator of equation (14) with $`𝐰_j^2\sigma ^2`$. Therefore, if we have two images of the same object with different noise levels, we must use a larger value of $`\epsilon `$, hence a smaller number of iterations, for the noisier one.
However, in the case of very noisy images it is important to observe that, if the value of $`\epsilon `$ defined above is used, then the criterion stops the iterations too early. This effect is due to a property of the discrepancy principle which has been discovered empirically and is well documented in the literature (Bertero & Boccacci (1998)). The criterion can be corrected by using a value of $`\epsilon `$ smaller than that defined above (for instance by a factor of 2).
In the next Section, by applying both stopping criteria to real images, we show that the criterion based on equation (14) works better than that based on equation (13).
As shown in (BBR98) and (BBDR99), using both synthetic and real images, the method outlined above provides satisfactory restorations under many circumstances. However the restored images may exhibit a few typical artifacts. As discussed in (BBDR99) these artifacts are due to the particular structure of the imaging matrix $`[A]`$ and not to the various kinds of error discussed above (noise, non integer chopping, misalignment): these artifacts are present also in the restoration of synthetic images which are free of these errors (see BBDR99). For sake of simplicity they can be classified as Type A), Type B) and Type C) artifacts.
Type A) artifacts \- Multiple “ghost” images, spaced by $`K`$, of bright stars; they may appear as dark images over a bright background or as bright images over a dark background. These multiple images are a residual effect of the missing frequencies discussed at the beginning of Section 2 ( see equation (6)). Their presence means that the restoration method does not provide a complete interpolation of the missing frequencies, i.e. the Fourier components of the unknown object corresponding to these frequencies are not completely restored. Since the missing frequencies depend on the chopping amplitude $`K`$, the simplest way for overcoming this difficulty is to use other images with different chopping amplitudes. This raises the question of combining two (or more) images with different chopping amplitudes to avoid the zeros in (6) at the spatial frequencies $`\omega _{y,k}=\pm 2\pi k/\mathrm{\Delta }`$.
Type B) artifacts \- Regions where the restored image is exactly zero. They correspond to the negative counterparts of bright extended sources. If these regions contain faint positive sources, these sources can be lost.
Type C) artifacts \- Discontinuities of the restored brightness distribution at the rows corresponding to the following values of the index $`n`$ in the restored image: $`n=`$ $`K_1,K,K+K_1,2K,\mathrm{},K_1+(q+1)K,(q+2)K`$ ( $`2q+4`$ jumps), where $`q`$ is the integer part of the quotient of the division of $`N`$ by $`K`$ and $`K_1`$ is the remainder ($`N=qK+K_1,K_1<K`$). This effect is related to the finite size of the image and is especially evident when bright parts of the object remain outside the observation region.
The characteristics of these artifacts will be discussed through the next Section, whereas in Section 4 we shall treat methods for their reduction.
## 3 Application of the method to real images
### 3.1 MAX observations
In this section we show the results obtained by applying our algorithm to a sample of real mid-infrared images. The data, taken in different observing campaigns and in part still unpublished, have been obtained using MAX (Robberto & Herbst (1998)), the mid-IR imager developed by the Max-Planck-Institut für Astronomie (MPIA) for the United Kingdom Infrared Telescope (UKIRT). MAX is currently equipped with a Rockwell International $`128\times 128`$ SiAs BIB array optimized for high-background applications. The all-reflective optical design provides a scale of $`0.27`$ arcsec/pixel, corresponding to a field of view of $`35\times 35`$ arcsec.
Observations with MAX are usually performed under excellent weather conditions and take full advantage from the top-ring and the hexapod secondary mirror mount with tip-tilt adaptive control developed at the MPIA for UKIRT. Since the fast-guiding system operates on both chopping beams, MAX routinely provides images close to the diffraction limit ($`\lambda /D=0\stackrel{}{\mathrm{.}}54`$ at $`10\mu `$m on UKIRT) on hour-long integration times. To reach these performances, the chopping throw and nodding amplitude must be finely tuned to an integer multiple of $`0\stackrel{}{\mathrm{.}}84`$, the guide camera pixel size, corresponding to $`3`$ MAX’s pixels (this value has recently changed to $`0\stackrel{}{\mathrm{.}}94`$ \- N.Rees, private communication).
The data presented here were generally taken using the standard chopping and beam switching technique. However, since the possible use of the inversion algorithm was not considered at the time most of the observations were done, the alignment of the array with the chopping and nodding directions was never perfectly tuned. During image post-processing images were rotated to compensate for the original misalignments, so to satisfy the basic assumption of our restoration method. Due to excellent cosmetic quality of the detector, MAX raw data required minimal cleaning. On the other hand, we did not attempt to correct for flat field. Accurate flat fielding at mid-IR wavelengths is known to be a critical issue due to the fast and non-uniform variations of the background signal. However, we must notice that for various reasons the systematic errors that in principle would require flat-field correction in fact turn out to be less important in our wavelength range. Namely:
1. non linearity in pixel response is unimportant due to very low dynamic range of the images;
2. the small fields of view allow to build instruments with a fairly uniform detector illumination, i.e. vignetting is normally negligible (at least, this is the case of MAX);
3. differences in the detector’s pixel-to-pixel response are in general less than 5% and to some extent mitigated by the oversampling;
4. for faint sources, the flat-field uncertainty in subtracted images can be easily dominated by the background noise.
Taking chopping pairs with chopping throws small enough, we can (almost) simultaneously compare the flux from bright stars in different parts of the MAX’s field of view. This check, routinely done at the telescope, provides results that are consistent within $`1\%`$, i.e. less than the error typically associated to the absolute flux calibration in this wavelengths regime. It is clear, however, that any reduction of systematic effects through accurate flat fielding strategy should further improve the results obtained by our reconstruction algorithm.
### 3.2 Bright point sources
We show first the results obtained by applying the algorithm to a bright, isolated point source. Although our restoration method is not really needed in such a case, we will use it to investigate his effect on noise propagation and photometric accuracy as well as the generation of Type A artifacts.
Figure 1a) shows an image of the bright star BS 1370 obtained at UKIRT on 1996 August 26-27 through a broad-N band filter ($`\lambda _{eff}=10.16\mu `$m, $`\mathrm{\Delta }\lambda =5.20\mu `$m). The integration times were set to 6.1 milliseconds/frame and 12 seconds total, chopping at $`2`$Hz with $`\mathrm{\Delta }=10`$ arcseconds throw in the N-S direction ( $`K=36`$ ). Image post-processing consisted in filtering out some electronic noise in 1 out of 16 preamplifier channels and a counterclockwise rotation by $`2.3^{}`$. After post-processing the standard deviation of the noise is estimated using 10 columns in the blank region of the image. The value we obtain is $`\sigma 10.6`$ counts/pixel and this is the value to be used when applying the stopping rule based on equation (13). The corresponding value of $`\epsilon `$ to be used according to equation (14), is about $`0.033`$. Note that the maximum value of the star intensity is $`\mathrm{2.\; 10}^4`$ counts. Since the restoration method reduces this value by a factor of about 2 (see equation (5)), it should also reduce the noise by a similar factor. In the following we denote by $`\sigma _r^2`$ the variance of the noise in the restored image. It is estimated on the same columns used for the estimation of $`\sigma ^2`$.
We have applied the method to the image of Figure 1a), using the two stopping rules introduced in Section 2. Using the first one, the algorithm performs only one iteration when the column does not contain signal; the maximum number of iterations is $`44`$ in the case of the column through the maximum of the star. After restoration, the peak value becomes $`\mathrm{1.04\; 10}^4`$ and the variance of the noise is $`\sigma _r1.7`$ counts/pixel. Therefore noise is reduced by a factor $`6`$. This very high noise reduction is due to the strong smoothing effect of the first iterations of the Landweber method. As a consequence the high frequency components of the noise are supressed. Regarding photometric accuracy the integrated flux (estimated through standard multi-aperture photometry) equal to $`1.03`$ the flux of the original image, i.e. photometric accuracy is preserved within $`3\%`$.
Using the second stopping rule, with $`\epsilon =0.033`$, the algorithm stops after $`13`$ iterations. The $`\sigma _r`$ is smaller than $`\sigma `$ by a factor $`2.5`$ while photometric accuracy is now preserved within $`0.4\%`$. We emphasize that the noise reduction we obtain is just what we expect because, as already pointed out above, the restoration algorithm reduces the signal by a factor of 2.
This example clearly illustrates that the method assures noise reduction and rather good photometric accuracy. Both effects depend on the number of iterations and, using our two stopping rules, it results that the first one provides an excessive noise reduction and a poorer photometric accuracy with respect the second one. For these reasons the second stopping rule seems to be preferable.
Concerning artifacts, the two deep negatives counter-images created by the chopping and nodding technique are completely removed. Since the linear scale used in Figure 1b) makes difficult to notice the presence of any type of problem we plot in Figure 2 the profiles of the original and restored image along the column passing through the star maximum. The profile of the original image has been divided by 2 for comparison and the range of the ordinates reduced to $`\pm 10\%`$ the stellar maximum. It can be seen that the stellar profile is reproduced with great accuracy and that the two large negative counterparts of the original image are replaced by two small positive ghosts, similar to spikes. The integrated photometry of these spikes is about $`3.5\%`$ the stellar flux of the restored image. Besides these spikes two other positive ghosts appear at distances $`\pm 2\mathrm{\Delta }`$, outside the original target. Their integrated fluxes are about $`10\%`$ the stellar flux. We point out again that the positions of these ghosts can be exactly predicted since they are located at distances which are multiple of the chopping throw.
According to our stopping rules, changing the value of $`\epsilon `$ will also change the number of iterations: more precisely decreasing $`\epsilon `$ increases $`k`$. The comparison of the results obtained with different number of iterations allows us to provide the following rule of thumb: an increase of iterations will cause the noise of the restored image to increase, as well as the integrated flux of the star. This is first underestimated and then overestimated by the algorithm.
### 3.3 Faint point sources
In order to test the method near the sensitivity limit, we apply our algorithm to an image containing two faint and isolated point sources, the system DH/DI Tau (Meyer et al. (1997)). The observational parameters and post-processing are similar to those presented in Section 4.2, with a total integration time $`t_{int}=250`$ s. Again, since the chopping direction was not perfectly aligned to the array columns, the image has been properly rotated before applying the inversion algorithm.
In our analysis we have considered only the central part of the original image (more precisely the columns from $`26`$ to $`106`$) in order to eliminate the incomplete lateral columns present in the rotated image, which could prevent an accurate estimate of $`\epsilon `$. After rotation $`\sigma `$ is about $`4`$ counts/pixel and the corresponding value of $`\epsilon `$ is about $`.5`$. This large value of $`\epsilon `$ implies that we have an example of a very noisy image.
We have used again the two stopping rules of Section 2 with these values of $`\sigma `$ and $`\epsilon `$. The number of iterations allowed by the two criteria is now very small (in the case of the second one, just one iteration). The first criterion provides $`\sigma _r=0.6`$ and underestimates the integrated fluxes of DH Tau (the bright source) and DI Tau by about $`20`$%. The second criterion provides again $`\sigma _r=0.6`$ and underestimates the integrated flux of DH Tau by about $`15`$% while correctly reproduces that of DI Tau. Therefore both criteria provide a considerable noise reduction but the second criterion works better even if the integrated flux of the brighter star is not really satisfactory.
This result is due to the property of the discrepancy principle discussed in Section 2. For this reason we reapplied the algorithm by stopping the iterations when $`\epsilon _j^{(k)}\sigma /2=2`$, in the case of equation (13), and when $`\epsilon ^{(k)}\epsilon /2=2.5`$, in the case of equation (14). In this case, the first method provides $`\sigma _r=1`$ and the integrated flux of DH Tau is underestimated by $`5`$% while that of DI Tau is overestimated by $`5`$%. The second criterion provides $`\sigma _r=1.2`$ and an underestimation of about $`3.5`$% for DH Tau and of $`0.7`$% for DI Tau. Again the criterion based on the average relative discrepancy turns out to provide better results, since both integrated fluxes are in nice agreement with the flux measured before the inversion.
The original image (after rotation) as well as the best restored image are displayed in Figure 3. In the original image DH Tau is the brighter source, clearly visible with its two negative counterparts, while DI Tau is the fainter source $`15`$ arcseconds to the S-W. The derived fluxes, as reported by Meyer et al. (1997), are $`F_N=0.137\pm 0.005`$Jy (6.26 mag) for DH Tau and $`F_N=0.030\pm 0.005`$Jy (7.90 mag) for DI Tau. The restored image shows no spikes or ghosts so that the negative counterparts of the brightest star (DH Tau) appear properly cancelled.
It must be noted that the rotation introduces a certain amount of high-frequency noise filtering depending on the rotation angle and origin. To investigate the noise decrease independently of the particular rotation/filtering applied, we added, somewhat arbitrarily, that amount of uniform Gaussian noise needed to preserve the average noise level of the original (before rotation) image. This new image has $`\sigma 7`$ counts/pixel and $`\epsilon 0.6`$. By applying the method with the stopping based on the average relative discrepancy (using a reduction of $`\epsilon `$ by a factor of 2) we obtain a result very close to the previous one still with a correct noise reduction and a non-significant marginal variation of photometric accuracy.
From the analysis of the restoration of isolated sources, we conclude that the criterion based on the average relative discrepancy provides better results. Moreover, in the case of very noisy images, it is convenient to reduce by a factor $`2`$ the value of $`\epsilon `$ obtained from the $`\sigma `$ of the noise. We will adopt these rules in the following sections.
### 3.4 Bright extended sources
In this subsection we show how the algorithm performs when bright point sources coexist with fainter extended structures, i.e. when the field is characterized by high dynamic range signal. The central parsec of the Galaxy represents an ideal target in this respect, also because most of its prominent features lie within the MAX’s field of view. The image presented here (Figure 4) has been obtained on the night of 11 April 1998 on UKIRT with adaptive tip-tilt correction. We used a “N-narrow” filter with $`\lambda _c=11.6\mu `$m, $`\mathrm{\Delta }\lambda =2.5\mu `$m, chopping throw $`10\mathrm{"}`$ in the north-south direction, and 119.6 seconds total integration time on source (i.e. on the positive image). The airmass was $`z=1.55`$.
Once again, before applying the inversion algorithm, we had care of rotating the image by 0.7 degrees clockwise in order to align the chopping direction to the array columns. The image was also expanded to a $`256\times 256`$ size, in order to get a chopping throw $`74.9`$ pixels, allowing us to run the algorithm with $`K=75`$. Due to the complex structure of the source, almost entirely filling the field, we were unable to get a meaningful estimate of $`\sigma `$ using a sufficiently broad black part of the image. Given the brightness of the source, we have applied the criterion based on the average relative discrepancy assuming $`\epsilon =0.08`$. The algorithm stops after 63 iterations. In Figure 5 we present the reconstruction result with two different cuts, in order to evidence the full dynamic range of the image.
The reconstructed image clearly shows several details, both of the “northern arm” and of the “bar”, which were barely visible or completely lost in the negative features of the original image. Quite remarkably, the algorithm correctly finds the IRS 8 source at the very last rows (top) of the reconstructed image. Moreover photometry of sources like IRS 7 turns out to be more reliable once the local patchy negative background has been removed.
On the other hand, the reconstructed image is not entirely free from Type A) artifacts. Ghosts of the brightest sources are clearly visible in the lower and upper part of the image. Especially IRS 1 produces a bright artificial spot $`5\mathrm{"}`$ south of IRS 9, but virtually every bright IRS source generates some signature at the position of his negative counterparts, and in some case even at the corresponding multiple distances. To illustrate more in detail how artifacts look like, we plot in Figure 6 the vertical cut passing through IRS 1 on both the original and reconstructed image. The original image (thin line) shows the two negative “valleys” at pixels yr. 20 and yr. 94. At the same places, the reconstructed image presents two residual peaks, and especially at pixel yr. 20 the ghost feature appears prominent.
Methods for the reduction of these artifacts will be proposed in the next Section.
### 3.5 Faint extended sources
Here we apply the algorithm to the case of diffuse emission completely filling the field. It is clear that, in this situation, no algorithm can reconstruct the absolute value of the original image, lost due to the differential observing technique. In Figure 7a) we present the image of an area within the Orion nebula taken at UKIRT on the night of the 9th February 1997 through the standard N-band filter. The main observing parameters were: integration time 10 ms/frame and 491.52 s total on source, chopping throw $`5`$ arcseconds in N-S direction, airmass $`z1.23`$.
The field is located approximately 2 arcminutes S-E of the Trapezium stars and contains two point sources. That on top is the proplyd HST8=206-446 (O’Dell et al. (1993), O’Dell & Wong (1996)). The diffuse diagonal feature crossing the field is the Orion “bar”, i.e. the ionization front created by the Lyman-C photons produced by the Trapezium stars, $`\theta ^1C`$ in particular.
An estimate of the noise has been attempted by analysing regions of the original image characterized by a rather uniform intensity distribution. We have obtained $`\sigma 4000`$ counts/pixel and the corresponding value of $`\epsilon `$ is $`0.2`$. Therefore we have another example of a rather noisy image. For this reason we have used the second stopping rule with $`\epsilon =0.1`$. The corresponding number of iterations turns out to be $`49`$.
The result of the reconstruction, performed on the original image without any rotation/filtering, is presented in Figure 7b). It can be seen that the restored image results nicely free from Type A) artifacts: both stellar images appear properly reconstructed with no relevant ghosts. The bar structure turns out to be more detailed, but the effect of its negative counterpart on the northern side is a flat reconstruction with a large number of zeros. This is a typical example of a Type B) artifact better evidenced in Figure 8 where we plot vertical cuts through the HST8 maximum both in the original image and in two restored images obtained with $`\epsilon =0.2`$ and $`\epsilon =0.1`$ respectively. The Type B) artifact corresponding to the large negative counterpart of the “bar” looks the same in the two restorations. Their comparison allows us to conclude that an increase in the number of iterations enhances both the star peak and the “bar”. Again the noise is reduced by the restoration algorithm.
### 3.6 Bright sources outside the field
Before concluding this section, we must refer to that kind of artifacts that often arise when the image to be restored contains the negative counterpart of bright extended sources lying outside the field. At the end of Section 3 they were classified as Type C) artifacts. They appear as a series of horizontal jumps in the reconstructed images, with periodicity that depends on $`K`$ and $`K_1`$. As an example, we show in Figure 9a) an image of the Orion nebula centered on the dark-silhouette proplyd 167-231 (Robberto, Beckwith & Herbst (1999)) taken with similar parameters of Figure 7, except for the 10″chopping throw. The corresponding restored image, obtained with average relative discrepancy $`ϵ=0.02`$, is displayed in Figure 9b) and clearly shows the horizontal jump artifacts. In the next Section we will describe a simple method for efficiently reduce this kind of artifacts.
## 4 Artifacts reduction
We propose three methods which can be used to eliminate the artifacts illustrated in the previous Section. The first is both observational and computational and it is a way for reducing Type A) and Type B) artifacts. The second is only computational since it is based on manipulations performed on a single chopped and nodded image and acts only on Type C) artifacts. The last one is again observational and computational and can be also used for reducing the Type C) artifacts.
### 4.1 Inversion of multiple images of the same object
The analysis of the previous Section indicates that the ghosts of a very bright source, i.e. Type A) artifacts, or the areas with zero value, i.e. Type B) artifacts, are very frequent and may reduce the quality of the restorations. As we observed in Section 2, these artifacts are due to a lack of information in the chopped and nodded image (the missing frequencies). Images containing different pieces of information can then be restored by means of the projected Landweber method and recombined. It is clear that no pair of values of $`K`$ should have a common divisor. In practice, this procedure is similar to that suggested by Beckers (Beckers (1994)), but is completely different in what concerns data processing. Note that this approach reduces all types of artifacts.
The most simple kind of recombination is to take the arithmetic mean of the various restored images over the common domain, coincident with the size of the restored image with the smallest value of $`K`$. This approach was applied to simulated images, finding that it allows to reduce significantly the restoration error (Bertero et al. (1999)).
If more than two images are available, we have verified that the median usually provides better results. If only two images are available, it seems more convenient to take the smallest value. In case of images containing only compact sources, this recombination removes completely the Type A) artifacts. In case of images containing both compact and extended sources, the bright ghosts over dark background are removed while the dark ghosts over bright background are preserved.
In order to validate this approach, four other images of the Galactic Center (besides that one presented in Section 3.4) were taken in April 1998. The five images have $`7\mathrm{},10\mathrm{},13\mathrm{},17\mathrm{},25\mathrm{}`$ chopping throws, always in the N-S direction, corresponding to $`K=29,38,51,67,95`$. After the inversion, the size of the common region is $`128\times 186`$ pixel. Figure 10 shows the original images with $`7\mathrm{},13\mathrm{},25\mathrm{}`$ chopping throw, as well as the corresponding restored images. The ghosts of the brightest point source can be easily identified. In Figure 11 we show the combination of the 5 reconstructed images using the mean (a) and the median (b) of the stack. As we anticipated, the median image provides the better reconstruction, just marginally affected by Type A) artifacts. Figure 11 shows that only the restoration in the observation region is entirely reliable, whereas the restoration outside strongly depends on the chopping amplitude.
It is certainly possible to get better results by using more refined procedures for recombining the various restored images. In particular, one can observe that the position of the ghosts can be readily foreseen once the reconstructed image is available. For each image it is possible to build a map of the areas possibly affected by ghosts where other images should better be used, and therefore a weight function describing the relative reliability of the various images may be introduced. The investigation of this approach is in progress.
### 4.2 Multiple inversions from the same image
This is a useful method for reducing the Type C) artifacts, i.e. jumps in the brightness distribution of the restored images but it does not modify the Type A) and Type B) artifacts whenever they appear. It is directly suggested by the observation that the positions of the discontinuities are related to the values of $`K_1`$ and $`K`$. Since $`K`$ is fixed, the only way for changing these positions is to change $`K_1`$. This can be obtained by reducing the value of $`N`$, i.e. by removing a certain number of rows from the original chopped and nodded image.
Starting with a chopped and nodded image with columns of length $`N`$ and chopping amplitude $`K`$, by applying the inversion algorithm one obtains a restored image with columns of length $`N+2K`$ and jumps at the rows indicated at the end of Section 3. Removing $`M`$ rows both from the top and from the bottom of the original image a reduced image with columns of length $`N2M`$ is produced. Let us assume $`K_1>2M`$, so that the value of $`q`$ does not change while $`K_1`$ is replaced by $`K_1^{}=K_12M`$. By applying again the inversion method to the reduced image, one obtains a new restored image with columns of length $`N+2K2M`$ and jumps at the rows $`K_1^{},K,K+K_1^{},\mathrm{},K_1^{}+(q+1)K,(q+2)K`$. The pixels of one column of this image, with $`n`$ ranging from 1 to $`N+2K2M`$, correspond to the pixels of the same column of the previously restored image with $`n`$ ranging from $`M+1`$ to $`N+2KM`$, so that the jumps of the second restored image occur at the rows $`K_1M,K+M,K+K_1M,\mathrm{},K_1+(q+1)K+M,(q+2)KM`$ of the first one. In other words, the effect produced by the reduction of the original image is a shift of length $`M`$ towards the top of the restored image, for the jumps of odd order and a shift of length $`M`$ towards the bottom, for the jumps of even order. If the positions of the jumps in the two restored images do not coincide, by taking the arithmetic mean of the two images over the common domain one obtains a reduction of the jumps by a factor of 2. The same result holds true also when $`2M>K_1`$, as it can be deduced by observing that, for the reduced image, $`q`$ is replaced by $`q^{}=q1`$ and $`K_1`$ by $`K_1^{}=K_1+K2M`$.
This procedure can be extended to the case where several reduced images are used in addition to the original one. If the total number of images is $`p`$ and if the rows are removed in such a way that all positions of the jumps in the various restored images are different, then, by taking the arithmetic mean of these images over the common domain (which coincides with that of the shortest restored image) we obtain a reduction of the jumps by a factor of $`p`$. The resulting average image will produce a much better result than that obtained by a single inversion.
Figure 12 shows the result obtained by applying this procedure to the image presented in Figure 9a). The average image is the arithmetic mean of six restored images, one directly obtained from the original image and five obtained by removing respectively 1,2,3,4, and 5 rows from the top and the bottom of the original one. The jumps are considerably attenuated while the multiple ghosts of the bright stars remain unchanged.
### 4.3 Inversion of pasted images
The Type C) artifacts in the restored image do not appear or, at least, are very weak if no significant extended source exists outside the observation region. However, it is in general possible to take images of adjacent regions in order to build a mosaic encompassing the brightest sources of the region. If no significant extended source exists below and above the mosaic, the algorithm can be run on the combined image to obtain a better reconstruction.
We have tested this approach on a couple of images of the giant HII region W51. The observations were carried out with MAX on August 29-30 1997 through the broad N-band filter. The chopping frequency was 2.2 Hz and the chopping throw $`30\mathrm{}`$ in the N-S direction. The integration time was set to 10.2 ms per frame and 40 seconds total on source (Ligori, Robberto & Herbst (1999)).
The two images (Figure 13) are centered respectively on W51 IRS1 (Figure 13b)) and 29 arcseconds north of W51 IRS1 (Figure 13a)). Figure 13b) shows that, due to the large chopping throw, the bright star on top has a negative counterpart close to the bottom of the image. Another negative counterpart is visible nearby, $`8\mathrm{}`$ to the south of IRS 1. From this image alone there is no way to specify if this second source lies above or belove the field. Figure 13a) reveals that it lies above.
The reconstruction of these two images produces the results displayed in Figure 14a) and Figure 14b) respectively. These results are clearly unacceptable. On the other hand, the two images can be combined to form a $`128\times 224`$ pixel mosaic (Figure 13c)). Figure 14c) shows the corresponding reconstruction with dynamic range emphasizing the lowest counts. With respect to Figure 14a,b), the improvement in image is striking: the horizontal discontinuities within the central field have disappeared, and also the other artifacts are significantly reduced. Only discontinuities at the border of the observation region defined by the mosaic are still visible. Note that the “hole” present between IRS1 and the north rim is real, as near-IR data clearly indicate the presence of a dark ridge North of IRS1. Note also the source $`25\mathrm{}`$ to the south of IRS 1, visible in the 20 $`\mu `$m map of Genzel et al. (1982) and correctly reconstructed by our algorithm.
## 5 Comments and Conclusions
We have considered the problem of the reconstruction of astronomical data taken at mid-IR wavelengths in chopping and nodding mode. Studying the mathematical properties of the corresponding under-determined linear system, we have proposed an iterative method for approximating the positive solution of minimal r.m.s. value. We have implemented the algorithm and tested it on astronomical data taken in various observing runs at the UKIRT telescope with the MAX camera. We have investigated the nature of the artifacts affecting the restored images and suggested various computational and observational strategies for their reduction. We find that if an extended source is observed with a few different chopping throws, possibly building a mosaics if the source extends beyond the detector field of view, our algorithm can provide a reliable reconstruction of the source brightness.
We finally remark that on the basis of the MAX experience the proper setup of the camera at the telescope is crucial to get the best results from our reconstruction method. It can be useful in this respect to provide our check-list for the setup of MAX on UKIRT.
Assuming that the chopping-nodding direction has been chosen to be N-S:
1. Select chopping throws close to a common multiple of the instrument pixel size and of the guide/tip-tilt camera pixel size. Use preferably chopping throws larger than 1/6 the array size for best reconstruction (see BBDR99).
2. Align the chopping direction to the detector orientation: point to a bright star at the edge of the field, and make sure that for large chopping throws $`x`$ coordinates of the positive and negative centroids are as close as possible within the same column. The chopping direction will be in general different from the real N-S direction.
3. Align the nodding direction to the chopping direction. If the telescope software implements the nodding as a simple jump of the telescope to an “offset” position, be sure that both the right ascension and declination (or amplitude and position angle) of the offset beam have been given. Note that both values will change with the chopping throw.
4. If the guide camera/cross-head moves when the telescope is pointed to the offset beam, make sure that also for the camera offset both right ascension and declination have been introduced.
There is a last point that deserves some attention as a potential limit to the accuracy of the method: field distortion. Off-axis reflectors are a preferred choice for mid-IR imagers, as they allow a compact and achromatic optical design. However, these systems are in general strongly affected by field distortion. Optical designers often trade field distortion for the ultimate optical quality. For thermal-IR work, it must be kept in mind that is still very difficult to obtain a good astrometric calibration, given also the scarcity of good calibration fields in this wavelengths regime.
Our code is free and is available on request to the authors. Two versions are available, in C and in IDL. For the IDL version, the typical execution time for the inversion of a $`128\times 128`$ image is of the order of a few tens of seconds on a Ultra 10 SPARC station.
## 6 Acknowledgements
This work was partially funded by the Italian CNAA (Consorzio Nazionale per l’Astronomia e l’Astrofisica) under the contract nr. 16/97.
The authors are indebted to S.Ligori for providing the W51 images in advance of publication. M.R. acknowledges Steven Beckwith and Tom Herbst for discussions on this subject during the long nights spent observing on UKIRT.
|
warning/0005/hep-ph0005090.html
|
ar5iv
|
text
|
# References
CONFIRMATOIN OF THE $`\sigma `$-MESON BELOW 1 GEV AND INDICATION FOR THE $`f_0(1500)`$ GLUEBALL
Yu.S. Surovtsev
Bogoliubov Laboratory of Theoretical Physics, Joint Institute for Nuclear Research, Dubna 141 980, Moscow Region, Russia
D. Krupa and M. Nagy
Institute of Physics, Slov.Acad.Sci., Dúbravská cesta 9, 842 28 Bratislava, Slovakia
> On the basis of a simultaneous description of the isoscalar $`s`$-wave channel of the $`\pi \pi `$ scattering (from the threshold up to 1.9 GeV) and of the $`\pi \pi K\overline{K}`$ process (from the threshold to $``$ 1.4 GeV) in the model-independent approach, a confirmation of the $`\sigma `$-meson at $``$ 660 MeV and an indication for the glueball nature of the $`f_0(1500)`$ state are obtained.
A problem of scalar mesons is most troublesome and long-lived in the light meson spectroscopy. Among difficulties in understanding the scalar-isoscalar sector there is the one related to a strong model-dependence of information on multichannel states obtained in analyses based on the specific dynamic models or using an insufficiently-flexible representation of states (e.g., the standard Breit – Wigner form). Earlier, we have shown that an inadequate description of multichannel states gives not only their distorted parameters when analyzing data but also can cause the fictitious states when one neglects important (even energetic-closed) channels. In this paper we are going, conversely, to demostrate that the large background (e.g., that happens in analyzing $`\pi \pi `$ scattering), can hide low-lying states, even such important for theory as a $`\sigma `$-meson . The latter is required by most of the models (like the linear $`\sigma `$-models or the Nambu – Jona-Lasinio models ) for spontaneous breaking of chiral symmetry. Since earlier all the analyses of the $`s`$-wave $`\pi \pi `$ scattering gave a large $`\pi \pi `$-background, it was said that this state (if exists) is ”unobservably”-wide. Recently, new analyses of the old and new experimental data have been performed which give a very wide scalar-isoscalar state in the energy region 500-850 MeV . However, these analyses use either the Breit – Wigner form (even if modified) or specific forms of interactions in a quark model, or in a multichannel approach to the considered processes; therefore, there one cannot talk about a model independence of results. Besides, in these analyses, a large $`\pi \pi `$-background is obtained. We are going to show that a proper detailing of the background (as allowance for the left-hand branch-point) permits us to extract from the latter a very wide (but observable) state below 1 GeV.
An adequate consideration of multichannel states and a model-independent information on them can be obtained on the basis of the first principles (analyticity, unitarity and Lorentz invariance) immediately applied to analyzing experimental data. The way of realization is a consistent allowance for the nearest singularities on all sheets of the Riemann surface of the $`S`$-matrix. The Riemann-surface structure is taken into account by a proper choice of the uniformizing variable. Earlier, we have proposed this method for 2- and 3-channel resonances and developed the concept of standard clusters (poles on the Riemann surface) as a qualitative characteristic of a state and a sufficient condition of its existence as well as a criterion of a quantitative description of the coupled-process amplitudes when all the complifications of the analytic structure due to a finite width of resonances and crossing channels and high-energy “tails” are accumulated in quite a smooth background . Let us stress that for a wide state, the pole position (the pole cluster one for multichannel states) is a more stable characteristic than the mass and width which are strongly dependent on a model. The cluster kind is determined from the analysis of experimental data and is related to the state nature. At all events, we can, in a model-independent manner, discriminate between bound states of particles and the ones of quarks and gluons, qualitatively predetermine the relative strength of coupling of a state with the considered channels, and obtain an indication on its gluonium nature.
In this work, we restrict ourselves to a two-channel approach when considering simultaneously the coupled processes $`\pi \pi \pi \pi ,K\overline{K}`$. Therefore, we have the two-channel $`S`$-matrix determined on the 4-sheeted Riemann surface. The $`S`$-matrix elements $`S_{\alpha \beta }`$, where $`\alpha ,\beta =1(\pi \pi ),2(K\overline{K})`$, have the right-hand (unitary) cuts along the real axis of the $`s`$-variable complex plane, starting at the points $`4m_\pi ^2`$ and $`4m_K^2`$ and extending to $`\mathrm{}`$, and the left-hand cuts, which are related to the crossing-channel contributions and extend along the real axis towards $`\mathrm{}`$ and begin at $`s=0`$ for $`S_{11}`$ and at $`4(m_K^2m_\pi ^2)`$ for $`S_{22}`$ and $`S_{12}`$. We number the Riemann-surface sheets according to the signs of analytic continuations of the channel momenta $`k_1=(s/4m_\pi ^2)^{1/2},k_2=(s/4m_K^2)^{1/2}`$ as follows: signs $`(\text{Im}k_1,\text{Im}k_2)=++,+,,+`$ correspond to the sheets I,II,III,IV.
To elucidate the resonance representation on the Riemann surface, we express analytic continuations of the matrix elements to the unphysical sheets $`S_{\alpha \beta }^L`$ ($`L=II,III,IV`$) in terms of them on the physical sheet $`S_{\alpha \beta }^I`$. Those expressions are convenient for our purpose because, on sheet I (the physical sheet), the matrix elements $`S_{\alpha \beta }^I`$ can have only zeros beyond the real axis. Using the reality property of the analytic functions and the 2-channel unitarity, one can obtain
$`S_{11}^{II}={\displaystyle \frac{1}{S_{11}^I}},S_{11}^{III}={\displaystyle \frac{S_{22}^I}{detS^I}},S_{11}^{IV}={\displaystyle \frac{detS^I}{S_{22}^I}},`$
$`S_{22}^{II}={\displaystyle \frac{detS^I}{S_{11}^I}},S_{22}^{III}={\displaystyle \frac{S_{11}^I}{detS^I}},S_{22}^{IV}={\displaystyle \frac{1}{S_{22}^I}},`$ (1)
$`S_{12}^{II}={\displaystyle \frac{iS_{12}^I}{S_{11}^I}},S_{12}^{III}={\displaystyle \frac{S_{12}^I}{detS^I}},S_{12}^{IV}={\displaystyle \frac{iS_{12}^I}{S_{22}^I}},`$
Here $`detS^I=S_{11}^IS_{22}^I(S_{12}^I)^2`$. In the matrix element, a resonance with the only decay mode is represented by a pair of complex-conjugate poles on sheet II as the nearest singularities and a pair of conjugate zeros on sheet I at the same points of complex energy. In the 2-channel case, the mentioned formulae of analytical continuations immediately give the resonance representation by poles and zeros on the 4-sheeted Riemann surface. One must discriminate between three types of resonances described by a pair of conjugate zeros on sheet I: (a) in $`S_{11}`$, (b) in $`S_{22}`$, (c) in each of $`S_{11}`$ and $`S_{22}`$. A resonance of every type is represented by a pair of complex-conjugate clusters (of poles and zeros on the Riemann surface) of size typical of strong interactions. Thus, we arrive at the notion of three standard pole-clusters which represent two-channel bound states of quarks and gluons. Note that this resonance division into types is not formal. In paricular, the resonance, coupled strongly with the first ($`\pi \pi `$) channel, is described by the pole cluster of type (a); if the resonance is coupled strongly with the $`K\overline{K}`$ and weakly with $`\pi \pi `$ channel (say, if it has a dominant $`s\overline{s}`$ component), then it is represented by the cluster of type (b); finally, since a most noticeable property of a glueball is the flavour-singlet structure of its wave function and, therefore, (except the factor $`\sqrt{2}`$ for a channel with neutral particles) practically equal coupling with all the members of the nonet, then a glueball must be represented by the pole cluster of type (c) as a necessary condition.
Just as in the 1-channel case, the existence of a particle bound-state means the presence of a pole on the real axis under the threshold on the physical sheet, so in the 2-channel case, the existence of a bound state in channel 2 ($`K\overline{K}`$ molecule), which, however, can decay into channel 1 ($`\pi \pi `$ decay), would imply the presence of a pair of complex conjugate poles on sheet II under the threshold of the second channel without an accompaniment of the corresponding shifted pair of poles on sheet III .
In our previous 2-channel analysis of the $`\pi \pi `$ scattering, we have obtained satisfactory description ($`\chi ^2/\text{ndf}1.00`$) with two resonances ($`f_0(975)`$ and $`f_0(1500)`$) and with the large $`\pi \pi `$-background. There, in the uniformizing variable, we have taken into account only the right-hand branch-points at $`s=4m_\pi ^2`$ and $`s=4m_K^2`$.
Now, to take also the left-hand branch-point at $`s=0`$ into account, we use the uniformizing variable
$$v=\frac{m_K\sqrt{s4m_\pi ^2}+m_\pi \sqrt{s4m_K^2}}{\sqrt{s(m_K^2m_\pi ^2)}},$$
(2)
which maps the 4-sheeted Riemann surface onto the $`v`$-plane, divided into two parts by a unit circle centered at the origin. The sheets I (II), III (IV) are mapped onto the exterior (interior) of the unit disk on the upper and lower $`v`$-half-plane, respectively. The physical region extends from the point $`i`$ on the imaginary axis ($`\pi \pi `$ threshold) along the unit circle clockwise in the 1st quadrant to point 1 on the real axis ($`K\overline{K}`$ threshold) and then along the real axis to point $`b=\sqrt{(m_K+m_\pi )/(m_Km_\pi )}`$ into which $`s=\mathrm{}`$ is mapped on the $`v`$-plane. The intervals $`(\mathrm{},b],[b^1,b^1],[b,\mathrm{})`$ on the real axis are the images of the corresponding edges of the left-hand cut of the $`\pi \pi `$-scattering amplitude. The type (a) resonance is represented in $`S_{11}`$ by two pairs of the poles on the images of the sheets II and III, symmetric to each other with respect to the imaginary axis, by zeros, symmetric to these poles with respect to the unit circle.
Note that the variable $`v`$ is uniformizing for the $`\pi \pi `$-scattering amplitude, however, the amplitudes of the $`K\overline{K}`$ scattering and $`\pi \pi K\overline{K}`$ process do have the cuts on the $`v`$-plane, which arise from the left-hand cut on the $`s`$-plane, starting at $`s=4(m_K^2m_\pi ^2)`$. Under conformal mapping (2), this left-hand cut is mapped into cuts which begin at the points $`v=(m_K\sqrt{m_K^22m_\pi ^2}\pm im_\pi )/(m_K^2m_\pi ^2)`$ on the unit circle on the $`v`$-plane, go along it up to the imaginary axis, and occupy the latter. This left-hand cut will be neglected in the Riemann-surface structure, and the contribution on the cut will be taken into account in the $`K\overline{K}`$ background as a pole on the real $`s`$-axis on the physical sheet in the sub-$`K\overline{K}`$-threshold region; on the $`v`$-plane, this pole gives two poles on the unit circle in the upper half-plane, symmetric to each other with respect to the imaginary axis, and two zeros, symmetric to the poles with respect to the real axis, i.e. at describing the process $`\pi \pi K\overline{K}`$, one additional parameter is introduced, say, a position $`p`$ of the zero on the unit circle.
For the simultaneous analysis of experimental data on the coupled processes it is convenient to use the Le Couteur-Newton relations expressing the $`S`$-matrix elements of all coupled processes in terms of the Jost matrix determinant $`d(k_1,k_2)d(s)`$, the real analytic function with the only square-root branch-points at the process thresholds $`k_i=0`$.
On $`v`$-plane the Le Couteur-Newton relations are
$$S_{11}=\frac{d(v^1)}{d(v)},S_{22}=\frac{d(v^1)}{d(v)},S_{11}S_{22}S_{12}^2=\frac{d(v)}{d(v)}$$
(3)
with the $`d`$-function that on the $`v`$-plane already does not possess branch-points is taken as
$$d=d_Bd_{res},$$
(4)
where $`d_B=B_\pi B_K`$; $`B_\pi `$ contains the possible remaining $`\pi \pi `$-background contribution, related to exchanges in crossing channels; $`B_K`$ is that part of the $`K\overline{K}`$ background which does not contribute to the $`\pi \pi `$-scattering amplitude
$$B_K=v^1(1pv)(1+p^{}v).$$
(5)
The function $`d_{res}(v)`$ represents the contribution of resonances, described by one of three types of the pole-zero clusters, i.e., except for the point $`v=0`$, it consists of zeros of clusters:
$$d_{res}=v^M\underset{n=1}{\overset{M}{}}(1v_n^{}v)(1+v_nv),$$
(6)
where $`n`$ runs over the independent zeros; therefore, for resonances of the types (a) and (b), $`n`$ has two values, for the type (c), four values; $`M`$ is the number of pairs of the conjugate zeros.
On the basis of these formulas, we analyze simultaneously the available experimental data on the $`\pi \pi `$-scattering and the process $`\pi \pi K\overline{K}`$ in the channel with $`I^GJ^{PC}=0^+0^{++}`$.
To obtain the satisfactory description ($`\chi ^2/\text{ndf}2.2`$) of the $`s`$-wave $`\pi \pi `$ scattering from the threshold to 1.89 GeV and $`|S_{12}|`$ from the threshold to 1.4 GeV (where the 2-channel unitarity is valid), we have taken $`B_\pi =1`$ in eq.(4), and three multichannel resonances turned out to be sufficient: the two ones of the type (a) ($`f_0(660)`$ and $`f_0(980)`$) and $`f_0(1500)`$ of the type (c). A satisfactory description of the phase shift of the $`\pi \pi K\overline{K}`$ matrix element is obtained to 1.5 GeV with the value of the parameter $`p=0.9936130.112842i`$ (this corresponds to the position of the pole on the $`s`$-plane at $`s=4(m_K^2m_\pi ^2)0.06`$).
In the table, the obtained parameter values of poles on the corresponding sheets of the Riemann surface are cited on the complex energy plane ($`\sqrt{s_r}=\mathrm{E}_ri\mathrm{\Gamma }_r/2`$). We stress that these are not masses and widths of resonances. Since, for wide resonances, values of masses and widths are very model-dependent, it is reasonable to report characteristics of pole clusters which must be rather stable for various models.
Let us indicate the constant values of the obtained-state couplings with the $`\pi \pi `$ and $`K\overline{K}`$ systems calculated through the residues of amplitudes at the pole on sheet II. Taking the resonance part of the amplitude in the form $`T_{ij}^{res}=_rg_{ir}g_{rj}D_r^1(s)`$, where $`D_r(s)`$ is an inverse propagator ($`D_r(s)ss_r`$), we obtain (we denote the coupling constants with the $`\pi \pi `$ and $`K\overline{K}`$ systems through $`g_\pi `$ and $`g_K`$, respectively) for $`f_0(660)`$: $`g_\pi =0.7376\pm 0.12`$ GeV and $`g_K=0.37\pm 0.1`$ GeV, for $`f_0(980)`$: $`g_\pi =0.158\pm 0.03`$ GeV and $`g_K=0.86\pm 0.09`$ GeV, for $`f_0(1500)`$: $`g_\pi =0.347\pm 0.028`$ GeV. In this 2-channel approach, there is no point in calculating the coupling constant of the $`f_0(1500)`$ state with the $`K\overline{K}`$ system, because the 2-channel unitarity is valid only to 1.4 GeV, and, above this energy, there is a considerable disagreement between the calculation of the amplitude modulus $`S_{12}`$ and the experimental data.
Let us indicate also scattering lengths calculated in our approach. For the $`K\overline{K}`$ scattering, we obtain $`a_0^0(K\overline{K})=0.932\pm 0.11+(0.706\pm 0.09)i,m_{\pi ^+}^1.`$ A presence of the imaginary part in $`a_0^0(K\overline{K})`$ reflects the fact, that already at the threshold of the $`K\overline{K}`$ scattering, other channels ($`2\pi ,4\pi `$ etc.) are opened.
For the $`\pi \pi `$ scattering, we obtain: $`a_0^0=0.27\pm 0.06,m_{\pi ^+}^1.`$ Compare with results of some other works both theoretical and experimental: the value $`0.26\pm 0.05`$ (L. Rosselet et al.), obtained in the analysis of the decay $`K\pi \pi e\nu `$ with using Roy’s model; $`0.24\pm 0.09`$ (A.A. Bel’kov et al. from analysis of the process $`\pi ^{}p\pi ^+\pi ^{}n`$ with using the effective range formula; $`0.23`$ (S. Ishida et al.\[Ishida\], modified approach to analysis of $`\pi \pi `$ scattering with using Breit-Wigner forms; $`0.16`$ (S. Weinberg , current algebra (non-linear $`\sigma `$-model)); $`0.20`$ (J. Gasser, H. Leutwyler , the theory with the non-linear realization of chiral symmetry); $`0.26`$ (M.K. Volkov , the theory with the linear realization of chiral symmetry).
So, an existence of the low-lying state with the properties of the $`\sigma `$-meson and the obtained value of the $`\pi \pi `$-scattering length seem to suggest the linear realization of chiral symmetry.
Here, a parameterless description of the $`\pi \pi `$ background in the channel with $`I^GJ^{PC}=0^+0^{++}`$ is first given.
The $`f_0(1500)`$ state is represented by the pole cluster on the Riemann surface of the $`S`$-matrix which corresponds to a flavour singlet, e.g. the glueball.
Finally, a minimum scenario of the simultaneous description of the processes $`\pi \pi \pi \pi ,K\overline{K}`$ in the channel with $`I^GJ^{PC}=0^+0^{++}`$ does not require the $`f_0(1370)`$ resonance; therefore, if this meson exists, it must be weakly coupled with the $`\pi \pi `$ channel, e.g. be the $`s\overline{s}`$ state (as to that assignment of the $`f_0(1370)`$ resonance, we agree with the work ).
This work has been supported by the Grant Program of Plenipotentiary of Slovak Republic at JINR. Yu.S. and M.N. were supported in part by the Slovak Scientific Grant Agency, Grant VEGA No. 2/7175/20; and D.K., by Grant VEGA No. 2/5085/99.
|
warning/0005/hep-ph0005267.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
In our recent papers,, we proposed the democratic-type mass matrix which contains six real parameters and found that this mass matrix predicts
$`\theta _{23}={\displaystyle \frac{\pi }{4}},\delta ={\displaystyle \frac{\pi }{2}},`$ (1)
where $`\theta _{23}`$ and $`\delta `$ are the mixing angle between the mass eigenstates, $`\nu _2`$ and $`\nu _3`$, and the CP violation phase, in the parameterization of the mixing matrix given in the particle data group (see the matrix given in the Appendix A).
If we take the CHOOZ bound, $`|V_{13}|<0.16`$ or $`|\mathrm{sin}\theta _{13}|<0.16`$, we find almost maximum atmospheric neutrino mixing,
$`\mathrm{sin}^22\theta _{atm}=4|V_{23}|^2(1|V_{23}|^2)=1\mathrm{sin}^4\theta _{13}>0.999,`$ (2)
where $`V`$ is the neutrino mixing matrix. If the experimental data turns out to show that $`\mathrm{sin}^22\theta _{atm}`$ is really close to unity, our model will become a good candidate. Another special feature of the model is the prediction of the value of the CP violation phase. Both Dirac CP phase ($`\delta `$) and Majorana CP phases are predicted. In particular, the maximal value of the CP violation phase $`\delta `$ is predicted. Our prediction gives the great encouragement for experiments to measure the CP violation in the oscillation processes in the near future. The theoretical study has become an urgent topic.
In Ref.2, we made a further investigation on the democratic-type mass matrix. We constructed $`Z_3`$ invariant Lagrangian with two or three up-type Higgs doublets and derived the democratic-type mass matrix by using the see-saw mechanism. We also considered one up-type Higgs model. By considering the $`Z_3`$ symmetric Lagrangian together with the $`Z_2`$ invariant $`Z_3`$ breaking terms, we found the further prediction,
$`|\mathrm{tan}\theta _{12}|=\sqrt{23\mathrm{sin}^2\theta _{13}},`$ (3)
which we shall explain in the next section. By using the CHOOZ bound, this relation leads to
$`0.87<\mathrm{sin}^22\theta _{sol}=4|V_{11}|^2|V_{12}|^2<{\displaystyle \frac{8}{9}}.`$ (4)
In Refs.1 and 2, we assumed that the above predictions are valid at the weak scale $`m_Z`$, although the neutrino mass matrix is assumed to be defined at the right-handed neutrino mass scale $`M_R`$. The stability of mixing angles under the change of energy scale has been discussed\[7-10\]. According to their result, in many occasions, the predictions at $`m_Z`$ are essentially the same as those at $`M_R`$. In some special cases where $`m_1m_2`$, the prediction of $`\mathrm{sin}^22\theta _{sol}`$ becomes unstable. That is, the predicted large value of $`\mathrm{sin}^22\theta _{sol}`$ at $`M_R`$ becomes the small value at $`m_Z`$.
The purpose of this paper is to examine the stability of our predictions. In particular, we are interested in the possibility that the large solar neutrino mixing at $`M_R`$ becomes small to be consistent with the small angle MSW solution at $`m_Z`$. We found that the angle can become small, but unfortunately this possibility does not realize the small angle MSW solution.
In Sec.2, we briefly explain our model. In Sec.3, we analytically examine the renormalization effect on the neutrino mass matrix and the effect to our predictions. The numerical analysis to supplement the discussions in Sec.3 is given in Sec.4. In Sec.5, the summary is given.
## 2 The model
We consider the following dimension five Lagrangian in the mass eigenstate basis of charged leptons
$`_Y=(m_1^0+\stackrel{~}{m}_1)\overline{(\mathrm{\Psi }_1)^C}\mathrm{\Psi }_1{\displaystyle \frac{H_uH_u}{u_u^2}}2\stackrel{~}{m}_1\overline{(\mathrm{\Psi }_2)^C}\mathrm{\Psi }_3{\displaystyle \frac{H_uH_u}{u_u^2}},`$ (5)
where $`\stackrel{~}{m}_1`$ and $`m_1^0`$ are real parameters, and $`u_u`$ is the vacuum expectation value of the neutral component of the doublet Higgs $`H_u`$. This Lagrangian is invariant under the $`Z_3`$ transformation
$`\mathrm{\Psi }_1\omega \mathrm{\Psi }_1,\mathrm{\Psi }_2\omega ^2\mathrm{\Psi }_2,\mathrm{\Psi }_3\mathrm{\Psi }_3,H_u\omega ^2H_u,`$ (6)
where, the irreducible representation $`\mathrm{\Psi }_i`$ $`(i=1,2,3)`$ are defined by
$`\mathrm{\Psi }_1`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{3}}}(\mathrm{}_e+\omega ^2\mathrm{}_\mu +\omega \mathrm{}_\tau ),`$
$`\mathrm{\Psi }_2`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{3}}}(\mathrm{}_e+\omega \mathrm{}_\mu +\omega ^2\mathrm{}_\tau ),`$
$`\mathrm{\Psi }_3`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{3}}}(\mathrm{}_e+\mathrm{}_\mu +\mathrm{}_\tau ).`$ (7)
The $`Z_3`$ transformation for $`\mathrm{\Psi }_i`$ is induced by the cyclic permutation among $`\mathrm{}_i`$, which are the left-handed lepton doublets defined by $`\mathrm{}_e=(\nu _{eL},e_L)^T`$ and so on.
Then, we introduce the $`Z_3`$ symmetry breaking term, but it preserves the $`Z_2`$ symmetry
$`\mathrm{\Psi }_1\mathrm{\Psi }_1,`$ (8)
and all other fields are unchanged. Now, we find
$`_{SB}=m_2^0\overline{(\mathrm{\Psi }_2)^C}\mathrm{\Psi }_2{\displaystyle \frac{H_uH_u}{u_u^2}}m_3^0\overline{(\mathrm{\Psi }_3)^C}\mathrm{\Psi }_3{\displaystyle \frac{H_uH_u}{u_u^2}}.`$ (9)
After $`H_u`$ acquires the vacuum expectation value, the neutrino mass term is derived. In the flavor basis, $`(\nu _e,\nu _\mu ,\nu _\tau )`$, the mass matrix is given in a democratic-type form,
$`m_\nu (M_R)`$ $`=`$ $`{\displaystyle \frac{m_1^0}{3}}\left(\begin{array}{ccc}1& \omega ^2& \omega \\ \omega ^2& \omega & 1\\ \omega & 1& \omega ^2\end{array}\right)+\stackrel{~}{m}_1\left(\begin{array}{ccc}1& 0& 0\\ 0& \omega & 0\\ 0& 0& \omega ^2\end{array}\right)`$ (10)
$`+{\displaystyle \frac{m_2^0}{3}}\left(\begin{array}{ccc}1& \omega & \omega ^2\\ \omega & \omega ^2& 1\\ \omega ^2& 1& \omega \end{array}\right)+{\displaystyle \frac{m_3^0}{3}}\left(\begin{array}{ccc}1& 1& 1\\ 1& 1& 1\\ 1& 1& 1\end{array}\right),`$
where $`\omega `$ is the element of $`Z_3`$ symmetry and we take $`\omega =\mathrm{exp}(i2\pi /3)`$, i.e., $`\omega ^3=1`$. We consider that this mass matrix is given at the right-handed mass scale $`M_R`$.
The unitary matrix $`V_2`$ which diagonalizes $`m_\nu (M_R)`$ is derived in the Appendix A and the result is
$`V_2=\left(\begin{array}{ccc}1& 0& 0\\ 0& \omega & 0\\ 0& 0& \omega ^2\end{array}\right)\left(\begin{array}{ccc}\frac{1}{\sqrt{3}}& \sqrt{\frac{2}{3}}c^{}& i\sqrt{\frac{2}{3}}s^{}\\ \frac{1}{\sqrt{3}}& \frac{1}{\sqrt{6}}(c^{}+i\sqrt{3}s^{})& \frac{1}{\sqrt{6}}(\sqrt{3}c^{}+is^{})\\ \frac{1}{\sqrt{3}}& \frac{1}{\sqrt{6}}(c^{}i\sqrt{3}s^{})& \frac{1}{\sqrt{6}}(\sqrt{3}c^{}is^{})\end{array}\right)\left(\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 0& 0& i\end{array}\right),`$ (11)
where $`c^{}=\mathrm{cos}\theta ^{}`$ and $`s^{}=\mathrm{sin}\theta ^{}`$ and
$`\mathrm{tan}\theta ^{}={\displaystyle \frac{\mathrm{\Delta }_{}}{\stackrel{~}{m}_1+\sqrt{\stackrel{~}{m}_1^2+\mathrm{\Delta }_{}^2}}},`$ (12)
with $`\mathrm{\Delta }_{}=(m_3^0m_2^0)/2`$.
It should be noted that predictions in Eqs.(1) and (3) are derived from $`V_2`$. The phase in $`\mathrm{diag}(1,1,i)`$ represents the Majorana phase, while the phases in $`\mathrm{diag}(1,\omega ,\omega ^2)`$ are the irrelevant phases which are absorbed by the charged lepton fields.
From our later analysis, there is essentially no effect to $`V_{13}`$. As a result, we can impose the CHOOZ bound, $`|V_{13}|<0.16`$ at $`m_Z`$. We find
$`|s^{}|<0.2.`$ (13)
We define the mass eigenstate neutrinos at $`M_R`$ as $`(\nu _1^R,\nu _2^R,\nu _3^R)`$ and their masses are
$`m_1^R`$ $`=`$ $`m_1^0+\stackrel{~}{m}_1,`$
$`m_2^R`$ $`=`$ $`m_2^0+\mathrm{\Delta }_{}+\sqrt{\stackrel{~}{m}_1^2+\mathrm{\Delta }_{}^2},`$
$`m_3^R`$ $`=`$ $`m_2^0+\mathrm{\Delta }_{}\sqrt{\stackrel{~}{m}_1^2+\mathrm{\Delta }_{}^2}.`$ (14)
We take the convention, $`\stackrel{~}{m}_1>0`$. Since $`m_2^0`$ and $`m_3^0`$ are parameters for the symmetry breaking terms, we expect that $`\stackrel{~}{m}_1>>|m_2^0|`$, $`|m_3^0|`$. Then, we find $`m_2^R>0`$ and $`m_3^R<0`$. The parameter $`\stackrel{~}{m}_1`$ controls the overall normalization neutrino masses, and $`m_1^0`$ and $`m_2^0`$ (or $`m_3^0`$) control the mass of $`m_1^R`$ and the mass splitting between $`m_2^R`$ and $`m_3^R`$, while the parameter $`\mathrm{\Delta }_{}=(m_3^0m_2^0)/2`$ does the size of $`V_{13}`$.
## 3 The renormalization group analysis
We consider the renormalization group effect on the dimension five interaction in Eqs.(5) and (9) in the MSSM model. The general feature of the stability of mixing angles has been extensively discussed\[7-10\]. Here, we take the special mass matrix, the democratic-type mass matrix and examine the stability in detail.
(3-1) Neutrino mass matrix at $`m_Z`$
In the basis where charged lepton mass matrix is diagonal and thus the Yukawa coupling matrix which induces masses of charged leptons is diagonal, the neutrino mass matrices at $`M_R`$ and $`m_Z`$ are related as
$`m_\nu (M_R)A\left(\begin{array}{ccc}1& 0& 0\\ 0& \sqrt{\frac{I_\mu }{I_e}}& 0\\ 0& 0& \sqrt{\frac{I_\tau }{I_e}}\end{array}\right)m_\nu (m_Z)\left(\begin{array}{ccc}1& 0& 0\\ 0& \sqrt{\frac{I_\mu }{I_e}}& 0\\ 0& 0& \sqrt{\frac{I_\tau }{I_e}}\end{array}\right),`$ (15)
where
$`I_i=\mathrm{exp}\left({\displaystyle \frac{1}{8\pi ^2}}{\displaystyle _{\mathrm{ln}(m_z)}^{\mathrm{ln}(M_R)}}y_i^2𝑑t\right)(i=e,\mu ,\tau ),`$ (16)
with the Yukawa coupling for charged leptons $`y_i`$ and $`A=(I_e/I_\tau )(m_{\nu 33}(M_R)/m_{\nu 33}(m_Z))`$.
After absorbing $`A`$ into the overall normalization of parameters in $`m_\nu (M_R)`$ and by using the approximation,
$`\sqrt{{\displaystyle \frac{I_\mu }{I_\tau }}}\sqrt{{\displaystyle \frac{I_e}{I_\tau }}}{\displaystyle \frac{1}{\sqrt{I_\tau }}},`$ (17)
we obtain
$`m_\nu (m_Z)=\left(\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 0& 0& \alpha \end{array}\right)m_\nu (M_R)\left(\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 0& 0& \alpha \end{array}\right),`$ (18)
where
$`\alpha 1/\sqrt{I_\tau }=\left({\displaystyle \frac{m_Z}{M_R}}\right)^{\frac{1}{8\pi ^2}(1+\mathrm{tan}^2\beta )(m_\tau /v)^2}<1.`$ (19)
Here we neglect the radiative correction on $`y_\tau `$, and $`m_\tau `$ is the $`\tau `$ lepton mass, $`v^2=u_u^2+u_d^2`$ and $`\mathrm{tan}\beta =u_u/u_d`$ with $`u_i`$ being the vacuum expectation value of MSSM Higgs doublet $`<H_i>(i=u,d)`$.
Now we define the small parameter $`ϵ=1\alpha `$. In order to estimate the value of $`ϵ`$, we consider the right-handed mass scale $`M_R`$ and the region of $`\mathrm{tan}\beta `$ as
$`M_R=10^{13}(\mathrm{GeV}),2<\mathrm{tan}\beta <60.`$ (20)
Then, with $`m_Z=91.187(\mathrm{GeV})`$, $`m_\tau =1.777(\mathrm{GeV})`$ and $`v=245.4(\mathrm{GeV})`$, we find
$`8\times 10^5<ϵ<6\times 10^2.`$ (21)
(3-2) The diagonalization
By transforming $`m_\nu (m_Z)`$ in Eq.(18) by $`V_2`$, we find
$`\stackrel{~}{m}_\nu `$ $``$ $`V_2^T\left(\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 0& 0& \alpha \end{array}\right)V_2^{}\left(V_2^Tm_\nu (M_R)V_2\right)V_2^{}\left(\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 0& 0& \alpha \end{array}\right)V_2`$ (22)
$`=`$ $`(1ϵS^T)D_\nu ^R(1ϵS),`$
where $`D_\nu ^R=V_2^Tm_\nu (M_R)V_2=\mathrm{diag}(m_1^R,m_2^R,m_3^R)`$ and
$`S=V_2^{}\left(\begin{array}{ccc}0& 0& 0\\ 0& 0& 0\\ 0& 0& 1\end{array}\right)V_2={\displaystyle \frac{1}{3}}\left(\begin{array}{ccc}1& \frac{1}{\sqrt{2}}ae^{i\varphi _1}& i\sqrt{\frac{3}{2}}be^{i\varphi _2}\\ \frac{1}{\sqrt{2}}ae^{i\varphi _1}& \frac{1}{2}a^2& i\frac{\sqrt{3}}{2}abe^{i(\varphi _1\varphi _2)}\\ i\sqrt{\frac{3}{2}}be^{i\varphi _2}& i\frac{\sqrt{3}}{2}abe^{i(\varphi _1\varphi _2)}& \frac{3}{2}b^2\end{array}\right),`$
where $`s^{}`$ and $`c^{}`$ are given in Eq.(12), and $`a`$, $`b`$ and phases $`\varphi _i`$ are defined by
$`a=\sqrt{1+2s^2},`$ $`b=\sqrt{1{\displaystyle \frac{2}{3}}s^2},`$
$`\mathrm{tan}\varphi _1=\sqrt{3}\mathrm{tan}\theta ^{},`$ $`\mathrm{tan}\varphi _2={\displaystyle \frac{1}{\sqrt{3}}}\mathrm{tan}\theta ^{}.`$ (24)
By keeping $`ϵ`$ up to the first order, we find
$`\stackrel{~}{m}_\nu \left(\begin{array}{ccc}(1\frac{2}{3}ϵ)m_1^R& \frac{1}{3\sqrt{2}}ϵa(m_1^Re^{i\varphi _1}+m_2^Re^{i\varphi _1})& iϵp\\ \frac{1}{3\sqrt{2}}ϵa(m_1^Re^{i\varphi _1}+m_2^Re^{i\varphi _1})& (1\frac{1}{3}a^2ϵ)m_2^R& iϵq\\ iϵp& iϵq& (1b^2ϵ)m_3^R\end{array}\right),`$ (25)
where
$`p`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{6}}}b(m_1^Re^{i\varphi _2}m_3^Re^{i\varphi _2}),`$
$`q`$ $`=`$ $`{\displaystyle \frac{1}{2\sqrt{3}}}ab(m_2^Re^{i(\varphi _1\varphi _2)}m_3^Re^{i(\varphi _1\varphi _2)}).`$ (26)
(3-3) The general discussion on the stability
Hereafter, we do not discuss the fully degenerate case, $`|m_1^R||m_2^R||m_3^R|`$, because this case is quite unstable and it is hard to have the definite predictions. Therefore, we focus our discussions on hierarchical cases; (a) $`|m_3^R|>>|m_2^R|>>|m_1^R|`$ or $`|m_3^R|>>|m_1^R|>>|m_2^R|`$ and (b) $`|m_1^R||m_2^R|>>|m_3^R|`$ or $`|m_3^R|>>|m_2^R||m_1^R|`$.
The case (a): The fully hierarchical case
With the use of the analogy of the analysis by Haba et al., we expect that all mixing angles and the CP violation phase are essentially unchanged by the scale change from $`M_R`$ to $`m_Z`$. This may be simply understood by the consideration that the see-saw mechanism can be used to evaluate the mixings and the neutrino masses, and thus the effect is suppressed by the order of $`ϵ`$. We checked this result by the numerical computations also.
The case (b): The hierarchical case with $`|m_1^R||m_2^R|`$
The situation is slightly complicated in comparison with the case (a), because of the degeneracy $`|m_1^R||m_2^R|`$. Firstly, we notice that the off diagonal terms are much small than $`(\stackrel{~}{m}_\nu )_{33}`$, or $`|(\stackrel{~}{m}_\nu )_{11}||(\stackrel{~}{m}_\nu )_{22}|`$. Therefore, we can use the see-saw calculation between $`(\nu _1^R,\nu _2^R)`$ and $`\nu _3^R`$, where $`\nu _i^R`$ are mass eigenstates at $`M_R`$. That is, we can safely neglect the contributions from $`p`$ and $`q`$ terms in the matrix and thus we do not need to consider the mixing between $`(\nu _1^R,\nu _2^R)`$ and $`\nu _3^R`$.
Now, the element $`V_{i3}`$ and $`V_{3i}`$ ($`i=1,2,3`$) at $`M_R`$ is still valid at $`m_Z`$. That is, $`V_{i3}=(V_2)_{i3}`$ and $`V_{3i}=(V_2)_{3i}`$. As a result, the prediction of $`\mathrm{sin}^22\theta _{atm}>0.999`$ in Eq.(2) and the CHOOZ constraint, $`|s_{13}|<0.16`$ are valid at $`m_Z`$.
The situation changes depending on the relative sign between $`m_1^R`$ and $`m_2^R`$.
(b-1) The case where $`m_1^R<0`$ and $`m_2^R>0`$
We denote the submatrix for $`(\nu _1^R,\nu _2^R)`$ as $`\stackrel{~}{m}_\nu ^{}`$ with the approximation $`a1`$ because $`s^2<0.04`$ is small,
$`\stackrel{~}{m}_\nu ^{}`$ $``$ $`\left(\begin{array}{cc}(1+\mathrm{\Delta }\frac{2}{3}ϵ)& i\frac{\sqrt{2}}{3}ϵ\mathrm{sin}\varphi _1\\ i\frac{\sqrt{2}}{3}ϵ\mathrm{sin}\varphi _1& 1\frac{1}{3}ϵ\end{array}\right)m_2^R`$ (27)
$`=`$ $`\left(\begin{array}{cc}1& 0\\ 0& i\end{array}\right)\left(\begin{array}{cc}(1+\mathrm{\Delta }\frac{2}{3}ϵ)& \frac{\sqrt{2}}{3}ϵ\mathrm{sin}\varphi _1\\ \frac{\sqrt{2}}{3}ϵ\mathrm{sin}\varphi _1& (1\frac{1}{3}ϵ)\end{array}\right)m_2^R\left(\begin{array}{cc}1& 0\\ 0& i\end{array}\right),`$
where we defined
$`\mathrm{\Delta }={\displaystyle \frac{|m_1^R|m_2^R}{m_2^R}}.`$ (28)
The matrix $`\stackrel{~}{m}_\nu ^{}`$ is diagonalized by
$`\left(\begin{array}{cc}1& 0\\ 0& i\end{array}\right)\left(\begin{array}{cc}c& s\\ s& c\end{array}\right),`$ (29)
where $`c=\mathrm{cos}\theta `$ and $`s=\mathrm{sin}\theta `$ and
$`\mathrm{tan}\theta `$ $`=`$ $`{\displaystyle \frac{\pm \frac{2\sqrt{2}}{3}ϵ\mathrm{sin}\varphi _1}{\frac{1}{3}ϵ\mathrm{\Delta }+\sqrt{(\frac{1}{3}ϵ\mathrm{\Delta })^2+\frac{8}{9}ϵ^2\mathrm{sin}^2\varphi _1}}},`$
$`||m_1|m_2|`$ $`=`$ $`m_2^R\sqrt{({\displaystyle \frac{1}{3}}ϵ\mathrm{\Delta })^2+{\displaystyle \frac{8}{9}}ϵ^2\mathrm{sin}^2\varphi _1},`$ (30)
and $`m_2m_2^R`$.
The mixing matrix at $`m_Z`$ is now obtained by multiplying this matrix to $`V_2`$ in Eq.(11). By looking at the structure of $`V_2`$, we find
$`V_{11}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{3}}}(ci\sqrt{2}c^{}s),`$
$`V_{12}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{3}}}(si\sqrt{2}c^{}c),`$ (31)
aside from the irrelevant phases. By neglecting the small $`s^2<0.04`$, we have $`c^{}=1`$ and thus we find
$`\mathrm{sin}^22\theta _{sol}{\displaystyle \frac{8}{9}}(1+s^2)(1{\displaystyle \frac{s^2}{2}}),`$ (32)
which takes a value from 8/9 to 1 independent of the mixing angle $`\theta `$. This is due to the phase matrix $`\mathrm{diag}(1,i)`$ in Eq.(29).
By the transformation of the matrix in Eq.(29), the CP violation phase $`\delta `$ changes, due to the phase matrix $`\mathrm{diag}(1,i)`$. The effect is examined by considering the Jarlskog parameter which takes the value as
$`|J_{CP}||\mathrm{Im}(V_{11}V_{12}^{}V_{21}^{}V_{22})|={\displaystyle \frac{1}{3\sqrt{3}}}|s^{}c^{}(c^2s^2)|,`$ (33)
and we find
$`|\mathrm{sin}\delta |={\displaystyle \frac{|\mathrm{cos}2\theta |}{\sqrt{1+\frac{1}{8}\mathrm{sin}^22\theta \left(\frac{\mathrm{cos}2\theta ^{}}{\mathrm{cos}\theta ^{}}\right)^2}}}.`$ (34)
It should be noted that $`\theta =0`$ at $`M_R`$ so that $`|\mathrm{sin}\delta |=1`$. Now we examine the value at $`m_Z`$ from Eq.(34). The angle $`\theta `$ depends on $`\mathrm{\Delta }`$ which is defined in Eq.(28), as we can see in Eq.(30). For $`\mathrm{\Delta }>>ϵ`$ or $`\mathrm{\Delta }<0`$, $`|\mathrm{tan}\theta |>>1`$ or $`|\mathrm{tan}\theta |<<1`$. Therefore, $`|\mathrm{sin}\delta |1`$ is realized from Eq.(34). In special cases where $`\mathrm{\Delta }ϵ/3`$, $`\mathrm{sin}\delta `$ can become small at $`m_Z`$. In particular, for $`\mathrm{\Delta }=ϵ/3`$, we find $`\mathrm{tan}\theta =\pm 1`$ and thus we find $`\mathrm{sin}\delta =0`$.
Finally, we find
$`\mathrm{\Delta }_{sol}^2=|m_2^2m_1^2|2m_2^2\sqrt{({\displaystyle \frac{1}{3}}ϵ\mathrm{\Delta })^2+{\displaystyle \frac{8}{9}}ϵ^2\mathrm{sin}^2\varphi _1},`$ (35)
which depends on $`m_2`$ and $`\mathrm{\Delta }`$. Therefore, we can reproduce all three mass squared differences for the large angle MSW, the LOW mass and the Just so (Vacuum) solutions. For example, when $`|\mathrm{\Delta }|>>ϵ`$, we find $`\mathrm{\Delta }_{sol}^22m_2^2\mathrm{\Delta }(\mathrm{\Delta }_{sol}^2)_{M_R}`$, where the value at $`M_R`$, $`(\mathrm{\Delta }_{sol}^2)_{M_R}`$ is a free parameter that we can choose as an input.
(b-2) The case where $`m_1^R>0`$ and $`m_2^R>0`$
In order to simplify the calculation and to see the essence of the analysis, we neglect the term $`s^2<0.04`$. Thus we take $`a=b=1`$ and $`\mathrm{cos}\varphi _1=1`$. Then, the submatrix relevant to $`\nu _1^R`$ and $`\nu _2^R`$ is given by
$`\stackrel{~}{m}_\nu ^{}\left(\begin{array}{cc}(1+\mathrm{\Delta }\frac{2}{3}ϵ)& \frac{\sqrt{2}}{3}ϵ\\ \frac{\sqrt{2}}{3}ϵ& 1\frac{1}{3}ϵ\end{array}\right)m_2^R.`$ (36)
After the diagonalization, we find
$`m_1`$ $`=`$ $`\left(1+{\displaystyle \frac{\mathrm{\Delta }}{2}}{\displaystyle \frac{1}{2}}ϵ+\mathrm{sign}(\mathrm{\Delta }){\displaystyle \frac{1}{2}}\sqrt{D}\right)m_2^R,`$
$`m_2`$ $`=`$ $`\left(1+{\displaystyle \frac{\mathrm{\Delta }}{2}}{\displaystyle \frac{1}{2}}ϵ\mathrm{sign}(\mathrm{\Delta }){\displaystyle \frac{1}{2}}\sqrt{D}\right)m_2^R,`$ (37)
where $`\mathrm{sign}(\mathrm{\Delta })`$ takes 1 for $`\mathrm{\Delta }>0`$ and $`1`$ for $`\mathrm{\Delta }<0`$ and
$`D=({\displaystyle \frac{1}{3}}ϵ\mathrm{\Delta })^2+{\displaystyle \frac{8}{9}}ϵ^2.`$ (38)
The mass of the third one is $`m_3=(1ϵ)m_3^R`$. The mixing matrix is
$`{\displaystyle \frac{1}{N}}\left(\begin{array}{cc}\mathrm{sign}(\mathrm{\Delta })\sqrt{D}(\frac{1}{3}ϵ\mathrm{\Delta })& \frac{2\sqrt{2}}{3}ϵ\\ \frac{2\sqrt{2}}{3}ϵ& \mathrm{sign}(\mathrm{\Delta })\sqrt{D}(\frac{1}{3}ϵ\mathrm{\Delta })\end{array}\right),`$ (39)
where $`N`$ is the normalization factor.
Now we multiply the above matrix to $`V_2`$. Aside from the unimportant phase and by taking $`c^{}1`$, we find
$`V_{11}`$ $``$ $`{\displaystyle \frac{1}{\sqrt{3}N}}\left\{\mathrm{sign}(\mathrm{\Delta })\sqrt{D}({\displaystyle \frac{1}{3}}ϵ\mathrm{\Delta })+{\displaystyle \frac{4}{3}}ϵ\right\},`$
$`V_{12}`$ $``$ $`{\displaystyle \frac{1}{\sqrt{3}N}}\left\{{\displaystyle \frac{2\sqrt{2}}{3}}ϵ+\sqrt{2}\left[\mathrm{sign}(\mathrm{\Delta })\sqrt{D}({\displaystyle \frac{1}{3}}ϵ\mathrm{\Delta })\right]\right\}.`$ (40)
Now we find
$`\mathrm{sin}^22\theta _{sol}={\displaystyle \frac{8}{9}}\left[{\displaystyle \frac{\left(\mathrm{sign}(\mathrm{\Delta })\sqrt{D}+\mathrm{\Delta }\right)^2ϵ^2}{\left(\mathrm{sign}(\mathrm{\Delta })\sqrt{D}+\mathrm{\Delta }\frac{ϵ}{3}\right)^2+\frac{8}{9}ϵ^2}}\right]^2.`$ (41)
Firstly, since the mass matrix in Eq.(36) is real matrix, the CP violation phase $`\delta `$ are stable and takes $`\delta =\pi /2`$ at $`m_Z`$. Needless to say, the atmospheric neutrino mixing and $`s_{13}`$ are stable.
(i) The stable $`\mathrm{sin}^22\theta _{sol}`$
We focus on the solar neutrino mixing. From Eq.(41), we see that if $`|\mathrm{\Delta }|>>ϵ`$, $`\mathrm{sin}^22\theta _{sol}8/9`$. For $`\mathrm{\Delta }>0`$, this condition is relaxed to the condition $`\mathrm{\Delta }>3ϵ/2`$, where $`\mathrm{sin}^22\theta _{sol}8/9`$ is realized.
(ii) The unstable $`\mathrm{sin}^22\theta _{sol}`$
Now we consider the case where $`\mathrm{sin}^22\theta _{sol}`$ becomes small at $`m_Z`$. We observe that $`\mathrm{sin}^22\theta _{sol}0`$ as $`\mathrm{\Delta }0`$. This implies that $`\mathrm{sin}^22\theta _{sol}`$ becomes small for $`\mathrm{\Delta }<<ϵ`$, while it remains large value for $`\mathrm{\Delta }>ϵ`$.
Below, we examine the case $`\mathrm{\Delta }<<ϵ`$ to see the $`\mathrm{\Delta }`$ dependence of $`\mathrm{sin}^22\theta _{sol}`$ in detail. We expand $`\mathrm{sin}^22\theta _{sol}`$ in terms of $`\mathrm{\Delta }/ϵ`$. We obtain
$`\mathrm{sin}^22\theta _{sol}{\displaystyle \frac{8}{9}}\left({\displaystyle \frac{\mathrm{\Delta }}{ϵ}}\right)^2.`$ (42)
The small angle which is consistent with the angle for the small angle MSW solution, $`\mathrm{sin}^22\theta _{sol}10^2`$, is realized if we take $`|\mathrm{\Delta }|\frac{1}{10}ϵ`$. Next we examine the sign of $`(m_2^2m_1^2)\mathrm{cos}2\theta _{sol}`$. For $`\mathrm{\Delta }>0`$, we find $`|V_{11}|>>|V_{12}|`$ at $`m_Z`$, i.e., $`\mathrm{cos}2\theta >0`$, from Eq.(40). Then, as we can see from Eq.(37) with $`m_2^Rm_2`$,
$`m_1m_2=m_2\sqrt{({\displaystyle \frac{1}{3}}ϵ\mathrm{\Delta })^2+{\displaystyle \frac{8}{9}}ϵ^2}>0,`$ (43)
which means $`m_2^2m_1^2<0`$. Therefore we obtain $`(m_2^2m_1^2)\mathrm{cos}2\theta _{sol}<0`$. The same conclusion holds for $`\mathrm{\Delta }<0`$ where $`|V_{11}|<<|V_{12}|`$ at $`m_Z`$ ($`\mathrm{cos}2\theta <0`$). That is, in both cases, we find $`(m_2^2m_1^2)\mathrm{cos}2\theta _{sol}<0`$. It is the standard lore that the small angle MSW solution is realized when $`(m_2^2m_1^2)\mathrm{cos}2\theta _{sol}>0`$, which is in conflict with our result. Recently, Gouvêa, Friedland and Murayama have examined the case $`\mathrm{cos}2\theta <0`$ (dark side) for $`m_2^2m_1^2>0`$. They found that the region $`\mathrm{cos}2\theta 0.2`$ is still possible to explain the solar neutrino problem. However, this case corresponds to the large mixing case, $`\mathrm{sin}^22\theta 0.96`$ which is not our case. In conclusion, when $`\mathrm{\Delta }\frac{1}{10}ϵ`$, $`\mathrm{sin}^22\theta _{sol}0.01`$ can be realized, but in this case the MSW mechanism does not work. Therefore, this case is not applicable to solve the solar neutrino problem.
## 4 Examples -Numerical analysis-
Since it is hard to search all parameter regions, we set $`m_2^0=0`$ and then varied other parameters, $`\stackrel{~}{m}_1`$, $`m_1^0`$ and $`m_3^0`$. Here, we exhibit two examples, one for the stable case where the large angle MSW solution for the solar neutrino mixing is realized at $`m_Z`$ and Dirac CP phase $`\mathrm{sin}\delta `$ remain the maximal value, and the other for the case where $`\mathrm{sin}\delta `$ becomes to be small at $`m_Z`$.
* An example for the stable case
As an example, we adopted input values, $`(\stackrel{~}{m}_1,m_1^0,m_3^0)=(0.0699,0.0117,0.025)[\mathrm{eV}]`$ which give neutrino masses at $`M_R`$ as $`(m_1^R,m_2^R,m_3^R)=(0.058200,0.058509,0.083509)[\mathrm{eV}]`$. The values of observables at $`M_R`$ and at $`m_Z`$ are given in Table 1, for various values of $`\mathrm{tan}\beta `$.
Among various parameters, the parameters relevant to atmospheric neutrino mixings, $`\mathrm{\Delta }_{atm}^2`$ and $`\mathrm{sin}^22\theta _{atm}`$, $`\mathrm{sin}\theta _{13}`$ and $`\mathrm{sin}\delta `$ are almost unchanged against the energy scale change for various values of $`\mathrm{tan}\beta `$. The scale dependence for $`\mathrm{\Delta }_{sol}^2`$ and $`\mathrm{sin}^22\theta _{sol}`$ depend on the values of $`\mathrm{tan}\beta `$. From the data,
$`0.5<`$ $`\mathrm{sin}^22\theta _{sol}`$ $`<1,`$
$`1\times 10^5[\text{eV}^2]<`$ $`\mathrm{\Delta }_{sol}^2`$ $`<1\times 10^4[\text{eV}^2],`$ (44)
we have the restriction on $`\mathrm{tan}\beta `$,
$`\mathrm{tan}\beta =313,`$ (45)
which we can see from Table 1.
* An example to give a small Dirac CP phase at $`m_Z`$
We took input values, $`(\stackrel{~}{m}_1,m_1^0,m_3^0)=(0.3,0.59651,0.007)[\mathrm{eV}]`$ where neutrino masses at $`M_R`$ are $`(m_1^R,m_2^R,m_3^R)=(0.29651,0.29652,0.30352)[\mathrm{eV}]`$. We show the values of observables at $`M_R`$ and at $`m_Z`$ in Table 2, for various values of $`\mathrm{tan}\beta `$.
As we can see from Table 2, $`\mathrm{\Delta }_{atm}^2`$, $`\mathrm{sin}^22\theta _{atm}`$ and $`\mathrm{sin}^22\theta _{sol}`$ are almost unchanged. On the other hand, $`\mathrm{sin}\theta _{13}`$, $`\mathrm{\Delta }_{sol}^2`$ and $`\mathrm{sin}\delta `$ change depending on $`\mathrm{tan}\beta `$. In particular, $`\mathrm{sin}\delta `$ does not change much for small $`\mathrm{tan}\beta `$, while changes substantially for large $`\mathrm{tan}\beta `$. This result is consistent with the discussion given for the case $`m_1^Rm_2^R<0`$ and $`\mathrm{\Delta }ϵ/3`$.
In Fig.1 and Fig.2, we show the energy scale dependence of $`m_i^2(i=1,2)`$ and $`\mathrm{sin}\delta `$ for $`\mathrm{tan}\beta =4`$ and 10, for the parameter set in Table 2. From Fig.1, we see that $`\mathrm{\Delta }_{sol}^2`$ increases as the energy scale becomes small and also as $`\mathrm{tan}\beta `$ becomes large. In Fig.2, we see that $`\mathrm{sin}\delta `$ decreases for both $`\mathrm{tan}\beta =4`$ and 10. However, much faster decrease is observed for the larger $`\mathrm{tan}\beta `$.
## 5 Summary and discussions
We considered the stability of the predictions by some special democratic-type neutrino mass matrix, which has the quite interesting intrinsic predictions as given in Eqs.(1) and (3). We assumed that this mass matrix is derived at the right-handed mass scale $`M_R`$ by the see-saw mechanism, and then considered the mass matrix at the weak scale $`m_Z`$ and its predictions by using the renormalization group.
We summarize the result as follows:
* The case (a): The fully hierarchical case
This is the case where the neutrino masses at $`M_R`$ are either $`|m_3^R|>>|m_1^R|>>|m_2^R|`$ or $`|m_3^R|>>|m_2^R|>>|m_3^R|`$. In this case, all predictions are stable and the predictions at $`M_R`$ are valid at $`m_Z`$.
* The case (b): The hierarchical case with $`|m_1^R||m_2^R|`$
If $`m_1^Rm_2^R<0`$, $`\mathrm{sin}^22\theta _{atm}`$ and $`\mathrm{sin}^22\theta _{sol}`$ are stable. The CP violation phase $`\mathrm{sin}\delta `$ is also stable for $`\mathrm{\Delta }>>ϵ`$ or $`\mathrm{\Delta }<0`$. For $`\mathrm{\Delta }ϵ/3`$, $`\mathrm{sin}\delta `$ becomes unstable.
If $`m_1^Rm_2^R>0`$, $`\mathrm{sin}^22\theta _{atm}`$ and the CP violation phase $`\delta `$ are stable. The solar mixing angle $`\mathrm{sin}^22\theta _{sol}`$ is also stable for $`|\mathrm{\Delta }|>>ϵ`$. For $`|\mathrm{\Delta }|<ϵ`$, $`\mathrm{sin}^22\theta _{sol}`$ becomes unstable. In particular, for $`\mathrm{\Delta }ϵ/10`$, $`\mathrm{sin}^22\theta _{sol}`$ at $`m_Z`$ becomes small enough to be consistent with the mixing angle for the small angle MSW solution. However, this case does not realize the small angle MSW solution.
Our model based on the $`Z_3`$ symmetry gives quite special predictions as given in Eqs.(1) and (3). We emphasize that our matrix is intrinsically complex matrix and contains the CP violation phase. In particular, our model predicts the maximal CP violation phase, which is in contrast to most of works where the real neutrino mass matrices are treated so that the prediction for the CP violation phase is out of reach. The prediction for the CP violation phase in the neutrino mass matrix will become a quite important topic in view of the near future projects to observe the neutrino oscillations, for example, in the neutrino factory.
It is our belief that $`Z_3`$ symmetry is not only useful for describing the neutrino mass matrix, but also for the quark mass matrix. The work in this direction will be interesting, because we would like to embed the $`Z_3`$ symmetry in the grand unification scheme.
Acknowledgment This work is supported in part by the Japanese Grant-in-Aid for Scientific Research of Ministry of Education, Science, Sports and Culture, No.12047218.
Appendix A: Detailed derivations
(a) The standard parameterization of the mixing matrix
The particle data group defines the mixing matrix as
$`V_{SF}=\left(\begin{array}{ccc}c_{12}c_{13}& s_{12}c_{13}& s_{13}e^{i\delta }\\ s_{12}c_{23}c_{12}s_{23}s_{13}e^{i\delta }& c_{12}c_{23}s_{12}s_{23}s_{13}e^{i\delta }& s_{23}c_{13}\\ s_{12}s_{23}c_{12}c_{23}s_{13}e^{i\delta }& c_{12}s_{23}s_{12}c_{23}s_{13}e^{i\delta }& c_{23}c_{13}\end{array}\right).`$ (A.1)
(b) Diagonalization of $`m_\nu (M_R)`$ in Eq.(10)
Here, we diagonalize the neutrino mass matrix at $`M_R`$ and thus the predictions are given at $`M_R`$. In order to clarify the property of the democratic-type mass matrix, we first transform $`m_\nu (M_R)`$ by the trimaximal matrix $`V_T`$
$`V_T={\displaystyle \frac{1}{\sqrt{3}}}\left(\begin{array}{ccc}1& 1& 1\\ \omega & \omega ^2& 1\\ \omega ^2& \omega & 1\end{array}\right),`$ (A.2)
where $`\omega =e^{i2\pi /3}`$ ($`\omega ^3=1`$) and the result is
$`V_T^Tm_\nu (M_R)V_T=\left(\begin{array}{ccc}m_1^0+\stackrel{~}{m}_1& 0& 0\\ 0& m_2^0& \stackrel{~}{m}_1\\ 0& \stackrel{~}{m}_1& m_3^0\end{array}\right).`$ (A.3)
Then, we transform further by
$`O_1=\left(\begin{array}{ccc}1& 0& 0\\ 0& \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\\ 0& \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\end{array}\right),`$ (A.4)
and we find
$`(V_TO_1)^Tm_\nu (M_R)V_TO_1=\left(\begin{array}{ccc}m_1^0+\stackrel{~}{m}_1& 0& 0\\ 0& \stackrel{~}{m}_1+m_2^0+\mathrm{\Delta }_{}& \mathrm{\Delta }_{}\\ 0& \mathrm{\Delta }_{}& \stackrel{~}{m}_1+m_2^0+\mathrm{\Delta }_{}\end{array}\right).`$ (A.5)
The matrix $`V_1=V_TO_1`$ is explicitly given by
$`V_1=\left(\begin{array}{ccc}1& 0& 0\\ 0& \omega & 0\\ 0& 0& \omega ^2\end{array}\right)\left(\begin{array}{ccc}\frac{1}{\sqrt{3}}& \sqrt{\frac{2}{3}}& 0\\ \frac{1}{\sqrt{3}}& \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{2}}\\ \frac{1}{\sqrt{3}}& \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{2}}\end{array}\right)\left(\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 0& 0& i\end{array}\right).`$ (A.6)
We have to transform further by the orthogonal matrix $`O_2`$
$`O_2=\left(\begin{array}{ccc}1& 0& 0\\ 0& c^{}& s^{}\\ 0& s^{}& c^{}\end{array}\right),`$ (A.7)
where $`s^{}`$ and $`c^{}`$ are defined by Eq.(12). Now the mixing matrix is given by $`V=V_TO_1O_2`$ which is given in Eq.(11).
Below, we give some special cases. (b-1) The $`m_3^0=m_2^0`$ case
We have $`s^{}=0`$ and $`c^{}=1`$ and the mixing matrix is now $`V=V_1`$. Then, the model predicts
$`\mathrm{sin}^22\theta _{atm}=1,\mathrm{sin}^22\theta _{sol}={\displaystyle \frac{8}{9}}.`$ (A.8)
There is no CP violation Dirac phase.
(b-2) The $`m_2^0=0`$ case
The angle $`\theta ^{}`$ is determined by the ratio of $`m_2`$ and $`m_3`$, and we have
$`\mathrm{sin}^22\theta _{sol}={\displaystyle \frac{4}{9}}{\displaystyle \frac{\beta +2}{\beta }},\mathrm{sin}^22\theta _{atm}={\displaystyle \frac{4}{9}}{\displaystyle \frac{(\beta +1)(2\beta 1)}{\beta ^2}},`$ (A.9)
where $`\beta =\sqrt{|m_2/m_3|}+\sqrt{|m_3/m_2|}2`$. If $`\beta `$ is close to 2, we have the large solar neutrino mixing and also the large atmospheric neutrino mixing.
|
warning/0005/quant-ph0005081.html
|
ar5iv
|
text
|
# I INTRODUCTION
## I INTRODUCTION
Thermal motion of radiating atoms leads, owing to the Doppler effect, to an isotropic broadening of the luminescence line. At the same time, the line width of Rayleigh scattering depends on the direction
$$\mathrm{\Delta }\omega =\omega \frac{\overline{v}}{c}2\mathrm{sin}\frac{\theta }{2},\overline{v}=\sqrt{2kT/m},$$
and vanishes in the case of forward scattering ($`\theta =0`$). This difference in the manifestation of the Doppler effect is due entirely to the deference between the frequency-correlation properties of the indicated processes . On the other hand, the frequency-correlation properties are strongly pronounced also in other two-photon processes (two-quantum absorption and luminescence, or Raman scattering). It is thereforenatural to expect the existence of anisotropy of the Doppler line width in this case, too. This question is discussed in Sec. 3.
We note now that an analysis of radiative processes within the framework of second order perturbation-theory leads to a delimitation of two-photon processes proper from stepwise (or cascade) processes, for example two-photon luminescence and cascade emission of two photons with a real intermediate state. Such a delimitation is based essentially on the frequency-correlation properties. On the other hand, if the energy of interaction of the atom with the field is larger than the level width, then the frequency-correlation properties of the radiative processes experience a strong metamorphosis (Sec. 2). In particular, two-photon and stepwise processes turn out to be physically Indistinguishable. As a consequence, in strong fields the manifestation of the Doppler broadening also changes strongly. The resultant phenomena are traced for the resonant-scattering doublet and resonant-fluorescence triplet (Secs. 4 and 5).
## II FREQUENCY – CORRELATION PROPERTIES OF RADIATIVE PROCESSES
We shall consider the radiation of an atom situated in an external field with two monochromatic components of frequencies $`\omega `$ and $`\omega _\mu `$ and amplitudes $`E`$ and $`E_\mu `$. We assume that each field component interacts only with one transition (resonance approximation). In the scheme corresponding to Raman scattering (Fig. 1), a photon $`\mathrm{}\omega `$ is absorbed and a photon $`\mathrm{}\omega _\mu `$ is emitted. The probability amplitudes $`a_i`$ of the states $`i=m,n,l`$ satisfy the system of equations
$$\dot{a}_m+\gamma _ma_m=iGe^{i\mathrm{\Omega }t}a_n+iG_\mu e^{i\mathrm{\Omega }_\mu t}a_l,$$
$$\dot{a}_n+\gamma _na_n=iG^{}e^{i\mathrm{\Omega }t}a_m,\dot{a}_l+\gamma _la_l=iG_\mu ^{}e^{i\mathrm{\Omega }_\mu t}a_m,$$
$`(2.1)`$
$$\mathrm{\Omega }=\omega \omega _{mn},\mathrm{\Omega }_\mu =\omega _\mu \omega _{ml},G=d_{mn}E/2\mathrm{},G_\mu =d_{ml}E_\mu /2\mathrm{},$$
where $`d_{ij}`$ are the matrix elements of the dipole moment.
We shall henceforth regard $`Ge^{i\mathrm{\Omega }t}`$ as a strong perturbation and $`G_\mu e^{i\mathrm{\Omega }_\mu t}`$ as a weak perturbation. We are interested in the probability $`w_\mu `$ of emission of the photon $`\mathrm{}\omega _\mu `$. The solution of the system (2.1) can be obtained by successive approximations in the parameter $`G_\mu `$ \[3-5\]: we consider the system of equations
$$\dot{a}_m+\gamma _ma_m=iGe^{i\mathrm{\Omega }t}a_n,\dot{a}_n+\gamma _na_n=iG^{}e^{i\mathrm{\Omega }t}a_m,$$
$`(2.2)`$
in the zeroth approximation and introduce Its exact solution into the right-hand side of the equation for $`a_l(t)`$ in (2.1). Integration of this equation yields the first approximation $`G_\mu `$ for $`a_l(t)`$, with the aid of which we calculate
$$w_\mu =2\gamma _l_0^{\mathrm{}}|a_l(t)|^2𝑑t.$$
The solution of the system (2.2) can be represented in the form
$$a_n(t)=[A_1e^{\alpha _1t}+A_2e^{\alpha _2t}]e^{i\mathrm{\Omega }t},$$
$$a_m(t)=iG\left[\frac{A_1e^{\alpha _1t}}{\gamma _m\alpha _1}+\frac{A_2e^{\alpha _2t}}{\gamma _m\alpha _2}\right],$$
$`(2.3)`$
$$\alpha _{1,2}=\frac{\mathrm{\Gamma }+i\mathrm{\Omega }}{2}\pm i\sqrt{G^2+\left(\frac{\mathrm{\Omega }i\gamma }{2}\right)^2},\mathrm{\Gamma }=\gamma _m+\gamma _n,\gamma =\gamma _n\gamma _m,$$
$`(2.4)`$
where $`A_{1,2}`$ are the integration constants. In the case of interest to us, $`a_n(0)=1`$, we have
$$A_{1,2}=\frac{\gamma _m\alpha _{1,2}}{\alpha _1\alpha _2},a_m(t)=\frac{iG}{\alpha _1\alpha _2}[e^{\alpha _1t}e^{\alpha _2t}],$$
$`(2.5)`$
$$w_\mu =\frac{2|GG_\mu |^2}{|\alpha _1\alpha _2|^2}\mathrm{Re}\{\frac{(\alpha _1+\alpha _1^{})^1(\alpha _2+\alpha _1^{})^1}{\gamma _l+\alpha _1^{}+i\mathrm{\Omega }_\mu }+$$
$$+\frac{(\alpha _2+\alpha _2^{})^1(\alpha _1+\alpha _2^{})^1}{\gamma _l+\alpha _2^{}+i\mathrm{\Omega }_\mu }\}.$$
$`(2.6)`$
The frequency-correlation properties consist in the fact that the frequencies $`\mathrm{\Omega }_\mu `$ at which there is maximum probability of emitting the photon $`\mathrm{}\omega _\mu `$ ($`\mathrm{\Omega }_{\mu 1}=\mathrm{Im}\alpha _1,\mathrm{\Omega }_{\mu 2}=\mathrm{Im}\alpha _2`$) turn out to depend on $`\mathrm{\Omega }`$, i.e., on the frequency of the absorbed photon. In the case of small $`G`$, we have in place of (2.4) and (2.6)
$$\alpha _1=\gamma _m,\alpha _2=\gamma _n+i\mathrm{\Omega },G|\mathrm{\Omega }i\gamma |$$
$$w_\mu =\frac{|GG_\mu |^2}{|\gamma _n\gamma _m+i\mathrm{\Omega }|^2}\times $$
$$\times \mathrm{Re}\left\{\frac{\gamma _m^12(\gamma _m+\gamma _n+i\mathrm{\Omega })^1}{\gamma _l+\gamma _m+i\mathrm{\Omega }_\mu }+\frac{\gamma _n^12(\gamma _m+\gamma _ni\mathrm{\Omega })^1}{\gamma _l+\gamma _n+i(\mathrm{\Omega }_\mu \mathrm{\Omega })}\right\}.$$
$`(2.7)`$
The second term in the expression for $`w_\mu `$ reaches a maximum at a frequency $`\omega _\mu =\omega \omega _{ln}`$ (Raman scattering). It can therefore be said that in the second stage of the two-process the atom ”remembers” which quantum was absorbed during the first stage. The Raman-scattering line width $`\gamma _l+\gamma _n`$ also ”remembers” from which level the atom arrived at the first stage. These indeed are the properties of frequency correlation. To the contrary, the first term in (2.7) gives resonance at $`\mathrm{\Omega }_\mu =\omega _\mu \omega _{ml}=0`$, i.e., at the frequency of the transition between the intermediate and final states of the atom. The width of the corresponding line is also determined by the levels $`m`$ and $`l`$. This term describes the cascade or stepwise transition $`nml`$, and there is no correlation in it at all between the absorption and emission acts.
The correlation properties of the emission processes are closely connected with the type of evolution of amplitude $`a_m(t)`$ of the intermediate state. Let $`|\mathrm{\Omega }|\gamma _m,\gamma _n`$; then the rapidly oscillating term $`\mathrm{exp}[(i\mathrm{\Omega }+\gamma _n)t]`$ (virtual state) carries information concerning the initial state ($`\gamma _n`$) and the absorbed quantum ($`\mathrm{\Omega }`$), and causes the appearance of a scattering line, i.e., the second term in (2.7) (the terms $`2(\gamma _m+\gamma _n\pm i\mathrm{\Omega })^1`$ can be discarded). The time dependence of the second term, $`exp(\gamma _mt)`$, contains no attributes of the absorption act and does not differ from the case when the state $`m`$ is the initial state (i.e., $`a_m(0)=1`$). We can therefore say with respect to this term and with respect to the corresponding unshifted line at the transition $`ml`$ that the intermediate state is a real state of the atom, having a finite lifetime $`(2\gamma _m)^1`$.
A separate examination of the transitions through the virtual and real states signifies that only squares of the moduli of the first and second terms remain in the expression for $`|a_l(t)|^2`$. The crossing term lead to the appearance of $`2(\gamma _m+\gamma _n\pm i\mathrm{\Omega })^1`$ in $`w_\mu `$, and it is legitimate to neglect them if $`|\mathrm{\Omega }|\gamma _m,\gamma _n`$. In the case when $`|\mathrm{\Omega }|\gamma _m,\gamma _n`$, the crossing terms, which reflect the interference between the real and virtual states, are significant and cannot be discarded. However, even here $`w_\mu `$ can be represented in the form of a sum of terms with and without ”memory”, and allowance for the interference only changes the coefficient preceding these terms. The only physical basis for contrasting the stepwise and two-photon processes is the difference between their frequency-correlation properties, which are uniquely connected with the singularities of the evolutions of the individual terms of the amplitude $`a_m(t)`$ of the intermediate state. On the other hand, formulas (2.7) are valid within the framework of second-order perturbation theory, and are no longer valid at sufficiently large $`G^2`$. In the general case, $`a_m`$ contains two exponential terms (formula (2.5)), which are formally analogous to the ”virtual” and ”real” states. However both $`\alpha _1`$ and $`\alpha _1`$ depend on the parameters of the field $`(G^2,\mathrm{\Omega })`$ and of the two combining levels $`(\gamma _m,\gamma _n)`$. With respect to $`w_\mu `$, this means that in both resonances the atom ”remembers” which quantum was observed during the first stage of the process. In the limiting case of a very strong field we have
$$\alpha _{1,2}[\gamma _m+\gamma _n+i\mathrm{\Omega }]/2\pm iG,G|\mathrm{\Omega }i\gamma |,$$
$`(2.8)`$
i.e., the differences in the temporal properties of the two exponentials in (2.5) have disappeared completely. At the same time, the differences in the frequency-correlation properties of the corresponding lines have also disappeared, and there are no grounds for distinguishing between the two. The concepts of stepwise and two-photon transitions or of the virtual and real states are likewise physically Indistinguishable. It is clear from the foregoing that these concepts are inseparably linked with perturbation theory and lose physical meaning outside the region of its applicability.
The external field levels out the differences between $`\alpha _1`$ and $`\alpha _2`$ both with respect to the yield of the resonance $`(\mathrm{\Omega })`$ and with respect to the difference In the damping of the states $`m,n`$. Let us consider in greater detail the case $`\gamma _m=\gamma _n=\mathrm{\Gamma }`$, when the leveling function of the field simplifies:
$$\alpha _{1,2}=\mathrm{\Gamma }+i\alpha _{1,2}^{^{\prime \prime }},\alpha _{1,2}^{^{\prime \prime }}=\frac{1}{2}[\mathrm{\Omega }\pm \sqrt{\mathrm{\Omega }^2+4G^2}].$$
$`(2.9)`$
Figure 2 shows plots of $`\alpha _{1,2}^{^{\prime \prime }}`$ as functions of $`\mathrm{\Omega }`$. The asymptotic approach of the plots to the abscissa axis and to the dashed line $`\alpha ^{^{\prime \prime }}=\mathrm{\Omega }`$ corresponds respectively to real and virtual states. As $`\mathrm{\Omega }`$ changes from positive to negative values, we have for $`\alpha _1^{^{\prime \prime }}`$, for example, a smooth transition from the properties of the virtual state to the properties of the real state, and for $`\alpha _2^{^{\prime \prime }}`$ the inverse sequence. In the region $`|\mathrm{\Omega }|<G`$, the states $`\alpha _1`$ and $`\alpha _2`$ differ little.
As the measure of the ”memory” of the absorbed quantum we can choose the quantity
$$M_{1,2}\frac{d\alpha _{1,2}^{^{\prime \prime }}}{d\mathrm{\Omega }}=\frac{1}{2}\left[1\pm \frac{\mathrm{\Omega }}{\sqrt{\mathrm{\Omega }^2+4G^2}}\right],$$
$`(2.10)`$
which varies from 0 to 1. The value $`M=0`$ means complete absence of memory (stepwise transition), while $`M=1`$ corresponds to total correlation between the frequencies of the absorbed and emitted photons (two-photon transition). Since $`\alpha _1+\alpha _2=\mathrm{\Gamma }+i\mathrm{\Omega }`$, it follows that $`M_1+M_2=1`$; this fact can be interpreted as follows: the intermediate state as a whole, without subdivision into two states, retains the entire information on the absorbed photon. In the case $`|\mathrm{\Omega }|G`$, we have $`M_{1,2}1/2`$, i.e., the ”memory” is equally divided between the two terms in $`a_m(t)`$.
If $`\alpha _1^{^{\prime \prime }}\alpha _2^{^{\prime \prime }}`$ is sufficiently small, the interference of the states $`\alpha _1`$ and $`\alpha _1`$ is quite significant. A separate analysis of the transitions through these intermediate states is meaningful if the distances between the resonances exceed their widths:
$$|\alpha _1^{^{\prime \prime }}\alpha _2^{^{\prime \prime }}|\alpha _1^{^{}}+\alpha _2^{^{}}+2\gamma _l.$$
Under these conditions, the emission spectrum for the transition $`ml`$ has the form of a well resolved doublet, the appearance of which can be interpreted as the splitting of the intermediate-state level into two sublevels \[3-9\]. In accordance with the foregoing, It is meaningless to attribute the components of this doublet to stepwise or two-photon transitions. One can only speak of the resonance-scattering doublet as a unit. We shall henceforth use this term. For concreteness, we have referred throughout to processes of the type of Raman scattering. All the physical conclusions pertain also to other processes in which two-photons take part, such as two-quantum luminescence, two-quantum absorption, and Raman scattering via a lower intermediate level. It is only necessary to reverse the signs of $`\mathrm{\Omega }`$ and $`\mathrm{\Omega }_\mu `$ in all the formulas, depending on whether the corresponding guantum is emitted or absorbed.
## III DOPPLER BROADENING OF RAMAN SCATTERING LINE
Allowance for the motion of the atom in the case of traveling waves reduces, as is well known, to the substitutions $`\mathrm{\Omega }\mathrm{\Omega }𝐤𝐯`$ and $`\mathrm{\Omega }_\mu \mathrm{\Omega }_\mu 𝐤_\mu 𝐯`$, where $`𝐤`$ and $`𝐤_\mu `$ are the wave vectors of the waves. This means that the amplitudes $`A_1`$ and $`A_2`$ also depend on the velocity of the atom. Within the framework of the second approximation of perturbation theory, this pertains only to the virtual sublevel. This case apparently has not been discussed in the literature, and will be considered in the present section. Averaging over the velocities will be carried with a Maxwellian distribution:
$$(\pi \overline{\mathrm{v}})^{3/2}\mathrm{exp}\{𝐯^2/\overline{\mathrm{v}}^2\}.$$
$`(3.1)`$
Let the deviation from resonance be larger not only than the natural width but also of the Doppler width $`(|\mathrm{\Omega }|𝐤\overline{𝐯})`$. Under this condition we can neglect the interference between the real and virtual states, and we can easily obtain from (2.7)
$$w_\mu =\frac{|GG_\mu |^2}{\mathrm{\Omega }^2}\mathrm{Re}\left\{\frac{\sqrt{\pi }}{\gamma _mk_\mu \overline{\mathrm{v}}}e^{p_1^2}[1\mathrm{\Phi }(p_1)]+\frac{\sqrt{\pi }}{\gamma _nq\overline{\mathrm{v}}}e^{p_2^2}[1+\mathrm{\Phi }(p_2)]\right\},$$
$`(3.2)`$
$$p_1=[\gamma _l+\gamma _m+i\mathrm{\Omega }_\mu ]/k_\mu \overline{\mathrm{v}},p_2=[\gamma _l+\gamma _n+i(\mathrm{\Omega }_\mu \mathrm{\Omega })]/q\overline{\mathrm{v}},$$
$$q=|𝐤_\mu 𝐤|=\sqrt{(k_\mu k)^2+4k_\mu k\mathrm{sin}^2(\theta /2)},$$
where $`\mathrm{\Phi }(z)`$ is the probability integral and $`\theta `$ is the angle between $`𝐤`$ and $`𝐤_\mu `$. If the Doppler broadening dominates over the natural broadening, $`k_\mu \overline{\mathrm{v}}\gamma _l+\gamma _m,q\overline{\mathrm{v}}\gamma _l+\gamma _n`$, then we can assume that $`\mathrm{\Phi }(p_{1,2})=0`$ and (3.2) contains two terms of Gaussian form
$$w_\mu =\frac{|GG_\mu |^2}{\mathrm{\Omega }^2}\{\frac{\sqrt{\pi }}{\gamma _mk_\mu \overline{\mathrm{v}}}\mathrm{exp}[\frac{\mathrm{\Omega }_\mu ^2}{(k_\mu \overline{\mathrm{v}})^2}]+$$
$$+\frac{\sqrt{\pi }}{\gamma _nq\overline{\mathrm{v}}}\mathrm{exp}[\frac{(\mathrm{\Omega }_\mu \mathrm{\Omega })^2}{(q\overline{\mathrm{v}})^2}]\}.$$
$`(3.3)`$
The first terms in (3.2) or (3.3) (stepwise-transition line) has a Doppler width $`k_\mu \overline{\mathrm{v}}`$, which does not depend on $`\theta `$. On the other hand, the Doppler width $`q\overline{\mathrm{v}}`$ of the Raman-scattering line depends strongly on the observation direction, changing from $`|k_\mu k|\overline{\mathrm{v}}`$ to $`(k_\mu +k)\overline{\mathrm{v}}`$ when $`\theta `$ changes from zero to $`\pi `$ (Fig. 3a). If $`|k_\mu k|\overline{\mathrm{v}}\gamma _\mathrm{l}+\gamma _\mathrm{n}`$, then the Raman scattering has a Lorentz shape in the angle interval $`\theta \gamma _l+\gamma _n)/k_\mu \overline{\mathrm{v}}`$
$$\frac{(\gamma _l+\gamma _n)/\gamma _n}{(\gamma _l+\gamma _n)^2+(\mathrm{\Omega }_\mu +\mathrm{\Omega })^2}$$
and its width is determined by the natural damping of the initial and final states. On the other hand, for the direction $`\theta =\pi `$, the width $`(2k\overline{\mathrm{v}})`$ is twice the width of the stepwise-transition line.
It is easy to show that the doublet-component intensities integrated with respect to $`\mathrm{\Omega }_\mu `$ do not depend on $`\theta `$. Consequently, the width anisotropy means also an angular dependence of the ratio of the intensities at the maxima of the lines in the range $`k_\mu \gamma _m/|k_\mu k|\gamma _n`$.
Formula (3.2) for $`q\overline{\mathrm{v}}`$ is analogous to the expression for the line width of the Rayleigh scattering in a gas, $`2k\overline{\mathrm{v}}\mathrm{sin}(\theta /2)`$ , which is obtained from (3.2) when $`k=k_\mu `$. Just as in the case of Rayleigh scattering, formulas (3.2) and (3.3) admit of a simple interpretation, if we consider Raman scattering as emission of a classical oscillator moving with velocity $`\overline{\mathrm{v}}`$. A change-over to the c.m.s. of the oscillator changes the frequency $`\omega `$ of the external field by $`\omega 𝐤𝐯`$. The forced oscillations induced by the field also have a frequency $`\omega 𝐤𝐯`$. The internal motion in the atom (or in the molecule) with natural frequency $`\omega _{ln}`$ modulates the forced oscillation and leads to the appearance in the emission spectrum of a component of frequency $`\omega 𝐤𝐯\omega _{ln}`$. Finally, for the wave emitted in the $`𝐤_\mu `$ direction, the inverse transition to the stationary system of coordinates yields a frequency $`\omega \omega _{ln}(𝐤_\mu 𝐤)𝐯`$, and averaging over $`𝐯`$ leads to a Doppler width $`|𝐤_\mu 𝐤|\overline{\mathrm{v}}=q\overline{\mathrm{v}}`$.
## IV DOPPLER BROADENING OF RESONANT – SCATTERING DOUBLET
Let us turn now to the strong-field problem and assume that $`Gk\overline{\mathrm{v}}`$. Here, too, we can neglect the ”interference” terms $`(\alpha _1+\alpha _2^{})^1`$ and $`(\alpha _2+\alpha _1^{})^1`$ in the expression (2.6) for $`w_\mu `$, and the Doppler shifts of $`\alpha _{1,2}`$ can be taken into account in the first nonvanishing approximation:
$$\alpha _j=\alpha _{j0}iM_j\mathrm{𝐤𝐯},j=1,2,$$
$`(4.1)`$
where $`\alpha _{j0}`$ are the values of $`a_j`$ at $`𝐤𝐯=0`$ and $`M_j`$ is the ”memory” factor determined by formula (2.10). Neglecting the difference between $`\alpha _j`$ and $`\alpha _{j0}`$ everywhere except in the resonant denominators, we get from (2.6)
$$w_\mu =\frac{|GG_\mu |^2}{\mathrm{\Omega }^2+4G^2}\mathrm{Re}\frac{\gamma _m^1}{\gamma _l+\alpha _{10}^{}+i[\mathrm{\Omega }_\mu (𝐤_\mu M_1𝐤)𝐯]}\times $$
$$\times \frac{\gamma _n^1}{\gamma _l+\alpha _{20}^{}+i[\mathrm{\Omega }_\mu (𝐤_\mu M_2𝐤)𝐯]}.$$
$`(4.2)`$
Averaging of this expression leads to a formula of the type (3.2), in which the quantities $`p_{1,2}`$ should be replaced by
$$p_j=[\gamma _l+\alpha _{j0}^{}+i\mathrm{\Omega }_\mu ]/q_j\overline{\mathrm{v}},q_j=\sqrt{(k_\mu M_jk)^2+4M_jkk_\mu \mathrm{sin}^2(\theta /2)},$$
$$j=1,2,$$
$`(4.3)`$
and the widths $`k_\mu \overline{\mathrm{v}}`$ and $`q\overline{\mathrm{v}}`$ should be replaced by $`q_1\overline{\mathrm{v}}`$ and $`q_2\overline{\mathrm{v}}`$. Thus, the widths and the positions of both lines in the $`ml`$ transition depend on the frequency and direction of propagation of the absorbed photon, and its role is determined by the ”memory” factor $`M_{1,2}`$. In the case of exact resonance we have $`M_1=M_2=1/2`$ and $`\alpha _{1,0}^{^{\prime \prime }}=\alpha _{2,0}^{^{\prime \prime }}=|G|`$, i.e., both lines are simmetrical relative to the frequency $`\omega _{ml}`$ and have an identical angular dependence of the width (Fig. 3b). It is interesting that when the condition $`k_\mu =M_1k`$ or $`k_\mu =M_2k`$ is satisfied, one of the lines has a natural width in the case of observation along $`\theta =0`$ (see the discussion of formulas (3.2) and (3.3)).
Thus, variation of $`\mathrm{\Omega }`$ leads to the following changes in the spectrum. When $`|\mathrm{\Omega }|G`$, one component of the doublet is near $`\omega _{ml}`$ and the other near $`\omega \omega _{ln}`$ . With decreasing $`|\mathrm{\Omega }|`$, the shifted component moves towards the unshifted one, the latter shifts in the same direction, and the rate of motion $`(M_{1,2})`$ is larger for the component that is farthest from $`\omega _{ml}`$. The distance between the components is $`\sqrt{\mathrm{\Omega }^2+4G^2}`$. When $`\mathrm{\Omega }=0`$, the splitting is symmetrical, and the distance between components is minimal $`(2G)`$. Further change of $`\mathrm{\Omega }`$ brings the previously-shifted component closer to $`\omega _{ml}`$, and moves the previously-unshifted component at an increased rate. In addition to the shift of the lines, their Doppler widths also change in accordance with the memory factors $`M_{1,2}`$ and the observation direction. The widths have minimal values $`|k_\mu M_{1,2}k|\overline{\mathrm{v}}`$ along the direction $`\theta =0`$, and a maximal value $`(k_\mu +M_{1,2}k)\overline{\mathrm{v}}`$ in the opposite direction.
We recall that the use of the obtained results for an analysis of other two-photon processes implies a reversal of the signs of $`\mathrm{\Omega }`$ and $`\mathrm{\Omega }_\mu `$ in accordance with whether a particular photon is absorbed or emitted. In the case of two-quantum luminescence and absorption, the quantities $`\pm (\mathrm{\Omega }+\mathrm{\Omega }_\mu )`$ are involved. Therefore, unlike Raman scattering, the minimum of the Doppler width will be reached at $`\theta =\pi `$. The magnitude of the narrowing will be the same as before.
## V DOPPLER BROADENING OF RESONANCE-FLUORESCENCE LINE ON EXCITED LEVELS
The results of Sec. 2 allow us to explain the correlation and frequency properties of the radiation also in the case of a transition between the levels that interact with the strong field $`(mn)`$. Within the framework of second-order perturbation theory, the radiation power $`\omega _\mu `$ is determined directly by formula (3.2) in which we put $`k=k_\mu `$ and $`\gamma _l=\gamma _n`$. In the angle interval $`\theta \gamma _n/k\overline{\mathrm{v}}`$, the second term will have a dispersion form with width $`2\gamma _n`$. Thus, the main conclusions of Sec. 3 concerning the width anisotropy, the line shift, etc. apply also to resonance fluorescence <sup>*</sup><sup>*</sup>*Resonance fluorescence is usually considered for the case when the lower level is the ground level $`(\gamma _n=0)`$, and transitions from $`m`$ are allowed only to $`n`$. Neither premise is satisfied in our problem. .
In the case of a strong field $`G`$, a singularity of the transition is the need for taking into account the field disturbance of both equations-both the upper and the lower. As a result, the formula for $`w_\mu `$, in the case of a strong field, is more complicated than (2.6). For our purposes it suffices, however, to use the general conclusions of Sec. 2. From formula (2.3) it is easy to conclude that $`w_\mu `$ will consist of four transitions between two sublevels of the upper state and two sublevels of the lower state. Each of these transitions contributes its own resonant term:
$$[2\alpha _1^{^{}}+i(\mathrm{\Omega }_\mu \mathrm{\Omega }\mathrm{𝐪𝐯})]^1,[\mathrm{\Gamma }+i(\mathrm{\Omega }_\mu 𝐤_\mu 𝐯2\alpha _1^{^{\prime \prime }})]^1$$
$$[2\alpha _2^{^{}}+i(\mathrm{\Omega }_\mu \mathrm{\Omega }\mathrm{𝐪𝐯})]^1,[\mathrm{\Gamma }+i(\mathrm{\Omega }_\mu 𝐤_\mu 𝐯2\alpha _2^{^{\prime \prime }})]^1$$
$`(5.1)`$
In the general case these terms differ in the positions of their maxima (as functions of $`\mathrm{\Omega }_\mu `$), in their widths, and in the coefficients with which they enter in $`w_\mu `$. We consider the simplest and most striking case when $`G\mathrm{\Omega },k\overline{\mathrm{v}},\mathrm{\Gamma },\gamma _l`$. The amplitudes of all the substates are the same here, and the fact (5.1) enter in $`w_\mu `$ with equal weights. Further, formula (2.8) is valid for $`\alpha _1`$ and $`\alpha _2`$, and consequently $`w_\mu `$ is given by
$$w_\mu [\mathrm{\Gamma }+i(\mathrm{\Omega }_\mu \mathrm{\Omega }2G\mathrm{𝐪𝐯})]^1+2[\mathrm{\Gamma }+i(\mathrm{\Omega }_\mu \mathrm{\Omega }\mathrm{𝐪𝐯})]^1+$$
$$+[\mathrm{\Gamma }+i(\mathrm{\Omega }_\mu \mathrm{\Omega }+2G\mathrm{𝐪𝐯})]^1,$$
$`(5.2)`$
$$𝐪=𝐤_\mu 𝐤,q=2k\mathrm{sin}(\theta /2).$$
The position of the maximum of one of the terms coincides with the frequency $`\omega `$ of the strong field, and the two others are shifted to the points $`\omega \pm 2G`$. The widths of all the components of the triplet have identical angular characteristics. At large observation angles, the contours of the lines have a Gaussian form with width $`q\overline{\mathrm{v}}`$. Inside the cone $`\theta <\mathrm{\Gamma }/k\overline{\mathrm{v}}`$, the lines have a dispersion form with natural width $`\mathrm{\Gamma }`$.
It is of interest to trace the connection between the components of the triplet (5.2) of the resonant fluorescence with the lines of the stepwise transition or Rayleigh scattering. If we successively increase the deviation from resonance, say in the direction of positive $`\mathrm{\Omega }`$, then the component in (5.2) of frequency $`\mathrm{\Omega }2G`$ will shift towards $`\omega _{mn}`$ change into a stepwise-transition line. The frequency of the central component will increase and coincide at all time with $`\omega `$, yielding a Rayleigh-scattering line. Finally, the third component will move away from $`\omega _{mn}`$ at a still larger length, and its amplitude will become of the order of $`G^4/\mathrm{\Omega }^4`$ (i.e., it disappears in the second approximation of perturbation theory). When $`\mathrm{\Omega }`$ changes in the opposite direction, the unshifted component, as before, remains at the frequency of the external field, and the roles of the shifted components are interchanged. Thus, the changes of the frequency-correlation properties due to an external field become manifest in the Doppler broadening of the resonance fluorescence lines to the same degree as in Raman scattering. In addition, there appears one more line that does not fit in the classification of second-order perturbation theory.
I.M. Fabelinskiy, Molekulyarnoe rasseyanie sveta (Molecular Scattering of Light), Nauka, 1965 \[Consultants Bureau, 1968\].
W. Heitler, The Quantum Theory of Radiation, Oxford, 1954.
S.G. Rautian and I.I. Sobel’man, Zh. Eksp. Teor. Fiz. 44, 934 (1963) \[Sov. Phys.-JETP 17, 635 (1963)\].
G.E. Notkin, S.G. Rautian, and A.A. Feoktistov, ibid. 52, 1673 (1967) \[25, 1112 (1967)\].
S.G. Rautian, Trudy FIAN 43, 3 (1968).
S.H. Autler and C.H. Townes, Phys. Rev. 100, 703 (1955).
V.M. Kontorovich and A.M. Prokhorov, Zh. Eksp. Teor. Fiz. 33, 1428 (1957) \[Sov. Phys.-JETP 6, 1100 (1958)\].
S.G. Rautian and I.I. Sobel’man, ibid. 41, 456 (1961) \[14 328 (1962)\].
A.M. Bonch-Bruevich and V.A. Khodovoi, Usp. Fiz. Nauk 93, 71 (1967) \[Sov. Phys.-Uspekhi 10, 637 (1968)\].
Translated by J. G. Adashko 51
|
warning/0005/astro-ph0005051.html
|
ar5iv
|
text
|
# Sgr A* Polarization: No ADAF, Low Accretion Rate, and Non-Thermal Synchrotron Emission
## 1. Introduction
The nearest supermassive black hole candidate lies at the center of the Milky Way galaxy, weighing in at $`2.6\times 10^6M_{}`$, as inferred from motions of stars near the galactic center (Ghez et al. 1998; Genzel et al. 1997). The low luminosity of the point source associated with the center, $`10^{37}`$ erg/s, is a conundrum since accretion from stellar winds of neighboring stars should create a luminosity of $`10^{41}`$erg/s. One possibility is that most of the energy is carried by the accreting matter into the black hole, as in the advection-dominated accretion flow solution (ADAF, Narayan & Yi 1994; Narayan, Yi, & Mahadevan 1995). Such a situation is achieved when most of the dissipated energy is channeled into protons which cannot radiate efficiently. Low efficiency also occurs when gas accretes spherically and carries its energy in as kinetic energy (Melia 1992). Alternatively, the accretion rate may be overestimated, and the emission may be due to a tenuous disk or jet (Falcke, Mannheim, & Biermann 1993).
The radio spectrum of Sgr A\* can be described by a power law, $`F_\nu \nu ^{1/3}`$ from centimeter to millimeter wavelengths. This is intriguingly close to the spectrum of optically thin, mono-energetic electrons emitting synchrotron radiation (Beckert & Duschl 1997). However, this explanation is not unique: a self-absorbed source which varies in size as a function of frequency may produce a similar spectral slope (Melia 1992; Narayan et al. 1995). A possible technique to distinguish these models is to measure the polarization of the emission: Faraday rotation and self-absorption can change the polarization magnitude and wavelength dependence (Jones & O’Dell 1977).
Only recently has linear polarization been detected at high frequency by Aitken et al. (2000: A00); previous searches at lower frequency showed only upper limits (Bower et al. 1999a,b). After correcting for contamination by dust and free-free emission, the inferred polarization is 10-20%, implying a synchrotron origin. This correction is made somewhat uncertain by the large beam size. Remarkably, the polarization shows a change in position angle of $`90^{}`$ around 1 mm, which A00 suggest might be due to synchrotron self-absorption. We first discuss the physics of synchrotron polarization (§2); we then apply it to various models in the literature (§3); next we discuss a model consistent with all of the observations (§4); and finally speculate on the physical implications of this model (§5).
## 2. Synchrotron Theory Background
In the synchrotron limit ($`\gamma 1`$) for an isotropic electron velocity distribution, some analytic results have been derived, which we now summarize (Ginzburg & Syrovatskii 1965 & 1969: GS ). For a uniform slab of electrons with a power-law distribution, $`dn_e/d\gamma \gamma ^\xi `$ (with $`\gamma _{min}\gamma \gamma _{max}`$ such that electrons with $`\gamma _{min}`$ and $`\gamma _{max}`$ do not contribute to the frequency of interest), we can relate the magnetic field strength and electron density in the slab to the fluid-frame brightness temperature and the spectral turnover due to self-absorption. For $`\xi =2`$ and a uniform field $`B_{}`$ (projected into the sky plane) we find $`B_{}2T_{11}^2\nu _{12}`$ G and $`\tau _C3\times 10^2T_{11}^4\nu _{12}\gamma _{min}^1`$, where $`T_{11}`$ is the brightness temperature in units of $`10^{11}`$ K at the self-absorption frequency $`\nu _t=10^{12}\nu _{12}`$ Hz and $`\tau _C`$ is the Compton scattering optical depth of the emission region. For $`\nu <\nu _t`$, the emission is self-absorbed so $`F_\nu \nu ^{5/2}`$, while above this frequency the emission is optically-thin and $`F_\nu \nu ^{(1\xi )/2}\mathrm{exp}(\nu /\nu _{max})`$ where $`\nu _{max}=3B_{}e\gamma _{max}^2/(4\pi m_ec)`$.
In the optically-thin regime, the polarization plane is perpendicular to the magnetic field with polarization $`\mathrm{\Pi }=(\xi +1)/(\xi +7/3)`$, up to 100% for $`\xi 1`$. In the optically-thick regime, $`\mathrm{\Pi }=3/(6\xi +13)`$ (for $`\xi >1/3`$); the radiation polarized perpendicular to the magnetic field is absorbed more strongly than the opposite polarization, causing the radiation polarized along the magnetic field to dominate, switching the polarization angle by $`90^{}`$, which changes the sign of $`\mathrm{\Pi }`$. Numerical calculations show that the optically-thick polarization peaks at $`|\mathrm{\Pi }|=20`$% for $`\xi =1/3`$, but remains large for $`0<\xi <2`$.
To compute the polarization near the self-absorption frequency requires a knowledge of the polarized opacity and emissivity, $`\mu _,,ϵ_,`$. For $`\xi =2`$, these can be approximated as (GS): $`\mu _,=r_s^1(\nu /\nu _t)^3(1\pm 3/4)`$ and $`(ϵ_,/\mu _,)=2S_t/9(\nu /\nu _t)^{5/2}(13\pm 9)/(4\pm 1)`$ where $`r_s`$ is the size of the emission region, $`\nu _t`$ is the frequency for which the total source has an optical depth of unity (i.e. $`\tau =\mu r_s=1/2(\mu _{}+\mu _{})r_s=1`$), $`S_t`$ is the source function near the frequency $`\nu _t`$, and the $`+`$ or $``$ signs go with the radiation emitted $``$ or $``$ to the magnetic field, respectively. GS then express the polarization and emission for a slab with uniform magnetic field strength and direction, constant density, and size $`r_s`$: $`I_{}=(ϵ_{}/\mu _{})(1\mathrm{exp}(\mu _{}r_s))`$, $`I_{}=(ϵ_{}/\mu _{})(1\mathrm{exp}(\mu _{}r_s))`$, and $`\mathrm{\Pi }=(I_{}I_{})/(I_{}+I_{}),`$ where $`I_{},I_{}`$ are the intensities (erg/cm<sup>2</sup>/s/Hz/sr) with polarization perpendicular and parallel to the projected direction of the magnetic field on the sky.
For electron distributions which are highly peaked at a single energy (such as mono-energetic or relativistic Maxwellian) the polarization for $`\nu \nu _t`$ is zero.
The Faraday effect rotates the polarization vector of photons emerging from different optical depths by different amounts, causing a cancellation in polarization (Agol & Blaes 1996). The differential Faraday rotation angle within the source scales as $`\mathrm{\Delta }\theta =3.6\times 10^{28}\tau _{phot}B\nu ^2\gamma _{min}^2`$ (Jones & O’Dell 1977), where $`\tau _{phot}`$ is the Compton optical depth of the photosphere. When optically thin, $`\tau _{phot}\tau _C`$ is constant, so rotation is largest at the self-absorbed wavelength. When self-absorbed, $`\tau _{phot}`$ of the photosphere scales as $`\nu ^{\xi /2+2}`$, so the differential Faraday rotation angle $`\nu ^{\xi /2}`$ (for $`\xi >1/3`$), again largest at the self-absorption wavelength. The differential rotation at $`\nu _t`$ is $`\mathrm{\Delta }\theta 2\pi g(\xi )(\theta _b/\gamma _{min})^\xi /\gamma _{min}`$, where $`\gamma _{min}`$ is the minimum electron Doppler factor, $`g(\xi )`$ is a dimensionless factor of order unity, and $`\theta _b`$ is the brightness temperature in units of $`m_ec^2/k_B`$.
## 3. Observational Constraints on Published Models
The observations of polarization in Sgr A\* provide the following constraints on emission models:
1) The differential Faraday rotation angle near the self-absorbed wavelength must be $`\pi `$.
2) The electron distribution must be non-thermal since the polarization due to a thermal electron distribution is suppressed when self-absorbed by a factor of $`\mathrm{exp}(\tau )`$. If the beam correction by A00 is correct, then $`\mathrm{\Pi }`$ 12% at self-absorbed wavelengths, requiring $`\xi 2`$.
3) The self-absorption frequency must lie near the change in polarization angle, $`1`$mm.
4) The component contributing at lower frequencies must have zero linear polarization.
5) The magnetic field must be ordered to prevent cancellation of polarization.
These constraints rule out several models proposed in the literature, as will be discussed in turn.
The low efficiency of an ADAF implies a higher accretion rate and thus higher density than for a high efficiency flow of the same luminosity and geometrical thickness. For Sgr A\*, an accretion rate of $`10^{(45)}M_{}/`$yr is inferred due to capture of gas in the vicinity of the black hole (Quataert, Narayan, & Reid 1999; Coker & Melia 1999), which is the value assumed in ADAF models. Assuming that the gas falls in at near the free-fall speed, one infers an electron density $`n_e=10^{10}`$ cm$`{}_{}{}^{3}\dot{m}_{5}^{}x^{3/2}`$ and a magnetic field strength of $`B=10^3\mathrm{G}\dot{m}_5^{1/2}x^{5/4}(v_A/0.1v_{ff})`$, where $`x`$ is the radius of the emission region in units of $`r_g=GM/c^2`$, $`\dot{m}_5`$ is the accretion rate in units of $`10^5M_{}`$/yr, and $`v_A/v_{ff}`$ is the ratio of the Alfvén speed to the free-fall speed. These values imply a total Faraday rotation angle at the self-absorption frequency $`\nu _t`$ of $`\mathrm{\Delta }\theta 10^4\dot{m}_5^{3/2}\nu _{12}^2(v_A/0.1v_{ff})`$. This value is so large that rotation of the emitted radiation leads to zero net polarization, so ADAFs are in direct conflict with the observed polarization. Only significant modifications of the model, such as a reduction in the accretion rate by a factor of $`10^3`$, can reduce the Faraday rotation angle $`\pi `$. An accretion rate of $`10^8M_{}/`$yr is consistent with the observed luminosity if the accretion flow has a higher efficiency $``$2%, no longer “advection-dominated.” In addition, ADAF models assume a Maxwellian electron distribution, which cannot produce the observed switch in polarization angle<sup>1</sup><sup>1</sup>1 Mahadevan (1999) and Özel, Psaltis, & Narayan (2000) have added a non-thermal electron component to ADAF models which contributes to the flux at wavelengths longer than 2 mm, not at the polarized wavelengths.. Finally, ADAFs predict a higher self-absorption frequency: Özel et al. (2000) find that $`\nu _t5\times 10^{12}\dot{m}_5^{5/9}\mathrm{Hz}`$, which implies $`\dot{M}4\times 10^7M_{}`$/yr to be consistent with the observed $`\nu _t5\times 10^{11}`$Hz. The accretion rate might be reduced if there is significant gas lost by a wind or jet (Begelman & Blandford 1999; Quataert & Narayan 1999) or if the Bondi rate is reduced by heating the infalling gas with heat carried outwards by a convection-dominated accretion flow, or “CDAF” (Igumenshchev & Abramowicz 1999, 2000; Stone, Pringle, & Begelman 1999; Quataert & Gruzinov 2000; Narayan, Igumenshchev, & Abramowicz 2000; Igumenshchev, Abramowicz, & Narayan 2000).
The model of Melia (1992) is rather similar to the ADAF model, and thus suffers the same problems: the high accretion rate implies high density which is inconsistent with the observed polarization.
Beckert & Duschl (1997) considered several 1-zone, quasi-monoenergetic and thermal emission models for the synchrotron emission. These electron distributions do not produce a swing in polarization angle by 90 degrees since the polarization is suppressed when self-absorbed. Their model does produce a self-absorption frequency near the correct frequency, however. Falcke, Mannheim, & Biermann (1993) present a disk-plus-jet model which assumes a tangled magnetic field topology which would erase any polarization. However, an ordered magnetic field would be a small change to their model which might bring it into line with the polarization observations.
## 4. A Phenomenological Model
Now, we attempt to construct a model consistent with all of the data, using uniform emission regions for simplicity. Typical optically-thin AGN spectra show $`\xi 23`$; since $`\xi =2`$ is consistent with the polarization from A00, we fix $`\xi =2`$ in our model fits. The model parameters for the polarized component are $`S_t=6`$ Jy, $`\nu _t=550`$ GHz (corresponding to $`\lambda =0.55`$ mm), and $`\nu _{max}5000`$GHz (Figure 1).
To explain the lack of polarization and spectral slope flatter than 5/2, we require an additional component which is unpolarized and has a cutoff near 1 mm so that it doesn’t dilute the polarization at shorter wavelengths. Since Sgr A\* has a spectral slope of 1/3 at mm wavelengths and appears to have a spectral turnover at 1 GHz, we model the spectrum as a monoenergetic electron distribution with energy $`\gamma `$ and zero polarization (due to Faraday depolarization or tangled magnetic field) which becomes self-absorbed at low frequency (Beckert & Duschl 1997). For the unpolarized component, we find $`F_\nu =1.3(\nu /\nu _{max})^{1/3}\mathrm{exp}(\nu /\nu _{max})\mathrm{Jy}`$ with $`\nu _{max}50`$GHz, and $`\nu _t1`$GHz (Figure 1).
Fig. 1: Polarization and spectral energy distribution of Sgr A\* compared to model. The dashed line shows the polarized component, the dotted line the unpolarized, mono-energetic component, and the solid line the sum of the two. The dot-dash line shows the maximum CDAF model (assumed to be unpolarized; the total polarization is similar if the CDAF replaces the monoenergetic component). The diamonds are the data compiled by Narayan et al. (1995), while the asterisks are the data from Bower et al. (1999a,b) and A00.
Figure 1 compares the model to the data. To compare the polarization, we have plotted the Stokes’ parameter that lies at $`83^{}`$. Remarkably, the polarization should rise to $`100`$% at even shorter wavelengths.
### 4.1. Physical conditions in Synchrotron emitting regions
Krichbaum et al. (1998) report a source radius of $`55\mu `$as at 1.4 mm from VLBI observations; this corresponds to 19$`r_g`$. The source size may be smaller at higher frequencies, but we expect the radius of the emission region to be greater than the size of the event horizon of the black hole, which has an apparent size of $`5r_g15\mu `$as projected on the sky (including gravitational bending, Bardeen 1973), so we use an intermediate size in further estimates. The flux of the fitted model at the self-absorption frequency, $`\nu _t=550`$ GHz, is $`9`$ Jy. This implies a brightness temperature in the emission frame $`T_b1.6\times 10^{10}(r_s/10r_g)^2\mathrm{\Gamma }^1`$ K, where $`r_s`$ is the size of the source (we have assumed the area of the source is $`\pi r_s^2`$) and $`\mathrm{\Gamma }`$ is the bulk Doppler boost parameter. For a steeply falling electron number distribution, $`kT_b4\gamma m_ec^2`$ (for $`\xi =2`$), where $`\gamma m_ec^2`$ is the energy of the emitting electrons, implying $`\gamma 10(r_s/10r_g)^2\mathrm{\Gamma }^1`$ for the electrons at the self-absorption frequency. Using the formulae from §2, we find: $`B_{}=350(r_s/10r_g)^4\mathrm{\Gamma }\mathrm{G}`$, $`\tau _C=10^5(r_s/10r_g)^8\mathrm{\Gamma }^5`$, and $`\gamma _{max}=50(r_s/10r_g)^2\mathrm{\Gamma }^1`$, implying $`n_e6\times 10^6(r_s/10r_g)^9\mathrm{\Gamma }^5`$cm<sup>-3</sup>. The ratio of magnetic to rest-mass energy density is $`B^2/(8\pi n_em_pc^2)1(r_s/10r_g)^{17}\mathrm{\Gamma }^{9/2}`$ for an electron-proton plasma, indicating a relativistic Alfvén speed. The Faraday rotation angle at $`\nu _t=5.5\times 10^{11}`$ Hz is $`\mathrm{\Delta }\theta 350(r_s/10r_g)^4\mathrm{\Gamma }^2\gamma _{min}^3`$, assuming $`B_{}B_{}`$. For $`r_s10r_g`$, $`\gamma _{min}`$ can be as large as 4, reducing $`\mathrm{\Delta }\theta `$ to 5; for $`r_s5r_g`$, $`\gamma _{min}`$ can be as large as 20 reducing $`\mathrm{\Delta }\theta `$ to $`0.6`$. Alternatively, if the synchrotron emission is due to a pair plasma, Faraday rotation will be reduced by the ratio of the proton number density to the pair number density. The rotation angle is further reduced at the observed wavelengths by a factor $`\nu /\nu _t`$. The high energy cutoff for the electron distribution may be due to synchrotron cooling since $`t_{cool}=8\times 10^8\gamma _{max}^1B^26(r_s/10r_g)^6\mathrm{\Gamma }^3`$ sec, similar to the dynamical time, $`t_D13x^{3/2}`$ sec. Given the strong scaling of quantities with the unknown $`r_s`$ and $`\mathrm{\Gamma }`$, the above estimates can only be improved with future observations.
The unpolarized emission component dominates at $`7`$ mm, where Lo et al. (1998) measure a source size of $`5\times 10^{13}`$ cm. The self-absorption frequency then requires $`\gamma 400`$, $`B0.1`$ G, and $`n_e4\times 10^5`$ cm<sup>-3</sup>. Though somewhat ad-hoc, this model reproduces the spectrum well. The Faraday rotation parameter is rather small, so depolarization requires field which is tangled on a scale $`100`$ times smaller than the size of the emission region.
### 4.2. Accretion Component
We have tried modeling the spectrum of the unpolarized component with a self-similar, self-absorbed accretion flow. We used the cyclo-synchrotron emission formulae from Mahadevan, Narayan, & Yi (1996) and we performed the radiation transfer in full general relativity (Kurpiewski & Jaroszyński 1997). We can place an upper limit on the accretion rate of an ADAF component (using the model of Özel et al. 2000) since its unpolarized flux must not dilute the polarized component: we find $`\dot{M}_{ADAF}3\times 10^6M_{}/`$yr. If the ADAF surrounds the polarized emission region, then it will depolarize, so the Faraday depolarization places a stronger upper limit (§2). We can place a similar limit on the CDAF model (using the structure from Quataert & Gruzinov, 2000, with equipartition $`B`$ field and $`p_{gas}=2n_ek_BT_e`$): we find $`\dot{M}_{CDAF}1.5\times 10^9M_{}/`$yr; this accretion rate can account for the unpolarized component at $`\nu 10`$GHz (see Figure 1) and is consistent with the Faraday rotation constraint. The CDAF luminosity is $`2\times 10^{34}`$erg/s and the self-absorption frequency is $`30`$GHz, so the polarized component would be visible through it. Finally, we can place a limit on a standard thin disk from the infrared upper limits: we find $`\dot{M}_{thin}2\times 10^{11}M_{}/`$yr; this upper limit can be increased to a maximum of $`3\times 10^7M_{}/`$yr if the inner edge of the disk is truncated at $`r=6000r_g`$.
## 5. Conclusions
The main success of advection-dominated accretion models for Sgr A\* is in explaining the high-frequency radio spectrum and skirting below the upper limits at infrared frequencies. However, the ADAF model is unpolarized at the same high frequencies, inconsistent with the recent detection of linear polarization. We have constructed a simple toy model for the millimeter polarization which predicts a rise towards shorter wavelengths: polarization of $`70\%`$ might be seen with SCUBA at 350 $`\mathrm{\mu m}`$ if this model is correct. The lack of polarization and spectral slope of 1/3 at wavelengths longer than 2 mm indicates that a different physical component may be contributing. The presence of two physical components can be confirmed by looking for a change in variability amplitude and time-scale or source size and morphology around 2 mm.
The high observed polarization implies a highly ordered magnetic field lying near the sky plane. This might be due to the poloidal field in a jet (Falcke, Mannheim, & Biermann 1993), or due to a toroidal field in a disk component seen edge-on. The non-thermal electron distribution might be produced by shock acceleration, reconnection, or electric field acceleration near the event horizon of a spinning black hole (Blandford & Znajek 1977). The Blandford-Znajek mechanism can generate a maximum luminosity of $`L_{BZ}10^{37}(B/600G)^2`$ erg/s (Thorne, Price, & MacDonald 1986), so the entire polarized luminosity of Sgr A\* might be powered by black hole spin.
The dynamics of the emission region will be controlled by the ratio of the magnetic field energy density to the matter energy density, $`B^2/(8\pi \rho c^2)`$; however, this ratio scales as $`r_s^{17}\mathrm{\Gamma }^{9/2}`$, while $`\mathrm{\Gamma }`$ and $`r_s`$ are unknown. Doppler boosting decreases the brightness temperature, which reduces Faraday rotation but makes the electrons trans-relativistic. Future sub-mm VLBI observations should accurately measure the $`r_s`$ as a function of frequency, and proper motions may constrain $`\mathrm{\Gamma }`$. Also uncertain are the pair fraction and minimum electron energy $`\gamma _{min}`$. The pair number density can be constrained by measuring the circular polarization; without pairs, the circular polarization may be as high as a few percent at optically-thin wavelengths (Jones & O’Dell 1977), while pure pair emission should have no circular polarization. The pair annihilation line should be looked for at higher spatial resolution; however, it will be strongly broadened by relativistic motions of the pairs. Once the source size is known, $`\gamma _{min}`$ and the pair fraction will be constrained by the Faraday rotation limit.
An ADAF model must have a low accretion rate, $`10^8M_{}/`$yr, to be consistent with the lack of Faraday rotation of the polarized emission. Such a low inferred accretion rate disagrees with estimates of the Bondi accretion rate inferred from stellar winds near the region of the black hole. If accretion is episodic due to outer-disk instabilities, then the current state might be one of low accretion rate in the inner disk. Alternatively, the accretion rate might be reduced by depositing energy from the accretion flow in the surrounding gas (either through outflow or convection), thus increasing the sound speed and decreasing the capture rate of gas by the black hole. The accretion flow must deposit energy $`\dot{M}_AGM/r_A6\times 10^{35}`$ erg/s, where $`\dot{M}_A`$ is the stellar mass loss rate which crosses the Bondi radius $`r_A`$ (Quataert et al. 1999). This can be supplied by accretion which releases energy $`5\times 10^{35}(\eta /0.01)\dot{m}_9`$ erg/s, where $`\eta `$ is the efficiency with which accretion deposits energy at large radius. As remarked above, a convection-dominated accretion flow with $`\dot{M}10^9M_{}/`$yr can explain part of the unpolarized component without diluting the polarized emission; the associated convection can carry the required energy outward to suppress the Bondi accretion rate.
Since the self-absorption frequency occurs at $`500\mathrm{\mu m}`$, it will be possible to image shadow of a black hole from the ground using VLBI, providing a direct confirmation of the existence of an event horizon (Falcke, Melia, & Agol 2000). Future sub-mm polarimetric VLBI observations might show rotation of the polarization angle near the black hole, a general relativistic effect which becomes stronger for a spinning black hole (Connors, Stark, & Piran 1980).
Eliot Quataert & Andrei Gruzinov have shown me work which reaches similar conclusions about ADAFs and the need for a very low accretion rate onto Sgr A\*.
I acknowledge Ski Antonucci, Julian Krolik, Colin Norman, and Eliot Quataert for ideas and corrections which greatly improved this letter. This work was supported by NSF grant AST 96-16922.
|
warning/0005/cond-mat0005217.html
|
ar5iv
|
text
|
# 𝑑-wave nonadiabatic superconductivity
## 1 Introduction
In ordinary low-temperature superconductors, the smallness of the relevant phonon frequency $`\omega _0`$ compared to the Fermi energy $`E_\mathrm{F}`$ permits to formulate a theory of superconductivity based on a closed set of formulas known as Migdal-Eliashberg (ME) equations migdal ; elia allowing quantitative agreements with experiments carbotte . The closed form of the ME equations stems from Migdal’s theorem which states that as long as $`\omega _0/E_\mathrm{F}1`$ the electron-phonon ($`e`$-$`ph`$) vertex corrections to the electron self-energy are at least of order $`\lambda \omega _0/E_\mathrm{F}`$, where $`\lambda `$ is the $`e`$-$`ph`$ coupling, and can therefore be neglected migdal .
A different situation is encountered in high-$`T_\mathrm{c}`$ superconductors such as cuprates and fullerides. These materials have in fact Fermi energies much smaller than those of conventional metals uemura ; gunna so that the energy scale $`\omega _0`$ associated to the mediator of the superconducting pairing can be comparable to $`E_\mathrm{F}`$. Hence, the quantity $`\omega _0/E_\mathrm{F}`$ is no longer negligible and in principle vertex corrections become relevant preventing the ordinary ME scheme to be a correct description of the superconducting state.
The possible breakdown of Migdal’s theorem in high-$`T_\mathrm{c}`$ superconductors inevitably calls for a generalization beyond the ME scheme to include the no longer negligible vertex corrections. A possible way to accomplish this goal is to rely on a perturbative scheme by truncating the infinite set of vertex corrections at a given order. In previous works, we have proposed a perturbative scheme in which the role of small parameter is played roughly by $`\lambda \omega _0/E_\mathrm{F}`$ leading to a generalized ME theory which includes the first nonadiabatic vertex corrections GPSprl ; PSG . For a single-electron Holstein model, such a first order perturbative approach leads to good agreements with exact results as long as the system is away from polaron formation, that is for $`\lambda <\lambda _c1`$ capone . The region of validity of the perturbative approach, which we could name nonadiabatic region, is characterized by quasi-free electrons ($`\lambda <1`$) coupled in a nonadiabatic way ($`\omega _0/E_\mathrm{F}`$ not negligible) to the lattice. According to this definition, our nonadiabatic region is different from the classic polaronic picture. Of course, larger values of $`\lambda `$ would render higher order vertex corrections important leading to the breakdown of our truncation scheme.
The key point of the nonadiabatic theory, is that the $`e`$-$`ph`$ effective interaction is described in terms of vertex corrections which depend on the momentum transfer $`|𝐪|=q`$ and the Matsubara exchanged frequency $`\omega `$ in a non-trivial way. For example, the $`e`$-$`ph`$ vertex correction appearing in the normal state self-energy becomes positive (negative) when $`v_\mathrm{F}q<\omega `$ ($`v_\mathrm{F}q>\omega `$), where $`v_\mathrm{F}`$ is the Fermi velocity PSG . The generalization to the superconducting transition reveals that this situation is also encountered for the class of diagrams beyond Migdal’s limit relevant for the Cooper channel. Concerning the critical temperature $`T_\mathrm{c}`$, as long as the momentum transfer is less than $`\omega _0/v_\mathrm{F}`$, the nonadiabatic corrections lead to a strong enhancement of $`T_\mathrm{c}`$ also for moderate values of the $`e`$-$`ph`$ coupling $`\lambda `$ GPSprl ; PSG ; CP . Such a strong momentum-frequency dependence of the vertex corrections is confirmed by numerical calculations within a tight-binding approach pera and general theoretical considerations on the physical interpretation of such nonadiabatic corrections GPSepj .
So far, nonadiabatic superconductivity has been studied by requiring the order parameter $`\mathrm{\Delta }`$ to be independent of the momenta. This situation is certainly suitable for the fullerene compounds which are $`s`$-wave superconductors. However, one striking characteristic of several high-$`T_\mathrm{c}`$ superconductors is the strong momentum dependence of the order parameter $`\mathrm{\Delta }(𝐤)`$. Among the several types of measurements aimed to resolve the pairing symmetry, the Josephson tunneling wollman and angle resolved photoemission experiments shen are the most convincing ones showing that the order parameter of several cuprates, maybe with the exception of the electron-doped NCCO, has a predominant $`d`$-wave symmetry: $`\mathrm{\Delta }(𝐤)\mathrm{\Delta }[\mathrm{cos}(k_x)\mathrm{cos}(k_y)]`$.
The origin of the $`d`$-wave symmetry in high-$`T_\mathrm{c}`$ cuprates is still debated. On one hand, the observed $`d`$-wave symmetry is regarded as an evidence against any purely electron-phonon pairing interaction so that the mechanism responsible for superconductivity should be sought among pairing mediators of electronic origin (like antiferromagnetic fluctuations) with eventually a minor electron-phonon component. On the other hand, several theoretical studies have shown that the $`e`$-$`ph`$ interaction could produce, under some quite general circumstances, a $`d`$-wave symmetry of the condensate lichten ; mierze . This could happen when for example charge carriers experience an on-site repulsive interaction together with a phonon induced attraction for large inter-electrons distances. The on-site repulsion inhibits the isotropic $`s`$-wave superconducting response leading the system to prefer order parameters of higher angular momenta. A quite general analysis of the interplay between on-site repulsion and neighbour and next-neighbour attraction has shown $`s`$-wave to $`d`$-wave crossover depending on the microscopic parameters of a model BCS Hamiltonian fehren .
The purpose of the present paper is to study how the $`d`$-wave superconducting response resulting from a strongly momentum dependent total interaction, is affected by the inclusion of nonadiabatic vertex corrections. In particular, we intend to clarify whether the complex momentum-frequency structure of the nonadiabatic contributions could sustain an underlying $`d`$-wave symmetry of the order parameter.
In the next section we introduce the model and the corresponding ME equations for $`s`$\- and $`d`$ wave symmetries of the gap. In Sec. 3 we generalize the ME equations to include the nonadiabatic terms for each symmetry channel and calculate the corresponding critical temperatures. We find that the theory of nonadiabatic superconductivity can lead to $`d`$-wave even for phonons in a broad parameter range which depends on the degree of electronic correlation.
## 2 The model
In this section we introduce a simple model interaction suitable for our investigation beyond Migdal’s limit and capable of providing for $`s`$\- or $`d`$-wave symmetries of the order parameter. Let us consider the anomalous self-energy at the critical temperature
$$\mathrm{\Sigma }_\mathrm{S}(k)=\underset{k^{}}{}V_{\mathrm{pair}}(kk^{})G(k^{})G(k^{})\mathrm{\Sigma }_\mathrm{S}(k^{}),$$
(1)
where $`G(k^{})`$ is the fermion dressed propagator:
$$G(k^{})=\frac{1}{i\omega _mϵ_𝐤^{}\mathrm{\Sigma }_\mathrm{N}(k^{})}$$
(2)
and $`\mathrm{\Sigma }_\mathrm{N}`$ is the normal self-energy. We use the compact notation $`k(𝐤,\omega _n)`$, $`k^{}(𝐤^{},\omega _m)`$ and $`_k^{}T_\mathrm{c}_m_𝐤^{}`$ where $`\omega _n`$, $`\omega _m`$ are fermionic Matsubara frequencies and $`𝐤`$, $`𝐤^{}`$ are electronic momenta (from now on, all momenta are two-dimensional vectors lying on the copper-oxygen plane).
To define the model interaction $`V_{\mathrm{pair}}(kk^{})`$ we have made use of a number of informations gathered from previous studies. First, in order to obtain order parameters with higher angular momenta than $`s`$-wave, it is sufficient to consider a pair interaction made of a repulsive part at short distances and an attractive one at higher distances (Fig. 1).
In momentum space, this interaction corresponds to an attractive coupling for small $`𝐪`$ and a repulsive one for large $`𝐪`$, where $`𝐪=𝐤𝐤^{}`$ is the momentum transfer.
Let us now try to interpret this strong momentum modulation in terms of $`e`$-$`ph`$ and electron-electron interactions. In strongly correlated systems, the $`e`$-$`ph`$ interaction acquires an important momentum dependence in such a way that for large values of the momentum transfer $`𝐪`$ the $`e`$-$`ph`$ interaction is suppressed, whereas for small values of $`𝐪`$ it is enhanced zey . A physical picture to justify this momentum modulation is the following dany . In many-electrons systems a single charge carrier is surrounded by its own correlation hole of size $`\xi `$ which can be much larger than the lattice parameter $`a`$ in the strongly correlated regime. This implies that one electron interacts with molecular vibrations of wavelength of order $`\xi `$ or larger, leading to an effective upper cut-off $`q_\mathrm{c}\xi ^1`$ in the momenta space. Thus we have a non zero electron-phonon interaction when $`|𝐪|<q_\mathrm{c}`$. The cut-off momentum $`q_\mathrm{c}`$ can also be regarded as a measure of the correlation in the system: $`aq_\mathrm{c}1`$ in strongly correlated systems while $`aq_\mathrm{c}1`$ in the case of free electrons.
From the above considerations, the attractive part at small $`𝐪`$ of our model pairing interaction finds a natural interpretation in the $`e`$-$`ph`$ coupling modified by the strong electron correlations. We introduce therefore the following simple form for the $`e`$-$`ph`$ part of the pairing interaction:
$`V(kk^{})`$ $`=`$ $`|g(𝐤𝐤^{})|^2D(\omega _n\omega _m)`$
$``$ $`g^2\left[{\displaystyle \frac{\pi k_\mathrm{F}}{q_\mathrm{c}}}\right]\theta (q_\mathrm{c}|𝐤𝐤^{}|)D(\omega _n\omega _m),`$
where
$$D(\omega _n\omega _m)=\frac{\omega _0^2}{(\omega _n\omega _m)^2+\omega _0^2},$$
(4)
is the phonon propagator for which we have adopted a simple Einstein spectrum with frequency $`\omega _0`$. In Eq.(2), $`\theta `$ is the Heaviside step function and the prefactor $`(\pi k_\mathrm{F}/q_\mathrm{c})`$ has been introduced in order to assure that the momentum average of $`|g(𝐤𝐤^{})|^2`$ becomes $`g^2`$ for relatively small values of the cut-off $`q_\mathrm{c}`$ regardless of the particular symmetry of the order parameter. In this way the comparison between $`s`$\- and $`d`$-wave solutions, especially in the nonadiabatic case treated in the next section, is more transparent.
Having defined the nature of the attractive part of the total pairing interaction $`V_{\mathrm{pair}}(kk^{})`$, we offer now a possible interpretation for the remaining repulsive part acting at large $`𝐪`$. This repulsion is given by the residual e-e interaction and its momentum dependence can be obtained, in analogy with the renormalization of the e-ph interaction, by using the above picture of correlation holes. In this picture, the residual e-e interaction should ensure that charge fluctuations with wavelength less than $`\xi `$ are unfavourable. This can be modelled by requiring that in momentum space the residual interaction is repulsive for $`|𝐤𝐤^{}|>q_\mathrm{c}`$ and, by using again the theta-function for later convenience, we introduce therefore the following residual repulsion:
$$U(𝐤𝐤^{})=U\left[\frac{\pi k_\mathrm{F}}{q_\mathrm{c}}\right]\theta (|𝐤𝐤^{}|q_\mathrm{c}).$$
(5)
In the above expression, $`U>0`$ and the factor $`\pi k_\mathrm{F}/q_\mathrm{c}`$ has been introduced for the same reason as in Eq.(2). Note that, in principle, $`q_\mathrm{c}`$ depends on $`U`$, however here we shall treat two quantities independently on each other by keeping in mind that small values of $`q_\mathrm{c}`$ correspond roughly to large values of $`U`$. An additional simplification of the following calculations is achieved by expressing the off-diagonal self-energy (1) in terms of a suitable pseudopotential $`U^{}`$ rather then $`U`$. It is then opportune to formally replace equation (5) by
$$U^{}(kk^{})=U^{}\left[\frac{\pi k_\mathrm{F}}{q_\mathrm{c}}\right]\theta (|𝐤𝐤^{}|q_\mathrm{c})\frac{\omega _0^2}{(\omega _n\omega _m)^2+\omega _0^2},$$
(6)
where $`U^{}`$ represents the dynamically screened Coulomb repulsion and the last factor is a cut-off over the Matsubara frequencies which has been chosen to have the same functional form of the phonon propagator for convenience.
By summarizing the above results, in the off-diagonal self-energy $`\mathrm{\Sigma }_\mathrm{S}`$, Eq.(1), the total pairing interaction $`V_{\mathrm{pair}}(kk^{})`$ is given by:
$$V_{\mathrm{pair}}(kk^{})=V(kk^{})+U^{}(kk^{}),$$
(7)
where $`V(kk^{})`$ and $`U^{}(kk^{})`$ are given by equation s (2) and (6), respectively. Finally, the normal state self-energy $`\mathrm{\Sigma }_\mathrm{N}`$ entering (2) is given by
$$\mathrm{\Sigma }_\mathrm{N}(\omega _n)=\underset{k^{}}{}V(kk^{})G(k^{}),$$
(8)
where the electron-electron interaction has been absorbed in a shift of the chemical potential.
In what follows we assume the Fermi surface to be a circle in the momenta space; thus the electronic energy $`ϵ_𝐤`$ depends only on $`|𝐤|`$. Moreover, we approximate equations (2) and (6) by keeping $`|𝐤|=|𝐤^{}|=k_\mathrm{F}`$ so that, for example, $`𝐤=k_\mathrm{F}(\mathrm{cos}\varphi ,\mathrm{sin}\varphi )`$. In this way both $`\mathrm{\Sigma }_\mathrm{S}`$ and $`\mathrm{\Sigma }_\mathrm{N}`$ depend on the momentum $`𝐤`$ only via the angle $`\varphi `$. At this point it is convenient to transform the momentum integrations appearing in $`\mathrm{\Sigma }_\mathrm{S}`$ and $`\mathrm{\Sigma }_\mathrm{N}`$ into energy integrations as follows:
$$\underset{𝐤}{}\frac{d\varphi }{2\pi }𝑑ϵN(ϵ)$$
(9)
where $`N(ϵ)`$ is the density of states for the electrons. We make the approximation of constant value for $`N(ϵ)=N_0`$ and finite bandwidth $`E`$ such that the energy is defined in the interval $`E/2ϵE/2`$. The chemical potential is $`\mu =0`$, so that we refer to the half-filled situations ($`E_\mathrm{F}=E/2`$).
On performing the integration over the energy, the anomalous self-energy $`\mathrm{\Sigma }_\mathrm{S}`$ reduces to:
$`\mathrm{\Sigma }_\mathrm{S}(\varphi ,\omega _n)=N_0\pi T_\mathrm{c}{\displaystyle \underset{\omega _m}{}}{\displaystyle \frac{d\varphi ^{}}{2\pi }\left[|g(\mathrm{cos}\theta )|^2U^{}(\mathrm{cos}\theta )\right]}`$
$`\times D(\omega _n\omega _m){\displaystyle \frac{\mathrm{\Sigma }_\mathrm{S}(\varphi ^{},\omega _m)}{|\omega _m|Z(\varphi ^{},\omega _m)}}{\displaystyle \frac{2}{\pi }}\mathrm{arctan}\left[{\displaystyle \frac{E/2}{|\omega _m|Z(\varphi ^{},\omega _m)}}\right],`$
$`Z(\varphi ,\omega _n)=`$ $`1`$ $`N_0{\displaystyle \frac{\pi T_\mathrm{c}}{\omega _n}}{\displaystyle \underset{\omega _m}{}}{\displaystyle \frac{d\varphi ^{}}{2\pi }|g(\mathrm{cos}\theta )|^2D(\omega _n\omega _m)}`$ (11)
$`\times `$ $`{\displaystyle \frac{\omega _m}{|\omega _m|}}{\displaystyle \frac{2}{\pi }}\mathrm{arctan}\left[{\displaystyle \frac{E/2}{|\omega _m|Z(\varphi ^{},\omega _m)}}\right],`$
where $`\mathrm{\Sigma }_\mathrm{N}(\varphi ,\omega _n)=i\omega _n[1Z(\varphi ,\omega _n)]`$ and $`\theta =\varphi \varphi ^{}`$. The wave function renormalization $`Z(\varphi ,\omega _n)`$ actually does not depend on $`\varphi `$ and reduces to:
$`Z(\omega _n)=`$ $`1`$ $`N_0g^2_0{\displaystyle \frac{\pi T_\mathrm{c}}{\omega _n}}{\displaystyle \underset{\omega _m}{}}D(\omega _n\omega _m){\displaystyle \frac{\omega _m}{|\omega _m|}}`$ (12)
$`\times `$ $`{\displaystyle \frac{2}{\pi }}\mathrm{arctan}\left[{\displaystyle \frac{E/2}{|\omega _m|Z(\omega _m)}}\right],`$
where $`g^2_0=_\pi ^\pi 𝑑\theta |g(\mathrm{cos}\theta )|^2/(2\pi )`$. Let us expand the off-diagonal self-energy (2) as follows:
$$\mathrm{\Sigma }_\mathrm{S}(\varphi ,\omega _n)=\underset{l=\mathrm{}}{\overset{+\mathrm{}}{}}\mathrm{\Sigma }_\mathrm{S}^{(l)}(\omega _n)Y_l(\varphi ),$$
(13)
where $`Y_l(\varphi )=e^{il\varphi }/\sqrt{2\pi }`$ are eigenfunctions of the operator $`L=id/d\varphi `$. By requiring $`\mathrm{\Sigma }_\mathrm{S}(\varphi ,\omega _n)`$ be real and invariant under $`\varphi \varphi \pm \pi `$ (singlet pairing) the above expansion reduces to:
$$\mathrm{\Sigma }_\mathrm{S}(\varphi ,\omega _n)=\frac{\mathrm{\Sigma }_\mathrm{S}^{(0)}(\omega _n)}{\sqrt{2\pi }}+\sqrt{\frac{2}{\pi }}\mathrm{\Sigma }_\mathrm{S}^{(2)}(\omega _n)\mathrm{cos}(2\varphi )+\mathrm{},$$
(14)
where we have singled out the $`s`$-wave and $`d`$-wave components since in the following we consider only these symmetries. By multiplying both sides of (2) by $`Y_l^{}^{}(\varphi )`$ and integrating over $`\varphi `$, it is straightforward to show that the equations for different values of the index $`l`$ are decoupled and that $`\mathrm{\Sigma }_\mathrm{S}^{(l)}(\omega _n)`$ reduces to:
$`\mathrm{\Sigma }_S^{(l)}(\omega _n)=`$ $``$ $`N_0\left[g^2_lU^{}_l\right]\pi T_\mathrm{c}{\displaystyle \underset{\omega _m}{}}D(\omega _n\omega _m)`$ (15)
$`\times `$ $`{\displaystyle \frac{\mathrm{\Sigma }_S^{(l)}(\omega _m)}{|\omega _m|Z(\omega _m)}}{\displaystyle \frac{2}{\pi }}\mathrm{arctan}\left[{\displaystyle \frac{E/2}{|\omega _m|Z(\omega _m)}}\right],`$
where
$`g^2_l`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle _\pi ^\pi }𝑑\theta |g(\mathrm{cos}\theta )|^2e^{il\theta }`$ (16)
$`U^{}_l`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle _\pi ^\pi }𝑑\theta U^{}(\mathrm{cos}\theta )e^{il\theta }`$ (17)
Finally, by introducing the coupling constants $`\lambda _l=N_0g^2_l`$, $`\mu _l^{}=N_0U^{}_l`$, and by setting $`\mathrm{\Delta }_l=\mathrm{\Sigma }_\mathrm{S}^{(l)}/Z`$, the Eliashberg equations assume the following more familiar form:
$`Z(\omega _n)=`$ $`1`$ $`\lambda _0{\displaystyle \frac{\pi T_\mathrm{c}}{\omega _n}}{\displaystyle \underset{\omega _m}{}}D(\omega _n\omega _m){\displaystyle \frac{\omega _m}{|\omega _m|}}`$ (18)
$`\times `$ $`{\displaystyle \frac{2}{\pi }}\mathrm{arctan}\left[{\displaystyle \frac{E/2}{|\omega _m|Z(\omega _m)}}\right],`$
$`Z(\omega _n)\mathrm{\Delta }_l(\omega _n)=`$ $``$ $`\left(\lambda _l\mu _l^{}\right)\pi T_\mathrm{c}{\displaystyle \underset{\omega _m}{}}D(\omega _n\omega _m){\displaystyle \frac{\mathrm{\Delta }_l(\omega _m)}{|\omega _m|}}`$ (19)
$`\times `$ $`{\displaystyle \frac{2}{\pi }}\mathrm{arctan}\left[{\displaystyle \frac{E/2}{|\omega _m|Z(\omega _m)}}\right].`$
For $`l=0`$ and $`l=2`$, equations (18) and (19) are Migdal-Eliashberg equations for s-wave and d-wave symmetry channels, respectively. The explicit expressions of the constants $`\lambda _l`$ and $`\mu _l^{}`$ follow from the models we adopted for the electron-phonon and electron-electron interactions. For $`l=0`$ ($`s`$-wave) they reduce to:
$$\lambda _0=\lambda \left[\frac{\pi k_\mathrm{F}}{q_\mathrm{c}}\right]\theta (q_\mathrm{c}|𝐤𝐤^{}|)_{l=0}=\lambda \frac{\mathrm{arcsin}Q_\mathrm{c}}{Q_\mathrm{c}},$$
(20)
$`\mu _0^{}`$ $`=`$ $`\mu ^{}\left[{\displaystyle \frac{\pi k_\mathrm{F}}{q_\mathrm{c}}}\right]\theta (|𝐤𝐤^{}|q_\mathrm{c})_{l=0}`$ (21)
$`=`$ $`\mu ^{}\left({\displaystyle \frac{\pi }{2Q_\mathrm{c}}}{\displaystyle \frac{\mathrm{arcsin}Q_\mathrm{c}}{Q_\mathrm{c}}}\right),`$
while for $`l=2`$ ($`d`$-wave):
$`\lambda _2`$ $`=`$ $`\lambda \left[{\displaystyle \frac{\pi k_\mathrm{F}}{q_\mathrm{c}}}\right]\theta (q_\mathrm{c}|𝐤𝐤^{}|)_{l=2}`$ (22)
$`=`$ $`\lambda (12Q_\mathrm{c}^2)\sqrt{1Q_\mathrm{c}^2},`$
$`\mu _2^{}`$ $`=`$ $`\mu ^{}\left[{\displaystyle \frac{\pi k_\mathrm{F}}{q_\mathrm{c}}}\right]\theta (|𝐤𝐤^{}|q_\mathrm{c})_{l=2}`$ (23)
$`=`$ $`\mu ^{}(12Q_\mathrm{c}^2)\sqrt{1Q_\mathrm{c}^2},`$
where $`\lambda =N_0g^2`$, $`\mu ^{}=N_0U^{}`$ and $`Q_\mathrm{c}=q_\mathrm{c}/2k_\mathrm{F}`$. Before we generalize the above expressions to include the nonadiabatic vertex corrections, it is useful to briefly examine qualitatively how the the magnitude of the cut-off parameter $`Q_\mathrm{c}`$ affects the gap symmetry. The total interaction in the gap equation (19) is weighted by $`\lambda _l\mu _l^{}`$. For the $`s`$-wave channel it reduces to:
$$\lambda _0\mu _0^{}=(\lambda +\mu ^{})\frac{\mathrm{arcsin}Q_\mathrm{c}}{Q_\mathrm{c}}\frac{\pi }{2Q_\mathrm{c}}\mu ^{},$$
(24)
while for the $`d`$-wave case $`l=2`$ it becomes:
$$\lambda _2\mu _2^{}=(\lambda +\mu ^{})(12Q_\mathrm{c}^2)\sqrt{1Q_\mathrm{c}^2}.$$
(25)
When $`Q_\mathrm{c}=1`$ (that is $`q_\mathrm{c}=2k_\mathrm{F}`$) the repulsive interaction (6) vanishes and the $`e`$-$`ph`$ coupling (2) becomes structureless. In this limit we expect the $`s`$-wave solution to dominate over the $`d`$-wave one. In fact, from (24) and (25), $`\lambda _0\mu _0^{}=\lambda `$ and $`\lambda _2\mu _2^{}=0`$. By lowering $`Q_\mathrm{c}`$, the total interaction acquires a momentum dependence, however $`\lambda _2\mu _2^{}`$ remains negative as long as $`1/\sqrt{2}<Q_\mathrm{c}<1`$. In this range therefore there is not $`d`$-wave solution. By further lowering of the cut-off parameter, the $`d`$-wave symmetry begins to compete with the $`s`$-wave one and for $`Q_\mathrm{c}1`$ $`\lambda _2\mu _2^{}\lambda +\mu ^{}`$ while $`\lambda _0\mu _0^{}\pi \mu ^{}/2Q_\mathrm{c}`$ signalling that the $`d`$-wave symmetry overcomes the $`s`$-wave ones.
In previous studies, we have shown that the nonadiabatic corrections lead to an enhancement of the critical temperature for small values of $`Q_\mathrm{c}`$ in the $`s`$-wave channel. However in the present model, small values of $`Q_\mathrm{c}`$ lead to a solution with $`d`$-wave symmetry and the question we face in the following sections is whether also in this symmetry the nonadiabatic corrections provide for an amplification of $`T_\mathrm{c}`$.
## 3 Non adiabatic vertex corrections
In this section we introduce the first corrections arising from the breakdown of Migdal’s theorem in the equations for the normal and anomalous self-energies. By following Ref.PSG , the first nonadiabatic corrections to the $`e`$-$`ph`$ interaction affect the normal state self-energy (8) in the following way:
$$\stackrel{~}{\mathrm{\Sigma }}_\mathrm{N}(k)=\underset{k^{}}{}\stackrel{~}{V}_\mathrm{N}(k,k^{})G(k^{}),$$
(26)
$$\stackrel{~}{V}_\mathrm{N}(k,k^{})=V(kk^{})\left[1+\underset{q}{}V(kq)G(qk+k^{})G(q)\right],$$
(27)
where the last term in the square bracket of Eq.(27) defines the vertex function. The off-diagonal self-energy in the nonadiabatic regime is instead modified as follows:
$`\stackrel{~}{\mathrm{\Sigma }}_\mathrm{S}(k)`$ $`=`$ $`{\displaystyle \underset{k^{}}{}}\left[\stackrel{~}{V}_\mathrm{S}(k,k^{})U^{}(kk^{})\right]`$ (28)
$`\times `$ $`G(k^{})G(k^{})\stackrel{~}{\mathrm{\Sigma }}_\mathrm{S}(k^{}),`$
$`\stackrel{~}{V}_\mathrm{S}(k,k^{})`$ $`=`$ $`V(kk^{})[1+{\displaystyle \underset{q}{}}V(kq)G(q)G(qk+k^{})`$
$`+`$ $`{\displaystyle \underset{q}{}}V(kq)G(q)G(q+kk^{})]`$
$`+`$ $`{\displaystyle \underset{q}{}}V(kq)V(qk^{})G(q)G(qkk^{}),`$
where $`q(𝐪,\omega _l)`$ and $`U^{}(kk^{})`$ is given by Eq.(6). The second and the third terms within the square brackets in Eq.(LABEL:vssum) correspond to the first order vertex corrections, while the last term corresponds to the cross scattering. These non adiabatic terms are shown in Fig. 2 in which we are only include these terms that give a finite contribution for $`T=T_\mathrm{c}`$.
In the vertex corrections there is simple one-phonon interaction, while in the cross term the sum refers to the product of both phonon propagators. This product of phonon propagators can be approximated as PSG :
$`{\displaystyle \frac{\omega _0^2}{(\omega _n\omega _l)^2+\omega _0^2}}{\displaystyle \frac{\omega _0^2}{(\omega _l\omega _m)^2+\omega _0^2}}`$
$`{\displaystyle \frac{\omega _0^2}{(\omega _n\omega _m)^2+\omega _0^2}}{\displaystyle \frac{\omega _0^2}{(\omega _n\omega _l)^2+\omega _0^2}}.`$ (30)
For the momentum dependence in the electron-phonon coupling we can approximate
$$|g(𝐤𝐪)|^2|g(𝐪𝐤^{})|^2|g(𝐤𝐤^{})|^2|g(𝐤𝐪)|^2.$$
(31)
This approximation is valid for relatively small values of the cut-off $`q_\mathrm{c}`$. Therefore the last term in Eq.(LABEL:vssum) reduces to:
$`{\displaystyle \underset{q}{}}V(kq)V(qk^{})G(q)G(qkk^{})`$
$`V(kk^{}){\displaystyle \underset{q}{}}V(kq)G(q)G(qkk^{}).`$ (32)
At this point it is useful to introduce a compact notation for the vertex and cross functions:
$$P_\mathrm{V}(k,k^{})\frac{1}{\lambda }\underset{q}{}V(kq)G(q)G(qk+k^{}),$$
(33)
$$P_\mathrm{C}(k,k^{})\frac{1}{\lambda }\underset{q}{}V(kq)G(q)G(qkk^{}).$$
(34)
Thus equations (27) and (LABEL:vssum) may be written in a simpler way as follows:
$`\stackrel{~}{V}_\mathrm{N}(k,k^{})`$ $`=`$ $`V(kk^{})\left[1+\lambda P_\mathrm{V}(k,k^{})\right],`$ (35)
$`\stackrel{~}{V}_\mathrm{S}(k,k^{})`$ $`=`$ $`V(kk^{})\left[1+2\lambda P_\mathrm{V}(k,k^{})+\lambda P_\mathrm{C}(k,k^{})\right].`$
(36)
In Eq.(28), the momentum dependence of $`\stackrel{~}{\mathrm{\Sigma }}_\mathrm{S}(k)`$ is transformed as in Eq.(13) and the interaction term $`\stackrel{~}{V}_\mathrm{S}(k,k^{})`$ is replaced by its angular weighted average:
$$\stackrel{~}{V}_\mathrm{S}(k,k^{})_l=\frac{1}{2\pi }_\pi ^\pi 𝑑\theta \stackrel{~}{V}_\mathrm{S}(\mathrm{cos}\theta )e^{il\theta },$$
(37)
which in terms of averaged vertex and cross corrections is expressed as:
$`\stackrel{~}{V}_\mathrm{S}(k,k^{})_l`$ $`=`$ $`V(kk^{})_l+2\lambda V(kk^{})P_\mathrm{V}(k,k^{})_l`$ (38)
$`+`$ $`\lambda V(kk^{})P_\mathrm{C}(k,k^{})_l.`$
The first term in the r.h.s. contains only the $`e`$-$`ph`$ interaction and the phonon propagator and it is simply given by:
$`V(kk^{})_l`$ $`=`$ $`g^2D(\omega _n\omega _m)\left[{\displaystyle \frac{\pi k_\mathrm{F}}{q_\mathrm{c}}}\right]\theta (q_\mathrm{c}|𝐤𝐤^{}|)_l`$ (39)
$`=`$ $`{\displaystyle \frac{\lambda _l}{N_0}}D(\omega _n\omega _m),`$
where $`\lambda _l`$ for $`l=0`$ and $`l=2`$ is given in equations (20) and (22), respectively. The second and third terms of (38) correspond instead to the momentum averages of the nonadiabatic corrections
$`V(kk^{})P_\mathrm{V}(k,k^{})_l`$ $`=`$ $`g^2D(\omega _n\omega _m)P_\mathrm{V}^l(\omega _n,\omega _m,q_\mathrm{c}),`$
$`V(kk^{})P_\mathrm{C}(k,k^{})_l`$ $`=`$ $`g^2D(\omega _n\omega _m)P_\mathrm{C}^l(\omega _n,\omega _m,q_\mathrm{c}),`$
where
$`P_\mathrm{V}^l(\omega _n,\omega _m,q_\mathrm{c})`$ $`=`$ $`\left[{\displaystyle \frac{\pi k_\mathrm{F}}{q_\mathrm{c}}}\right]\theta (q_\mathrm{c}|𝐤𝐤^{}|)P_\mathrm{V}(k,k^{})_l`$
$`P_\mathrm{C}^l(\omega _n,\omega _m,q_\mathrm{c})`$ $`=`$ $`\left[{\displaystyle \frac{\pi k_\mathrm{F}}{q_\mathrm{c}}}\right]\theta (q_\mathrm{c}|𝐤𝐤^{}|)P_\mathrm{C}(k,k^{})_l.`$
Analytic expressions of the vertex and cross functions together with their averages $`P_\mathrm{V}^l`$ and $`P_\mathrm{C}^l`$ for $`l=0`$ and $`l=2`$ are reported in Appendix and the results are shown in Figs. 2 and 3 as function of the adiabatic parameter $`\omega _0/E_\mathrm{F}`$ and for different values of dimensionless cut-off $`Q_\mathrm{c}=q_\mathrm{c}/2k_\mathrm{F}`$. All the curves have been obtained by setting $`\omega _n=0`$ and $`\omega _m=\omega _0`$ so that the exchanged frequency equals $`\omega _0`$. The behaviors, particularly at small $`Q_\mathrm{c}`$, of $`P_\mathrm{V}^l`$ and $`P_\mathrm{C}^l`$ are essentially independent of the particular symmetry. In fact for both $`l=0`$ ($`s`$-wave) and $`l=2`$ ($`d`$-wave) the nonadiabatic corrections are positive leading to an enhancement of the total $`e`$-$`ph`$ interaction. We expect therefore that, as for the $`s`$-wave case GPSprl ; PSG , also for the $`d`$-wave symmetry the vertex and cross corrections tend to amplify $`T_\mathrm{c}`$ when $`Q_\mathrm{c}`$ is sufficiently small.
To verify this point, we can write down the nonadiabatic Eliashberg equations for different symmetry channels. As in the previous section, the normal state self-energy (26) is averaged over the Fermi surface and, according to (35), $`\stackrel{~}{V}_\mathrm{N}(k,k^{})`$ reduces to:
$`\stackrel{~}{V}_\mathrm{N}(k,k^{})`$ $``$ $`\stackrel{~}{V}_\mathrm{N}(k,k^{})_{l=0}.`$
$`=`$ $`{\displaystyle \frac{D(\omega _n\omega _m)}{N_0}}\left[\lambda _0+\lambda ^2P_\mathrm{V}^{l=0}(\omega _n,\omega _m,q_\mathrm{c})\right],`$
and $`Z(\omega _n)=1\stackrel{~}{\mathrm{\Sigma }}_\mathrm{N}^{(0)}(\omega _n)/i\omega _n`$ becomes:
$`Z(\omega _n)=1{\displaystyle \frac{\pi T_\mathrm{c}}{\omega _n}}{\displaystyle \underset{m}{}}\left[\lambda _0+\lambda ^2P_\mathrm{V}^{l=0}(\omega _n,\omega _m,q_\mathrm{c})\right]`$
$`\times D(\omega _n\omega _m){\displaystyle \frac{\omega _m}{|\omega _m|}}{\displaystyle \frac{2}{\pi }}\mathrm{arctan}\left[{\displaystyle \frac{E/2}{|\omega _m|Z(\omega _m)}}\right].`$ (43)
Finally, the gap function for different symmetry channels is:
$`Z(\omega _n)\mathrm{\Delta }_l(\omega _n)=\pi T_\mathrm{c}{\displaystyle \underset{\omega _m}{}}[\lambda _l+2\lambda ^2P_\mathrm{V}^l(\omega _n,\omega _m,q_\mathrm{c})`$
$`+\lambda ^2P_\mathrm{C}^l(\omega _n,\omega _m,q_\mathrm{c})\mu _l^{}]D(\omega _n\omega _m){\displaystyle \frac{\mathrm{\Delta }_l(\omega _m)}{|\omega _m|}}`$
$`\times {\displaystyle \frac{2}{\pi }}\mathrm{arctan}\left[{\displaystyle \frac{E/2}{|\omega _m|Z(\omega _m)}}\right]`$ (44)
where $`\mu _l^{}`$ is given by Eqs.(21,23) and $`l=0,2`$.
To establish the range of $`Q_\mathrm{c}`$ values in which the $`d`$-wave symmetry is more stable than $`s`$-wave one and to quantify the effect of nonadiabaticity, we solve numerically the generalized Eliashberg equations (3) and (44) for $`l=0`$ and $`l=2`$. To find $`T_\mathrm{c}`$, we follow the maximum eigenvalue method described for example in PSG . The resulting values of $`T_\mathrm{c}`$ as a function of $`Q_\mathrm{c}`$ are shown in Fig. 4 for the $`d`$\- and $`s`$-wave symmetries. In the inset we show the critical temperature calculated without the vertex and cross corrections, i. e., for the ME equations (18) and (19). To display the crossover between the $`d`$\- and the $`s`$-wave symmetries more clearly, we show the results for $`\lambda =1`$ and $`\mu ^{}=0.1`$. When $`\mu ^{}\lambda `$ in fact the $`s`$-wave symmetry is suppressed by the strong repulsive interaction.
By reducing $`Q_\mathrm{c}`$, the $`s`$-wave solution (dashed lines) decreases monotonically when the vertex and cross corrections are not included (inset) while for the nonadiabatic case the corresponding $`T_\mathrm{c}`$ shows an upturn before falling to zero at $`Q_\mathrm{c}0`$. This latter feature is due to the nonadiabatic corrections which become more positive when $`Q_\mathrm{c}`$ is small. For lower values of $`Q_\mathrm{c}`$, however, the pseudopotential is dominant and $`T_\mathrm{c}`$ falls rapidly to zero. Contrary to the isotropic case, the $`d`$-wave solutions (solid lines) lead to critical temperatures which increase when $`Q_\mathrm{c}`$ is lowered. Since, as discussed before, the vertex corrections have a similar behavior both in $`d`$\- and in $`s`$ -wave symmetries when $`Q_\mathrm{c}`$ is small, the critical temperature in the nonadiabatic case is enhanced compared to the solution without vertex and cross corrections.
It is finally interesting to compare the present results with the phenomenology of the superconducting copper-oxides, which show $`d`$-wave, and the fullerides, which instead show $`s`$-wave. In our perspective there are important differencies between the two materials. A relevant one is that the oxides have their largest values of $`T_\mathrm{c}`$ when the Fermi surface is strongly influenced by Van Hove singularities. Then correlation effects can be estimated to be larger in the oxides and, finally, fullerides seem to have rotational disorder which would favour $`s`$-wave. Therefore, in principle, it could happen that in the oxides, going into the overdoped phase might lead to a crossover from $`d`$-wave to $`s`$-wave depending on the parameters.
## 4 Conclusions
In isotropic $`s`$-wave superconductors, the first nonadiabatic corrections to the $`e`$-$`ph`$ interaction such as vertex and cross functions are strongly dependent on the momentum transfer $`𝐪`$. In particular, small values of $`𝐪`$ leads to positive nonadiabatic corrections inducing an enhancement of the critical temperature $`T_\mathrm{c}`$ GPSprl ; PSG . Here, we have addressed the problem of the momentum dependence of the nonadiabatic corrections for a $`d`$-wave symmetry of the order parameter. By introducing a model interaction in which the $`e`$-$`ph`$ interaction is dominant at small values of $`\mathrm{q}`$ and the residual repulsion of electronic origin is instead important at larger momentum transfers, we have shown that also when the solution has $`d`$-wave symmetry, the inclusion of nonadiabatic corrections enhances $`T_\mathrm{c}`$ compared to the case without corrections. Therefore in a strongly correlated system, for which the $`e`$-$`ph`$ interaction is mainly of forward scattering, $`d`$-wave superconductivity driven by phonons can be sustained by the nonadiabatic corrections
## Appendix A Analytical calculation of vertex and cross functions
### A.1 Vertex function
The evaluation of the vertex function given in the Eq.(33) follows basically the same lines and approximations made in Ref.PSG , the main difference being that here we refer to a two-dimensional system rather than a three-dimensional one. Making use of the linear model for the electronic dispersion and considering the limit of $`T_\mathrm{c}/\omega _01`$, we obtain
$`P_\mathrm{V}(k,k^{})={\displaystyle \frac{\omega _0}{2\lambda }}{\displaystyle \underset{𝐪}{}}{\displaystyle \frac{|g(𝐤𝐪)|^2}{ϵ_𝐪ϵ_{𝐪𝐤+𝐤^{}}i\omega _n+i\omega _m}}`$
$`\times [{\displaystyle \frac{\theta (ϵ_𝐪)}{ϵ_𝐪+\omega _0i\omega _n}}{\displaystyle \frac{\theta (ϵ_𝐪)}{ϵ_𝐪\omega _0i\omega _n}}`$
$`+{\displaystyle \frac{\theta (ϵ_{𝐪𝐤+𝐤^{}})}{ϵ_{𝐪𝐤+𝐤^{}}+\omega _0i\omega _m}}+{\displaystyle \frac{\theta (ϵ_{𝐪𝐤+𝐤^{}})}{ϵ_{𝐪𝐤+𝐤^{}}\omega _0i\omega _m}}].`$ (45)
The main difficulty comes from $`ϵ_{𝐪𝐤+𝐤^{}}`$ which, within our model, is
$`ϵ_{𝐪𝐤+𝐤^{}}`$ $`=`$ $`v_\mathrm{F}[q^2+k^2+k^22qk\mathrm{cos}\alpha +2qk^{}\mathrm{cos}\beta `$ (46)
$``$ $`2kk^{}\mathrm{cos}\theta ]^{1/2}\mu ,`$
where $`\alpha `$, $`\beta `$ and $`\theta `$ are the angles between the directions of $`(𝐪,𝐤)`$, $`(𝐪,𝐤^{})`$ and $`(𝐤,𝐤^{})`$ respectively. In the limit of small $`q_\mathrm{c}`$, the presence of $`\theta `$-function in front of $`P_\mathrm{V}`$ \[Eq.(LABEL:mediaver2)\] and inside of the integral leads to $`|𝐤||𝐤^{}|`$ and $`|𝐤||𝐪|`$. Therefore Eq. (46) can be rewritten as follows:
$$ϵ_{𝐪𝐤+𝐤^{}}ϵ_𝐪+v_\mathrm{F}k_\mathrm{F}\left[1\mathrm{cos}\alpha +\mathrm{cos}\beta \mathrm{cos}\theta \right],$$
(47)
where we have taken $`|𝐤|k_\mathrm{F}`$. We can relate the angle $`\beta `$ to $`\alpha `$ and $`\theta `$ by means of relation $`\beta =\theta \alpha `$. Therefore the Eq. (47) becomes:
$$ϵ_{𝐪𝐤+𝐤^{}}ϵ_𝐪+v_\mathrm{F}k_\mathrm{F}\left[(1\mathrm{cos}\alpha )(1\mathrm{cos}\theta )+\mathrm{sin}\alpha \mathrm{sin}\theta \right].$$
(48)
If $`Q=|𝐤𝐤^{}|/2k_\mathrm{F}`$ then $`\mathrm{cos}\theta =12Q^2`$ and $`\mathrm{sin}\theta =2Q\sqrt{1Q^2}`$:
$$ϵ_{𝐪𝐤+𝐤^{}}ϵ_𝐪+EQ^2(1\mathrm{cos}\alpha )+EQ\sqrt{1Q^2}\mathrm{sin}\alpha .$$
(49)
Expanding $`\mathrm{cos}\alpha `$ and $`\mathrm{sin}\alpha `$ in powers of $`\alpha `$ and retaining only the lowest order term in $`\alpha `$ and $`Q`$ we finally obtain:
$$ϵ_{𝐪𝐤+𝐤^{}}ϵ_𝐪+EQ\alpha .$$
(50)
For small $`q_\mathrm{c}`$ we can replace
$`\theta (q_\mathrm{c}|𝐤𝐪|)`$ $``$ $`\theta (q_\mathrm{c}k_\mathrm{F}\sqrt{2(1\mathrm{cos}\alpha )})`$ (51)
$``$ $`\theta (2Q_\mathrm{c}|\alpha |).`$
At this point it is convenient to transform the momentum integration into an energy integration:
$$\frac{d^2q}{(2\pi )^2}=N_0_\pi ^\pi \frac{d\alpha }{2\pi }_{E/2}^{E/2}𝑑ϵ$$
(52)
where we have used a constant DOS $`N(ϵ)=N_0`$ in the range $`E/2ϵE/2`$. The integration over the energy $`ϵ`$ and over the angle $`\alpha `$ can be performed analytically and the final expression of the vertex function in the limit of small $`Q_\mathrm{c}`$ is given by
$`P_\mathrm{V}(k,k^{})`$ $`=`$ $`\omega _0B(\omega _n,\omega _m)+{\displaystyle \frac{\omega _0}{2Q_\mathrm{c}}}{\displaystyle \frac{1}{EQ}}\mathrm{arctan}\left({\displaystyle \frac{2Q_\mathrm{c}EQ}{|\omega _n\omega _m|}}\right)`$ (53)
$`\times `$ $`{\displaystyle \frac{A(\omega _n,\omega _m)B(\omega _n,\omega _m)(\omega _n\omega _m)^2}{|\omega _n\omega _m|}},`$
where $`Q=|𝐤𝐤^{}|/2k_\mathrm{F}`$, $`Q_\mathrm{c}=q_\mathrm{c}/2k_\mathrm{F}`$, and
$`A(\omega _n,\omega _m)`$ $`=`$ $`(\omega _n\omega _m)[\mathrm{arctan}\left({\displaystyle \frac{\omega _n}{\omega _0}}\right)\mathrm{arctan}\left({\displaystyle \frac{\omega _m}{\omega _0}}\right)`$
$`+`$ $`\mathrm{arctan}\left({\displaystyle \frac{\omega _m}{\omega _0+E/2}}\right)\mathrm{arctan}\left({\displaystyle \frac{\omega _n}{\omega _0+E/2}}\right)],`$
$$B(\omega _n,\omega _m)=(\omega _0+E/2)\frac{(\omega _0+E/2)^2+2\omega _m^2\omega _n\omega _m}{\left[(\omega _0+E/2)^2+\omega _m^2\right]^2}.$$
(55)
### A.2 Cross function
The function $`P_\mathrm{C}(k,k^{})`$, given by the Eq.(34), can be explicitly evaluated within the same scheme of calculation of the vertex function. In the limit of $`T_\mathrm{c}/\omega _01`$ we have:
$`P_\mathrm{C}(k,k^{})={\displaystyle \frac{\omega _0}{2\lambda }}{\displaystyle \underset{𝐪}{}}{\displaystyle \frac{|g(𝐤𝐪)|^2}{ϵ_𝐪ϵ_{𝐪𝐤𝐤^{}}i\omega _ni\omega _m}}`$ (56)
$`\times [{\displaystyle \frac{\theta (ϵ_𝐪)}{ϵ_𝐪+\omega _0i\omega _n}}{\displaystyle \frac{\theta (ϵ_𝐪)}{ϵ_𝐪\omega _0i\omega _n}}`$
$`+{\displaystyle \frac{\theta (ϵ_{𝐪𝐤𝐤^{}})}{ϵ_{𝐪𝐤𝐤^{}}+\omega _0+i\omega _m}}+{\displaystyle \frac{\theta (ϵ_{𝐪𝐤𝐤^{}})}{ϵ_{𝐪𝐤𝐤^{}}\omega _0+i\omega _m}}].`$
The electron energy $`ϵ_{𝐪𝐤𝐤^{}}`$ can be approximated for $`q_\mathrm{c}2k_\mathrm{F}`$ as follows:
$`ϵ_{𝐪𝐤𝐤^{}}`$ $`=`$ $`v_\mathrm{F}[q^2+k^2+k^22qk\mathrm{cos}\alpha 2qk^{}\mathrm{cos}\beta `$ (57)
$`+`$ $`2kk^{}\mathrm{cos}\theta ]^{1/2}\mu `$
$``$ $`ϵ_𝐪+E(1Q^2){\displaystyle \frac{\alpha ^2}{2}}EQ\sqrt{1Q^2}\alpha .`$
The integrations over the energy and the angle are elementary, the final expression of the cross function is however quite complicated:
$`P_\mathrm{C}(k,k^{})`$ $`=`$ $`\omega _0B(\omega _n,\omega _m){\displaystyle \frac{\omega _0}{2Q_\mathrm{c}}}{\displaystyle \frac{1}{E\sqrt{1Q^2}\rho (k,k^{})}}`$ (58)
$`\times `$ $`\left\{\mathrm{cos}[\eta (k,k^{})]C(k,k^{})+\mathrm{sin}[\eta (k,k^{})]D(k,k^{})\right\}`$
$`\times `$ $`{\displaystyle \frac{A(\omega _n,\omega _m)B(\omega _n,\omega _m)(\omega _n+\omega _m)^2}{|\omega _n+\omega _m|}},`$
where the functions $`A`$ and $`B`$ are the same of Eqs. (A.1) and (55) with $`\omega _m\omega _m`$. The function $`C`$, $`D`$, $`\eta `$, $`\rho `$ are given by
$$\rho (k,k^{})=\left[Q^4+\left(2\frac{\omega _n+\omega _m}{E}\right)^2\right]^{1/4},$$
(59)
$$\eta (k,k^{})=\frac{1}{2}\mathrm{arctan}\left(\frac{2|\omega _n+\omega _m|}{EQ^2}\right),$$
(60)
$`C(k,k^{})=b(k,k^{})^2\{\mathrm{arctan}\left[{\displaystyle \frac{a_+(k,k^{})}{b(k,k^{})}}\right]`$
$`+\mathrm{arctan}\left[{\displaystyle \frac{a_{}(k,k^{})}{b(k,k^{})}}\right]+\mathrm{arctan}\left[{\displaystyle \frac{2Q_\mathrm{c}a_+(k,k^{})}{b(k,k^{})}}\right]`$
$`+\mathrm{arctan}\left[{\displaystyle \frac{2Q_\mathrm{c}a_{}(k,k^{})}{b(k,k^{})}}\right]\},`$ (61)
$`D(k,k^{})=`$ $`{\displaystyle \frac{1}{2}}\mathrm{ln}\{{\displaystyle \frac{[2Q_\mathrm{c}a_+(k,k^{})]^2+b(k,k^{})^2}{[2Q_\mathrm{c}a_{}(k,k^{})]^2+b(k,k^{})^2}}`$ (62)
$`\times {\displaystyle \frac{a_{}(k,k^{})^2+b(k,k^{})^2}{a_+(k,k^{})^2+b(k,k^{})^2}}\},`$
$$a_\pm (k,k^{})=\frac{Q\pm \rho (k,k^{})\mathrm{cos}[\eta (k,k^{})]}{\sqrt{1Q^2}},$$
(63)
$$b(k,k^{})=\frac{\rho (k,k^{})}{\sqrt{1Q^2}}\mathrm{sin}[\eta (k,k^{})].$$
(64)
### A.3 $`s`$\- and $`d`$-wave averages
In what follows we report the final expressions of the $`s`$-wave and $`d`$-wave averages of the vertex and cross functions defined in Eq.(LABEL:mediaver2) for $`l=0`$ ($`s`$-wave) and $`l=2`$ ($`d`$-wave), where $`P_\mathrm{V}`$ and $`P_\mathrm{C}`$ are given by Eqs. (53) and (58), respectively.
$`P_\mathrm{V}^{l=0}(\omega _n,\omega _m;q_\mathrm{c})={\displaystyle \frac{A(\omega _n,\omega _m)B(\omega _n,\omega _m)(\omega _n\omega _m)^2}{|\omega _n\omega _m|}}`$
$`\times {\displaystyle \frac{\omega _0}{2EQ_c^2}}F_1(\omega _n,\omega _m,Q_\mathrm{c})+B(\omega _n,\omega _m){\displaystyle \frac{\mathrm{arcsin}Q_\mathrm{c}}{Q_\mathrm{c}}}`$ (65)
$`P_\mathrm{C}^{l=0}(\omega _n,\omega _m;q_\mathrm{c})`$
$`={\displaystyle \frac{A(\omega _n,\omega _m)B(\omega _n,\omega _m)(\omega _n+\omega _m)^2}{|\omega _n+\omega _m|}}`$
$`\times {\displaystyle \frac{\omega _0}{2EQ_c^2}}F_2(\omega _n,\omega _m,Q_\mathrm{c})+B(\omega _n,\omega _m){\displaystyle \frac{\mathrm{arcsin}Q_\mathrm{c}}{Q_\mathrm{c}}}`$ (66)
$`P_{rmV}^{l=2}(\omega _n,\omega _m;q_\mathrm{c})={\displaystyle \frac{A(\omega _n,\omega _m)B(\omega _n,\omega _m)(\omega _n\omega _m)^2}{|\omega _n\omega _m|}}`$
$`\times {\displaystyle \frac{\omega _0}{2EQ_c^2}}F_3(\omega _n,\omega _m,Q_\mathrm{c})+B(\omega _n,\omega _m)(12Q_c^2)\sqrt{1Q_c^2}`$
$`P_\mathrm{C}^{l=2}(\omega _n,\omega _m;q_\mathrm{c})`$
$`={\displaystyle \frac{A(\omega _n,\omega _m)B(\omega _n,\omega _m)(\omega _n+\omega _m)^2}{|\omega _n+\omega _m|}}`$
$`\times {\displaystyle \frac{\omega _0}{2EQ_c^2}}F_4(\omega _n,\omega _m,Q_\mathrm{c})+B(\omega _n,\omega _m)(12Q_c^2)\sqrt{1Q_c^2}`$
where
$$F_1(\omega _n,\omega _m,Q_\mathrm{c})=_0^{Q_\mathrm{c}}\frac{dQ}{Q\sqrt{1Q^2}}\mathrm{arctan}\left(\frac{2EQ_\mathrm{c}Q}{|\omega _n+\omega _m|}\right),$$
(69)
$`F_2(\omega _n,\omega _m,Q_\mathrm{c})={\displaystyle _0^{Q_\mathrm{c}}}{\displaystyle \frac{dQ}{Q}}\left({\displaystyle \frac{1}{1Q^2}}\right){\displaystyle \frac{1}{\rho (k,k^{})}}`$
$`\times \left\{C(k,k^{})\mathrm{cos}[\eta (k,k^{})]D(k,k^{})\mathrm{sin}[\eta (k,k^{})]\right\}`$ (70)
$`F_3(\omega _n,\omega _m,Q_\mathrm{c})`$ $`=`$ $`{\displaystyle _0^{Q_\mathrm{c}}}{\displaystyle \frac{dQ}{Q}}\left({\displaystyle \frac{1+8Q^48Q^2}{\sqrt{1Q^2}}}\right)`$ (71)
$`\times `$ $`\mathrm{arctan}\left({\displaystyle \frac{2EQ_\mathrm{c}Q}{|\omega _n+\omega _m|}}\right),`$
$`F_4(\omega _n,\omega _m,Q_\mathrm{c})={\displaystyle _0^{Q_\mathrm{c}}}{\displaystyle \frac{dQ}{Q}}\left({\displaystyle \frac{1+8Q^48Q^2}{1Q^2}}\right){\displaystyle \frac{1}{\rho (k,k^{})}}`$ (72)
$`\times `$ $`\left\{C(k,k^{})\mathrm{cos}[\eta (k,k^{})]D(k,k^{})\mathrm{sin}[\eta (k,k^{})]\right\}`$
The functions $`C`$, $`D`$, $`\eta `$ and $`\rho `$ are precedently defined.
|
warning/0005/nucl-th0005046.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The renormalization of the hadron properties in a nuclear medium is the object of continuous attentions. Particularly, the properties of the $`\rho `$ meson in nuclei have been thoroughly studied and many experiments are devoted to observe these modifications . Comparatively, the $`\varphi `$ meson has received little attention but the studies of Refs. predict a very small shift, if any, and a substantially increased width. Yet, the study of the $`\varphi `$ width in a nuclear medium is a very interesting question since it should offer information on the renormalization of the kaon properties in a medium, a subject itself which attracts much interest and which is related to the possible existence of kaon condensates in neutron stars .
The many body problem of the interaction of the $`\overline{K}`$ with a nuclear medium is a subtle one. The low density theorem leads to a repulsive $`\overline{K}`$ selfenergy, but analysis of kaonic atoms demands an attractive one . A step forward in the understanding of this peculiar feature is given in Refs. where it is shown that the Pauli blocking of intermediate states in the $`\overline{K}N`$ interaction leads to a shift of the $`\mathrm{\Lambda }(1405)`$ resonance (which lies just below the $`\overline{K}N`$ threshold) to higher energies. As a consequence the $`\overline{K}N`$ interaction becomes attractive. On the other hand, the attraction felt by the kaons has an opposite effect because it brings the resonance back to lower energies. This fact stimulated a selfconsistent calculation in Ref. which showed that the position of the $`\mathrm{\Lambda }(1405)`$ resonance was not changed in the medium but altogether the $`\overline{K}`$ still felt an attraction and the $`\mathrm{\Lambda }(1405)`$ resonance broadened considerably. The calculations of Ref. introduced in addition the renormalization of the pions and the baryons in the intermediate states, and found again that the position of the resonance barely changed when the $`\overline{K}`$ selfenergy was calculated selfconsistently while the width became even larger. The $`\overline{K}`$-nucleus potential of Ref. has been used in the study of Ref. where it was found to be compatible with the present data of kaonic atoms.
The important changes found in the kaon selfenergy in the medium, which is the key ingredient here, advise a reevaluation of the $`\varphi `$ selfenergy and this is the purpose of the present work. The framework for the evaluation of the $`\varphi `$ selfenergy in a nuclear medium was developed in Ref. and we will adhere to it, yet following a different technical approach.
## 2 $`\mathit{\varphi }`$ decay in the nuclear medium
We will use here the gauge vector representation of the vector field $`\varphi `$. The alternative tensor formulation of Ref. is equivalent to the gauge vector one but makes the chiral counting easier. Yet, in the present problem one needs Lagrangians for the coupling of the mesons to baryons which are available in the vector representation but not in the tensor one and hence we shall adhere to the vector representation as done in Refs. .
In free space the $`\varphi `$ meson decays into $`K\overline{K}`$ and $`3\pi `$. The latter channel is OZI forbidden and in spite of the large phase space only accounts for 15% of the total width. The coupling of the $`\varphi `$ to $`K\overline{K}`$ is large and it is only the reduced phase space what makes the $`\varphi `$ width small. However, we shall see that thanks to the $`\overline{K}`$ related channels in the medium, the $`\varphi `$ width becomes of the order of 22 MeV at normal nuclear matter density ($`\rho _0=0.17`$ fm<sup>-3</sup>) and the $`3\pi `$ channel represents only 3% of this width. Thus, we shall neglet it in our study.
The Lagrangian that describes the coupling of the $`\varphi `$ meson to kaons is given by
$$=ig_\varphi \varphi _\mu (K^{}_\mu K^+K^+_\mu K^{}+\overline{K}^0_\mu K^0K^0_\mu \overline{K}^0),$$
(1)
which provides the following $`\varphi K^+K^{}`$ vertex (matrix element of $`i`$)
$$V_{\varphi K^+K^{}}=ig_\varphi ϵ_\mu (p^\mu p^\mu ),$$
(2)
with $`p,p^{}`$ the momenta of the $`K^+,K^{}`$ mesons, respectively. An analogous vertex in found for $`K^0\overline{K}^0`$.
We shall work in a frame where the $`\varphi `$ is at rest and in a gauge where $`ϵ^0=0`$. If we consider the $`\varphi `$ selfenergy diagram of Fig. 1 with $`K^+K^{}`$ as intermediate states we obtain
$$\mathrm{\Pi }_\varphi =i\frac{d^4q}{(2\pi )^4}\frac{1}{q^2m_K^2+i\epsilon }\frac{1}{(Pq)^2m_K^2+i\epsilon }g_\varphi ^2ϵ_iϵ_j2q_i2q_j,$$
(3)
where $`P`$ is the fourmomentum of the $`\varphi `$. The imaginary part of $`\mathrm{\Pi }`$ is readily evaluated using Cutkosky rules
$`\mathrm{\Pi }_\varphi `$ $``$ $`2i\mathrm{Im}\mathrm{\Pi }_\varphi `$
$`D(q)`$ $``$ $`2i\theta (q^0)\mathrm{Im}D(q)`$ (4)
$`D(Pq)`$ $``$ $`2i\theta (P^0q^0)\mathrm{Im}D(Pq),`$ (5)
where $`D(q)`$ is the kaon propagator. Then, using the relationship
$$\mathrm{\Gamma }_\varphi =\frac{\mathrm{Im}\mathrm{\Pi }_\varphi }{M_\varphi },$$
(6)
with $`M_\varphi `$ the $`\varphi `$ mass, we obtain the free width of the $`\varphi `$ for $`K^+K^{}`$ decay
$$\mathrm{\Gamma }_\varphi =\frac{2}{3}\frac{1}{4\pi }g_\varphi ^2\frac{p^3}{M_\varphi ^2},$$
(7)
which requires the value $`g_\varphi =4.57`$ to obtain agreement with the experimental value of 2.18 MeV .
In the nuclear medium the kaon acquires a selfenergy, $`\mathrm{\Pi }_K(q^0,\stackrel{}{q},\rho )`$, and the kaon propagator reads
$$D(q,\rho )=\frac{1}{q^{\mathrm{0\hspace{0.17em}2}}\stackrel{}{q}^2m_K^2\mathrm{\Pi }_K(q^0,\stackrel{}{q},\rho )},$$
(8)
which can also be cast in terms of the Lehmann representation as
$$D(q,\rho )=_0^{\mathrm{}}𝑑\omega \mathrm{\hspace{0.17em}2}\omega \frac{S_K(\omega ,\stackrel{}{q},\rho )}{q^{\mathrm{0\hspace{0.17em}2}}\omega ^2+i\epsilon },$$
(9)
where $`S_K`$ is the kaon spectral function
$$S_K(p,\rho )=\frac{1}{\pi }\mathrm{Im}D(p,\rho ).$$
(10)
At this point it is worth differentiating between the $`\overline{K}(\overline{K}^0,K^{})`$ and the $`K(K^+,K^0)`$ meson doublets.
The $`K^+N`$ and $`K^0N`$ interactions are smooth, there are no resonances with strangeness $`S=1`$ and the selfenergy of the $`K^+`$ meson at small energies is well aproximated by the $`t\rho `$ approximation, which from Refs. is given by
$$\mathrm{\Pi }_{K^+}=\frac{1}{2}(t_{K^+p}+t_{K^+n})\rho _0\frac{\rho }{\rho _0}=0.13m_K^2\frac{\rho }{\rho _0}.$$
(11)
An identical expression is found for the $`K^0`$ meson.
The $`\overline{K}`$ case is far more subtle. As we pointed out, the $`\overline{K}N`$ interaction at low energies is dominated by the $`\mathrm{\Lambda }(1405)`$ resonance which appears just below the $`\overline{K}N`$ threshold. Any realistic approach to the problem has do deal properly with this resonance, which is actually one of the resonances which meets with more problems within quark models of baryons. Not surprisingly, the resonance is closely tied to the interaction of the $`\overline{K}N`$ system with related channels and is generated within unitary coupled-channel schemes . An important step forward has been given with the use of chiral Lagrangians within a coupled channel unitary approach , by means of which one is able to reproduce the $`\mathrm{\Lambda }(1405)`$ resonance and the cross sections for $`K^{}p`$ scattering to the different coupled channels. This succesful scheme is most appropriate to undertake the study of the kaon selfenergy in a nuclear medium and this was done in Refs. . As quoted above, with respect to the work of Ref. , the work of Ref. implemented the selfenergy of the kaon in the calculation in a selfconsistent way, with the important finding that the position of the $`\mathrm{\Lambda }(1405)`$ resonance was barely moved and the $`K^{}`$ experienced a moderate attraction. The work of Ref. improves on the one of Ref. by including the selfenergy of the pions and the baryons in the intermediate states, thus opening more channels for reactions of the $`K^{}`$ in the medium, consequently increasing the width of the $`\mathrm{\Lambda }(1405)`$ resonance and widening considerably the spectral function of the $`K^{}`$ (see also Ref. for a recent review on this and related issues). Clearly, the spread of the kaon strength necessarily should have repercussion on the $`\varphi `$ decay in the medium since it automatically enlarges the phase space for the $`\varphi `$ decay into the kaon related channels.
The low energy coupled channel equations studied in Refs. deal only with the s-wave part of the $`\overline{K}N`$ interaction. A recent study of the p-wave $`K^{}N`$ interaction at low energies within the context of chiral Lagrangians has been done in Ref. . Although the energies of the kaons in the $`\varphi `$ decay are not large and hence the s-wave selfenergy plays the major role in the process, we nevertheless take the p-wave into account by including the excitation of the $`\overline{K}`$ into $`\mathrm{\Lambda }h,\mathrm{\Sigma }h`$ and $`\mathrm{\Sigma }^{}h`$ states, as done in Refs. , where $`\mathrm{\Sigma }^{}`$ is the $`\mathrm{\Sigma }(1385)`$ resonance . The p-wave selfenergy plays a role in the decay of the $`\varphi `$ into $`K^+Yh`$ states, while the s-wave selfenergy contributes essentially to the decay into $`K^+\mathrm{\Sigma }\pi h`$ states.
In the present work we use the s-wave $`K^{}`$ selfenergy from Ref. and implement the p-wave selfenergy due to $`\mathrm{\Lambda }h,\mathrm{\Sigma }h`$ states, as also done in Ref. . As just mentioned, we also include here the coupling of the $`\overline{K}`$ into $`\mathrm{\Sigma }^{}h`$ excitations. Its addition in the kaon loops in the selfconsistent calculation of the s-wave $`K^{}`$ selfenergy barely modifies the results of Ref. . We will see, however, that as an extra contribution to the p-wave selfenergy it induces an increase in the $`\varphi `$ width, since it incorporates the $`\varphi K^+\mathrm{\Sigma }^{}h`$ decay channel.
In order to evaluate the $`\varphi `$ selfenergy in the medium we write the $`K^+`$ propagator as in eq. (8) using explicitly the $`K^+`$ selfenergy of eq. (11). For the $`K^{}`$ propagator we use instead the Lehmann representation of eq. (9) and thus, following the same steps that led us to eq. (7), we obtain a $`\varphi `$ width in the medium given by
$`\mathrm{\Gamma }_\varphi (P^0,\rho )`$ $`=`$ $`{\displaystyle \frac{1}{M_\varphi }}{\displaystyle _0^{\omega _{\mathrm{max}}^{(+)}}}𝑑\omega S_K^{}(\omega ,q^{(+)},\rho ){\displaystyle \frac{1}{3\pi }}g_\varphi ^2q^{(+)\mathrm{\hspace{0.17em}3}}`$ (12)
$`+`$ $`{\displaystyle \frac{1}{M_\varphi }}{\displaystyle _0^{\omega _{\mathrm{max}}^{(0)}}}𝑑\omega S_{\overline{K}^0}(\omega ,q^{(0)},\rho ){\displaystyle \frac{1}{3\pi }}g_\varphi ^2q^{(0)\mathrm{\hspace{0.17em}3}},`$
where
$`q^{(+)}=\sqrt{(P^0\omega )^2\mathrm{\Pi }_{K^+}m_{K^+}^2},`$ $`\omega _{\mathrm{max}}^{(+)}=P^0\sqrt{m_{K^+}^2+\mathrm{\Pi }_{K^+}}`$
$`q^{(0)}=\sqrt{(P^0\omega )^2\mathrm{\Pi }_{K^0}m_{K^0}^2},`$ $`\omega _{\mathrm{max}}^{(0)}=P^0\sqrt{m_{K^0}^2+\mathrm{\Pi }_{K^0}}.`$ (13)
The vertex $`\overline{K}NY`$ for an incoming kaon of momentum $`\stackrel{}{k}`$ is given by
$$V_{\overline{K}NY}=\stackrel{~}{V}_{\overline{K}NY}\stackrel{}{\sigma }\stackrel{}{k}=\left[\alpha \frac{D+F}{2f}+\beta \frac{DF}{2f}\right]\stackrel{}{\sigma }\stackrel{}{k},$$
(14)
where $`\alpha `$ and $`\beta `$ are given in Table 1 and $`D+F=g_A=1.257`$, $`DF=0.33`$. We take $`f=1.15f_\pi `$ with $`f_\pi =93`$ MeV, a value which lies in between the pion and kaon weak decay constants and that was chosen in the model of Ref. to optimize the position of the $`\mathrm{\Lambda }(1405)`$ resonance.
Similarly, the $`\overline{K}N\mathrm{\Sigma }^{}`$ vertex is given by
$$V_{\overline{K}N\mathrm{\Sigma }^{}}=\stackrel{~}{V}_{\overline{K}N\mathrm{\Sigma }^{}}\stackrel{}{S}^{}\stackrel{}{k}=\frac{g_\mathrm{\Sigma }^{}}{2M}A\stackrel{}{S}^{}\stackrel{}{k},$$
(15)
where $`\stackrel{}{S}^{}`$ is the spin transition operator from spin 1/2 to spin 3/2. The coefficient A is given in Table 2 and the coupling $`g_\mathrm{\Sigma }^{}/2M`$ is evaluated here by first using the SU(6) quark model to relate the $`\pi NN`$ coupling to the $`\pi N\mathrm{\Delta }`$ one and then using SU(3) symmetry to relate the $`\pi N\mathrm{\Delta }`$ coupling to the $`\overline{K}N\mathrm{\Sigma }^{}`$ one, since the $`\mathrm{\Sigma }^{}`$ belongs to the SU(3) decouplet of the $`\mathrm{\Delta }`$. We obtain:
$$\frac{g_\mathrm{\Sigma }^{}}{2M}=\frac{2\sqrt{6}}{5}\frac{D+F}{2f}.$$
(16)
There is hence a small diversion of our approach here with respect to the one of Ref. , where SU(3) arguments are used and, consequently, a different coupling was obtained. We note that with the coupling used in the present work one gets a good reproduction of the data for the decay of the $`\mathrm{\Sigma }^{}`$ into $`\mathrm{\Lambda }\pi `$ and $`\mathrm{\Sigma }\pi `$ final statesx, when the the value of $`f`$ is taken as $`f_\pi `$.
The p-wave $`\overline{K}`$ selfenergy in symmetric nuclear matter can then be written as
$`\mathrm{\Pi }_{\overline{K}}^{(p)}(q^0,\stackrel{}{q},\rho )`$ $`=`$ $`\stackrel{~}{\mathrm{\Pi }}_{\overline{K}}(q,\rho )\stackrel{}{q}^2`$
$`=`$ $`{\displaystyle \frac{1}{2}}\stackrel{~}{V}_{K^{}p\mathrm{\Lambda }}^2f_\mathrm{\Lambda }^2\stackrel{}{q}^2U_\mathrm{\Lambda }(q^0,\stackrel{}{q},\rho )`$
$`+`$ $`{\displaystyle \frac{3}{2}}\stackrel{~}{V}_{K^{}p\mathrm{\Sigma }^0}^2f_\mathrm{\Sigma }^2\stackrel{}{q}^2U_\mathrm{\Sigma }(q^0,\stackrel{}{q},\rho )`$
$`+`$ $`{\displaystyle \frac{1}{2}}\stackrel{~}{V}_{K^{}p\mathrm{\Sigma }^0}^2f_\mathrm{\Sigma }^{}^2\stackrel{}{q}^2U_\mathrm{\Sigma }^{}(q^0,\stackrel{}{q},\rho ),`$
where the Lindhard function $`U_Y(q)`$ $`(Y=\mathrm{\Lambda },\mathrm{\Sigma }`$ or $`\mathrm{\Sigma }^{}`$) is given by
$`\mathrm{Re}U_Y(q^0,\stackrel{}{q},\rho )`$ $`=`$ $`{\displaystyle \frac{3}{2}}\rho {\displaystyle \frac{M_Y}{qp_F}}\left\{z+{\displaystyle \frac{1}{2}}(1z^2)\mathrm{ln}{\displaystyle \frac{z+1}{z1}}\right\}`$
$`\mathrm{Im}U_Y(q^0,\stackrel{}{q},\rho )`$ $`=`$ $`\pi {\displaystyle \frac{3}{4}}\rho {\displaystyle \frac{M_Y}{qp_F}}\left\{(1z^2)\theta (1z)\right\}`$ (18)
$`z`$ $`=`$ $`\left(q^0{\displaystyle \frac{q^2}{2M_Y}}(M_YM)\right){\displaystyle \frac{M_Y}{qp_F}},`$
with $`\rho `$ the nuclear matter density, $`p_F`$ the Fermi momentum and $`f_\mathrm{\Lambda }=(1q^0/2M_\mathrm{\Lambda })`$, $`f_\mathrm{\Sigma }=(1q^0/2M_\mathrm{\Sigma })`$, $`f_\mathrm{\Sigma }^{}=(1q^0/M_\mathrm{\Sigma }^{})`$ relativistic recoil vertex corrections .
As in Ref. , we also incorporate a form-factor at the kaon-baryon vertices of dipole type, $`[\mathrm{\Lambda }^2/(\mathrm{\Lambda }^2q^2)]^2`$, with $`\mathrm{\Lambda }=1.05`$ GeV.
Inserting the p-wave $`K^{}`$ selfenergy at the lowest order in the nuclear density accounts for the contribution to the in medium $`\varphi `$ selfenergy depicted in diagram a) of Fig. 2. However, it was already noticed in Refs. that, when dealing with gauge vector mesons, there are other vertex corrections associated to these diagrams which are required by the gauge invariance of the model. These diagrams are depicted in Figs. 2b), c) and d). These terms appear from a contact $`\varphi KNY`$ term, similar to the Kroll Ruderman term for photons, and they can be evaluated systematically by substituting $`eQA_\mu `$ by $`gV_\mu /2`$ in the chiral Lagrangians, as done in Ref. , with $`g={\displaystyle \frac{\sqrt{2}}{3}}g_\varphi `$, where an extra minus sign is already incorporated in the Lagrangian of eq. (1). We thus obtain the following vertex function for these contact terms
$`V_{\varphi KNY}`$ $`=`$ $`g_\varphi \stackrel{~}{V}_{\overline{K}NY}\stackrel{}{\sigma }\stackrel{}{ϵ}(\varphi );Y=\mathrm{\Lambda },\mathrm{\Sigma }`$
$`V_{\varphi KN\mathrm{\Sigma }^{}}`$ $`=`$ $`g_\varphi \stackrel{~}{V}_{\overline{K}N\mathrm{\Sigma }^{}}\stackrel{}{S}^{}\stackrel{}{ϵ}(\varphi ).`$ (19)
The contribution of the vertex corrections can be easily evaluated. We can see that the addition of all the one-baryon loop diagrams in Fig. 2 just replaces the contribution of the one in Fig. 2a), $`D(q)\stackrel{~}{\mathrm{\Pi }}(q)\stackrel{}{q}^2D(q)`$, by
$$D(q)\stackrel{~}{\mathrm{\Pi }}(q)\stackrel{}{q}^2D(q)+\frac{1}{2}\stackrel{~}{\mathrm{\Pi }}(q)D(q)+\frac{1}{2}D(q)\stackrel{~}{\mathrm{\Pi }}(q)+\frac{3}{4}\frac{\stackrel{~}{\mathrm{\Pi }}(q)}{\stackrel{}{q}^2}.$$
(20)
One simplifying step forward is possible at this point since we are interested in the imaginary part of the selfenergy. By means of Cutkosky rules one knows that the contributions to the imaginary part are obtained when placing either the $`K^+K^{}`$ or the $`K^+Yh`$ of the intermediate states on shell. In both cases the $`K^+`$ appears on shell, which means that we can substitute $`P^0q^0`$ by $`\omega (q)`$, hence $`q^0=P^0\omega (q)`$, in eq. (20). It is then easy to see that the sum of all diagrams in Fig. 2 is equivalent to considering only diagram a) but with the following substitution for the kaon p-wave selfenergy
$$\stackrel{~}{\mathrm{\Pi }}(q)\stackrel{}{q}^2\stackrel{~}{\mathrm{\Pi }}(q)[(P^0\omega (q))^2m_K^2]\left\{1+\frac{3}{4}\frac{[(P^0\omega (q))^2\stackrel{}{q}^2m_K^2]^2}{\stackrel{}{q}^2[(P^0\omega (q))^2m_K^2]}\right\}.$$
(21)
The arguments given above apply to just the p-wave diagrams at lowest order in the density, i.e. those displayed in Fig. 2. In the actual calculation, however, we simply replace the p-wave selfenergy by means of eq. (21) in the $`K^{}`$ propagator which is used to evaluate the $`\varphi `$ decay width in the medium, hence generalizing the correction of eq. (21) to higher orders of the density. This approximation is well justified because when we pick up the imaginary part from the two kaon cut the momenta involved are small and the p-wave plays a minor role. On the other hand, the p-wave part becomes relevant when we pick up the $`K^+Yh`$ excitation, in which case there is plenty of phase space available and the momenta involved are large. In this case, however, the $`\overline{K}`$ is far off shell and an additional particle-hole insertion in the $`\overline{K}`$ selfenergy, i.e. a higher order in the density, does not modify much the $`\overline{K}`$ propagator.
The use of the technique explained above simplifies the analytical structure of the integrand of the $`\varphi `$ selfenergy and allows one to use the same formalism that leads to eq. (12), where there are only p(and s)-wave selfenergy insertions in the propagators and no vertex corrections.
## 3 Results
In Fig. 3 we show the different contributions to the $`\varphi `$ width as a function of the $`\varphi `$ energy. The dotted line represents the contribution to the $`\varphi `$ width when the $`\overline{K}`$ is dressed with only the s-wave selfenergy (the $`K`$ is dressed with the repulsive selfenergy of eq. (11) in all these curves). We can already see an appreciable increase of the width with respect to the free value of about 4 MeV. This is due on the one hand to the fact that the $`\overline{K}`$ feels an attraction of bigger strength than the repulsion felt by the $`K`$, hence the phase space for $`K\overline{K}`$ decay increases. On the other hand, the $`\overline{K}`$ in the nucleus decays in s-wave into the channels $`\pi \mathrm{\Sigma }h`$ and $`\pi \mathrm{\Lambda }h`$ and, therefore, the incorporation of the $`\overline{K}`$ s-wave selfenergy in the evaluation of the $`\varphi `$ selfenergy automatically accounts for the $`\varphi K\pi \mathrm{\Sigma }h`$ and $`\varphi K\pi \mathrm{\Lambda }h`$ channels or, conversely, the nucleon induced $`\varphi `$ decay reactions $`\varphi NK\pi \mathrm{\Sigma }`$ and $`\varphi NK\pi \mathrm{\Lambda }`$.
A next step in the figure incorporates the contribution of decay channels related to the p-wave $`\overline{K}`$ selfenergy through $`Yh`$ excitations, with $`Y=\mathrm{\Lambda }`$ (short dashed curve), $`Y=\mathrm{\Lambda },\mathrm{\Sigma }`$ (long-dashed curve) and $`Y=\mathrm{\Lambda },\mathrm{\Sigma },\mathrm{\Sigma }^{}`$ (dot-dashed curve). These correspond to the reactions $`\varphi NK\mathrm{\Lambda },K\mathrm{\Sigma },K\mathrm{\Sigma }^{}`$. We can see that the largest contribution corresponds to the $`\mathrm{\Lambda }h`$ excitation, as already noted in Ref. , since it has a larger phase space and in addition the $`\overline{K}N\mathrm{\Lambda }`$ coupling involves the $`D+F`$ combination, while the $`\overline{K}N\mathrm{\Sigma }`$ one involves the $`DF`$ combination, which is about a factor four smaller. The contribution of the $`\mathrm{\Sigma }^{}`$ is comparatively larger than the $`\mathrm{\Sigma }`$ one, due to the larger coupling, but still smaller than that of the $`\mathrm{\Lambda }`$ due to the reduced phase space.
Finally, the solid line in Fig. 3 shows the results obtained when the vertex corrections are also incorporated. They increase the width of the $`\varphi `$ by an additional 3 MeV at the $`\varphi `$ mass, $`M_\varphi =1019.413`$ MeV.
The final width of the $`\varphi `$ meson at the $`\varphi `$ mass in symmetric nuclear matter of density $`\rho =\rho _0`$ amounts to 22 MeV. All the different decay channels contributing to the final width increase smoothly with energy as a consequence of the increasingly larger available phase space.
Although not shown in the figure, the effect of the relativistic recoil corrections (factors $`f_Y`$ in eq. (2)) amounts to a reduction of about 5 MeV, also at the $`\varphi `$ mass. If the dipole form factor is omitted the $`\varphi `$ width raises to about 30 MeV. If, on the other hand, one uses a monopole factor of the type $`\mathrm{\Lambda }^2/(\mathrm{\Lambda }^2q^2)`$, with $`\mathrm{\Lambda }=1.3`$ GeV, as commonly used in phenomenological studies of the hyperon-nucleon interaction, the $`\varphi `$ width turns out to be around 26 MeV.
In Fig. 4 we show the total width of the $`\varphi `$ as a function of the $`\varphi `$ energy for three different densities. As one can see from the figure, the medium effects are not proportional to the density. The observed density dependence also differs for the different energies of the $`\varphi `$. We have checked that this highly nonlinear density behavior of the width comes mainly from the s-wave part of the $`\overline{K}`$ selfenergy, as already noted in Ref. . At $`\rho =\rho _0`$ this s-wave contribution to the width is reduced by a factor 2.5 with respect to a linear extrapolation from low densities. The p-wave contribution is more moderately density dependent, but there is still a reduction of about 30% at $`\rho =\rho _0`$ with respect to the linear extrapolation from low densities, as also estimated in Ref. . In the present work we take into account the full density dependence of the p-wave contribution, which in Refs. was only considered at the lowest order in the density.
The analysis done here about the density dependence and the p-wave contributions can be used to support the arguments given at the end of Sect. 2 justifying the use of the prescription of eq. (21) to incorporate the p-wave vertex corrections. We have carried out a different calculation in which $`\stackrel{}{q}^2`$ is substituted in the $`\varphi `$ vertices by the term on the right in expression (21) divided by $`\stackrel{~}{\mathrm{\Pi }}(q)`$. This new prescription is equivalent to the one used before in the case of just one particle-hole excitation in the p-wave part (it should not be used for the s-wave part where there are no vertex corrections). However, this new prescription does not modify the p-wave selfenergy at higher orders in the density, a feature we would like to have in our calculation. The differences between the two prescriptions are of the order of 2% at $`\rho =\rho _0`$, which gives us an idea of the accuracy of the calculation.
The width obtained in the present work is about a factor of two smaller than that obtained in Refs. . Had we used $`f=f_\pi `$ the width would have raised only up to 25 MeV. One might think that the differences with the result in Ref. should be attributed to the fact that the latter work relied on a $`\overline{K}`$ selfenergy that was not selfconsistently evaluated . Their in-medium $`\overline{K}N`$ amplitude included basically the Pauli-blocking effects on the nucleons in the intermediate $`\overline{K}N`$ loops. As a result, the $`\overline{K}`$ strength at $`\rho _0`$ presents a quite narrow peak , located 100 MeV below the kaon mass. In contrast, when the attractive and complex $`\overline{K}`$ selfenergy is incorporated selfconsistently in the in-medium scattering equation, as done in Refs. , the peak of the $`\overline{K}`$ spectral function moves to energies closer to the free $`\overline{K}`$ mass, which should result in a drastic reduction of the $`\varphi `$ width. However, when we use the $`\overline{K}`$ selfenergy which only incorporates the Pauli-blocking medium effects, an approximation which is also discussed in Ref. , the width is only 7% larger than that obtained using the selfconsistent $`\overline{K}`$ dressing. The reason is that the extra widening of the selfconsistent $`\overline{K}`$ spectral function weakens the reduction caused by the larger mass of the $`\overline{K}`$ in the medium. Another factor which contributes to obtaining a smaller result than that found in Refs. is the inclusion in the present work of the p-wave $`\overline{K}`$ selfenergy to all orders in the $`\overline{K}`$ propagator, while only the lowest order term in the density was considered in Refs. .
The tests done in this section give us an idea of the theoretical uncertainties of the present model, which we can estimate in a band between $`2228`$ MeV for the $`\varphi `$ width at normal nuclear matter density. Yet, when these numbers are compared with the free width of 4.4 MeV, the important message, shared with Ref. , is that the width of the $`\varphi `$ is appreciably enhanced in nuclei by nearly an order of magnitude.
Finally, the real and imaginary parts of the $`\varphi `$ propagator, normalized such that the later gives the $`\varphi `$ spectral function, are shown in Fig. 5 as functions of the $`\varphi `$ energy for three different densities. The mass of the $`\varphi `$ meson in the medium has been taken to be the same as the free one, an assumption that relies upon the results of Ref. where only a small change of about 1% was found. The novel elements introduced here in the kaon spectral function should not change this result qualitatively.
The $`\varphi `$ spectral function shown in the lower pannel of Fig. 5 gives us an idea of what one might expect for, let us say, the invariant mass of $`K\overline{K}`$ distributions in nuclei around the $`\varphi `$ mass in $`\varphi `$ production experiments, like $`\gamma `$-nucleus or $`\pi `$-nucleus collisions. These reactions could be easily done in experimental facilities like Spring8/Osaka or GSI, where programs to produce $`\varphi `$ mesons in the elementary reactions are already considered. Certainly, the results obtained here, extrapolated if needed at higher densities, should also be of much use in analyses of dilepton production in heavy ion collisions around the $`\varphi `$ energy, which is actually one of the regions studied experimentally.
## 4 Conclusions
We have studied the different mechanisms of $`\varphi `$ decay in a nuclear medium. We rely upon the OZI rule which makes the direct coupling of the $`\varphi `$ to nucleons or nonstrange mesons small, and thus the $`\varphi `$ decay is still driven by the $`K\overline{K}`$ or related channels in the medium. The $`\varphi `$ width in the medium is hence obtained from previously developed models for the interaction of kaons in nuclei, which have been tested in $`K^{}`$ atoms recently. The gauge nature of the $`\varphi `$ as a vector field also introduces some vertex corrections which lead to a moderate increase of the $`\varphi `$ decay width.
We have evaluated this width as a function of the energy of the $`\varphi `$ and the nuclear density of the medium and have separated the contribution of the different reaction channels, which can eventually be investigated in some experiments for $`\varphi `$ production in nuclei. The relevance of these findings for the dressing of kaons in nuclei and its relationship to the eventual kaon condensation in dense matter, as well as the implications for the running experiments of dilepton production in heavy ion collisions, should stimulate the performance of $`\varphi `$ production in nuclei with some particle nucleus collisions which could serve us to test the ideas and results obtained in the present work.
## 5 Acknowledgements
We would like to thank B. Friman for discussions . This work has been partially supported by the DGICYT contract numbers PB98-1247 and PB96-0753, and by the EEC-TMR Program, EURODAPHNE, under contract number CT98-0169.
|
warning/0005/nucl-th0005030.html
|
ar5iv
|
text
|
# Nuclear isotope thermometry
## I Introduction <br>
Due to the short range attraction between nucleons, nuclear matter is a Fermi liquid at low temperature and is expected to undergo a phase transition to a nucleonic gas within a mixed phase region bounded by a critical temperature of order 15 MeV . Experimental investigations of this phase transition have focused on a variety of experimental observables ranging from the mass, charge or multiplicity distributions for the emitted fragments to observables sensitive to the temperature of the system .
Temperature measurements, in particular, have been performed to search for evidence of the enhanced heat capacity predicted by statistical model calculations reflecting the latent heat for transforming the Fermi liquid to the nucleonic vapor . For example, the Statistical Multifragmentation Model (SMM) predicts a plateau of roughly constant temperature of $`T5MeV`$ for excitation energies of $`E^{}/A37MeV.`$ At these excitation energies the model predicts a mixed phase consisting of fragments (liquid) and nucleons and light particles (gas) corresponding to a mixed phase equilibrium. This is followed at higher excitation energies by a linear rise in the temperature with excitation energy as expected for a gas of small nuclei having negligible internal heat capacity . Similar effects are predicted by the Microcanonical Metropolis Monte Carlo (MMMC) model .
This trend was qualitatively reproduced in some experiments , but not in others . An essential part of these measurements is the determination of the temperature of the fragmenting system. Temperatures were extracted from the isotopic abundances of helium and lithium fragments, using the isotope thermometry method proposed by Albergo et al. . The idea of the method is to determine the double ratios of the yields of four suitably chosen isotopes, ($`A_1,Z_1`$), ($`A_1+1,Z_1`$), ($`A_2,Z_2`$), ($`A_2+1,Z_2`$),
$$\frac{Y(A_1,Z_1)/Y(A_1+1,Z_1)}{Y(A_2,Z_2)/Y(A_2+1.Z_2)}=C\mathrm{exp}(\mathrm{\Delta }B/T_{iso})$$
(1)
where the $`Y`$ are the yields of the different isotopes, $`C`$ is a constant related to spin values and kinematic factors, $`\mathrm{\Delta }B=B(A_1,Z_1)B(A_1+1,Z_1)B(A_2,Z_2)+B(A_2+1,Z_2)`$ is obtained from the binding energies of the isotopes appearing in Eq. (1), and $`T_{iso}`$ stands for the temperature deduced from this isotopic thermometer. In the case of the He-Li thermometer employed in ref. , $`A_1=6,Z_1=3,A_2=3,Z_2=2`$. For the C-Li thermometer, more recently considered by Xi et al. , $`A_1=6,Z_1=3,A_2=11,Z_2=6`$. For the Carbon thermometer studied in this work, $`A_1=12,Z_1=6,A_2=11,Z_2=6.`$
However, there are a few aspects which should be carefully analyzed when one wants to compare information on the breakup configuration of an excited system formed in a heavy-ion collision to multifragmentation models like the SMM approach. Some of these points are addressed below. In sect. II we briefly discuss the assumptions underlying this method. Variations in the temperature of the breakup stage, where the hot primary fragments decouple from the system, are intrinsic to finite systems and are explored within the SMM approach in sect. III. An analytic description of temperature variations is developed in the grand canonical limit in sect. IV; this description is consistent with the results from the SMM. In addition, there are finite size effects, discussed in sect. V, that make the concept of an overall chemical potential somewhat inaccurate. The influence of secondary decay is discussed in sect. VI. Conclusions are drawn in sect. VII.
## II Underlying assumptions <br>
The basic physical hypotheses of the isotope thermometry method are:
1. an equilibrated source is formed after the most violent stages of the reaction and it decays simultaneously and statistically,
2. for the experimental event selection employed in the analyses, all the events correspond to fragments formed at the same temperature, and
3. distortions on the isotopic temperature due to secondary decay of hot primary fragments may be neglected.
Although the Statistical Multifragmentation Model , used in the discussion below, is based on the first assumption, the last two hypotheses are not supported by the model, as we shall discuss in detail.
The SMM uses the Monte Carlo method and averages observables with the statistical weight over decay partitions. A multifragment decay partition is defined in the SMM approach as a specific set of emitted fragments and light particles. For simplicity, each partition in the SMM approach is weighted according to the entropy of the partition. This entropy is approximated by analytical expressions rather than by an event by event sampling of the phase space as in ref. These approximations rely upon that fact that the dominant contribution to this entropy comes from internal phase space of fragments which plays the role of a heat bath within the SMM approach just as an excited residue plays the role of a heat bath within compound nuclear decay theory .
For a given decay partition and by making a Wigner Seitz approximation to the Coulomb energy, energy conservation within the SMM approach leads to the expression ,
$$E_0^{gs}+E_0^{}=\frac{3}{5}\frac{Z_0^2e^2}{R_0}+\underset{\{A,Z\}}{}N_{AZ}E_{AZ}$$
(2)
where $`E_0^{}`$ is the total excitation energy and $`E_0^{gs}`$ is ground state energy of a nuclei having a mass and atomic number equal to that of the total system, $`A_0`$ and $`Z_0`$, respectively. The first term on the right hand side stands for the Coulomb energy of a homogeneous charge $`Z_0e`$ occupying the volume of the system of radius $`R_0`$ and $`N_{AZ}`$ indicates the number of fragments of mass number $`A`$ and atomic number $`Z`$ in the partition of the system.
In the equation above, $`E_{AZ}`$ is the kinetic plus internal energy for each of these fragments. It is related to the temperature by assuming all fragments are at a common temperature as follows,
$$E_{AZ}=\frac{3}{2}T+E_{AZ}^{}(T)+E_{AZ}^CB_{AZ}$$
(3)
where the internal excitation energy of the fragments, $`E_{AZ}^{}(T)`$, may be approximated by an extension of the semi-empirical mass formula to finite temperatures , and the extra Coulomb energy of the fragment in the fragmentation volume, $`E_{AZ}^C`$, may be calculated within the Wigner-Seitz approximation. $`B_{AZ}`$ stands for the ground state binding energy for the fragment. Eqs. 2 and 3 result from an average of the microcanonical expression for energy conservation over the phase space corresponding to the specific decay partition.
By applying the energy conservation relationship in Eqs. (2-3) one obtains a temperature $`T`$ that describes the internal excitation and translational energies of fragments within a given partition. Even though the overall system is assumed to be in equilibrium at a fixed excitation energy $`E_0^{},`$ different decay partitions have different Coulomb, binding, and translational energies and, consequently, different excitation energies of the emitted fragments. Consistency with Eqs. (2-3) therefore requires that the temperature $`T`$ of the fragments varies from one decay partition to another, reflecting the differences between the Coulomb, binding and translational energies of the various partitions.
Labeling the partition $`\{N_{AZ}\}`$ with the index, $`f,`$ the statistical weight associated with the partition,
$$W_f=\mathrm{exp}\left[\underset{\{A,Z\}}{}N_{AZ}S_{AZ}(T)\right]$$
(4)
may be found by expressing the entropy of the fragments, $`S_{AZ}`$, using approximations derived from the liquid drop model at finite temperature . Consequently the physical observables can be expressed by a weighted average over decay partitions as,
$$O_{AZ}=\frac{_fW_fO_{AZ}}{_fW_f}$$
(5)
where $`O_{AZ}`$ can be any interesting observables such as the yield of a fragment or the temperature (In the present work, the summation included $`10^8`$ partitions.)
This allows one to predict the various results from the SMM that are addressed in the next section with regard to the temperature variations.Because the SMM approach invokes a temperature to sample the microcanonical phase space, we denote the predicted observables as approximately microcanonical. Despite this caveat, we note that this procedure can, in principal, give accurate microcanonical predictions for experimental observables provided the thermal expressions for the free energies are accurate descriptions of the integration over the microcanonical phase space.
Before passing on to the various results of our investigation, it is important to clarify that we do not invoke the Grand Canonical approximations to the SMM approach introduced in ref. to allow Monte Carlo event simulations. Instead, we have adhered closely to original SMM approach outlined in ref. , with the exception that all calculations in this paper were performed at a constant freezeout density equal to one third that of the saturation density of nuclear matter.
## III Primary Temperatures
The SMM procedure expressed in Eqs.(2-5) leads to a distribution of the temperatures of the fragmenting system for a given excitation energy in the same sense that the temperature of the daughter nucleus in compound nuclear decay theory varies as a function of the Coulomb barrier and separation energy of each decay channel. The points in Fig. (1) denote the temperature distributions for the fragmentation of an excited $`{}_{}{}^{112}Sn`$ nucleus at three different excitation energies obtained with the SMM. These distributions are well fitted by gaussian functions, shown by the curves in the figure, with variances $`\sigma _T^2`$ that are fairly independent of the energy, $`\sigma _T0.4`$ MeV, in the range $`3\mathrm{MeV}E_0^{}/A10\mathrm{MeV}`$. At each excitation energy, we average over all of the partitions and define this average value as the ” approximate microcanonical” temperature T$`_{MIC}.`$
Since each of the isotopes employed in the thermometer has a specific mass, charge and binding energy, the application of conservation laws sets a constraint on the values available to the remainder of the system. Because of this finite size effect, the temperature distribution obtained when a specific isotope is present is slightly different from the one obtained when all partitions are considered. In particular, a small difference ($`0.1`$ MeV) is observed between the average temperatures for the various isotopes; this is illustrated in Fig. (2) for carbon isotopes from the fragmentation of a $`{}_{}{}^{112}Sn`$ nucleus at $`E_0^{}/A=6`$ MeV. Even though the average temperatures are different reflecting the different binding energies of the three isotopes, all these distributions are gaussians with nearly the same variances. We can extract another temperature $`T_{IMF\text{ }}`$ by averaging over partitions which contain an Intermediate Mass Fragment (IMF) with $`3Z10`$. It’s interesting to note that $`T_{MIC}`$ can exceed $`T_{IMF}`$ at low energies by as much as 0.2 MeV, in part because it takes more energy to emit an IMF than to emit an equivalent mass in the form of alpha particles, leaving less energy for thermal excitation.
The basic idea contained in Eq. (1) was derived under the assumption that the primary yields are well represented by the grand canonical approximation at a single breakup temperature; the double ratio was invoked to cancel out the contribution to the yields coming from the neutron and proton chemical potentials. In the SMM, however, the temperature varies from partition to partition and the chemical potentials, which appear within the grand canonical formalism as Lagrange multipliers that conserve charge and mass, are not explicitly invoked. Thus, we can not presume the validity of the Albergo’s formula ( Eq. 1) in the SMM and must test its validity instead.
We begin with a test of the validity of Eq. (1) when one employs the primary yields. For a given decay partition $`\{N_{AZ}\},`$ we take into account the internal free energy $`F_{AZ}^{\text{int}}(T)`$ which is parameterized as:
$`F_{AZ}^{\text{int}}`$ $`=`$ $`B(A,Z)+F_{AZ}^{}(T)+F_{AZ}^C,`$ (6)
$`F_{AZ}^{}(T)`$ $`=`$ $`F_{AZ}^B(T)+F_{AZ}^S(T)T\mathrm{ln}(g_{AZ}^{gs})`$ (7)
where $`g_{AZ}^{gs}`$ is the ground state spin degeneracy, and $`F_{AZ}^B`$, $`F_{AZ}^S`$, and $`F_{AZ}^C`$ correspond to the excitation energy dependent bulk, surface, and Coulomb contributions to the internal free energy after the binding energy part has been removed. The reader is referred to ref. for explicit expressions for the terms entering in the equation above. Then the primary yield for the ground state can be related to the total yield by
$$N_{AZ}^{gs}=N_{AZ}g_{AZ}^{gs}\mathrm{exp}\left[F_{AZ}^{}(T)/T\right]$$
(8)
for this partition. Following the procedure described in the previous section, we will use this expression and Eq. (5) to obtain the average g.s. yield distribution $`N_{AZ}^{gs}`$. This, in turn, can be used in Eq. (1) to extract isotopic temperatures as follows,
$$\frac{N_{A1,Z1}^{gs}/N_{A1+1,Z1}^{gs}}{N_{A2,Z2}^{gs}/N_{A2+1,Z2}^{gs}}=C\mathrm{exp}\left(\frac{\mathrm{\Delta }B}{T_{iso}^{smm}}\right).$$
(9)
In previous SMM calculations, experimental binding energies and spin degeneracy factors $`g_{AZ}^{gs}`$ were used for light nuclei with $`A<5`$. Liquid drop binding energies and spin degeneracy factors of unity were used for $`A5`$. In this work, we will retain these conventions on spin degeneracy factors so as to be consistent with prior calculations, but we will use empirical binding energies for all nuclei.
In Fig. (3), the isotopic temperatures $`T_{iso}^{smm}`$ for the carbon thermometer ($`Z_1=Z_2=6,A_1=11,A_2=12`$) are plotted as the stars for the multifragmentation of a $`{}_{}{}^{112}Sn`$ source at excitation energies $`E_0^{}/A=310`$ MeV. For comparisons, the corresponding $`T_{MIC}`$ and $`T_{IMF}`$ for the same system are also shown in Fig. (3) as the dashed and solid lines, respectively. While supporting the concept of isotopic thermometry, the good agreement between $`T_{IMF}`$ and $`T_{iso}^{smm}`$ is somewhat surprising, given the strong dependence of the Boltzmann factor on temperature for large $`\mathrm{\Delta }B`$ and the width of the temperature distribution shown in Fig. (1). As shown in the following section, it occurs in part due to a large cancellation involving the Boltzmann factor and the temperature dependences of the effective chemical potentials. Fig. (3) also reveals that fairly precise information about $`T_{IMF}`$ and somewhat less precise information about $`T_{MIC}`$ is provided by the primary yields. This suggests that given a precise relationship between primary to the final yields, it would be possible to determine the breakup temperature from the measured yields.
## IV Effects of Temperature Variations
The surprising consistency between $`T_{IMF}`$ and $`T_{iso}^{smm}`$ in Fig. (3) suggests that the corrections to the grand canonical prediction for the isotope temperatures are small, and one may utilize this approach to understand why the temperature variations have so little influence on the results. Taking this tact, we assume that the isotopic distributions are well approximated for each partition by the grand canonical limit, use this limit to gain insight into the finite size effects and at the same time, investigate the accuracy of this approximation. We take this approach to consider first the influence of the temperature variations and later the consequences of the finite size on the effective chemical potentials.
Considering the influence of the temperature variations in this approximation, we average the grand canonical approximation over the temperature distribution in Fig. (1). If the approximation works, the expressions that result from this average should be appropriate for the consideration of the effects of temperature distributions arising from other effects and within other equilibrium models of multifragmentation as well. Taking this approach, the yield of a particular isotope $`i`$ in the framework of Albergo’s method , when averaged over all possible partitions, becomes:
$`Y_i=`$ $`V{\displaystyle _0^{\mathrm{}}}𝑑Tf(T){\displaystyle \frac{A_i^{3/2}\zeta _i(T)}{\lambda _T^3}}`$ (11)
$`\mathrm{exp}\left[(Z_i\mu _{PF}\left(T\right)+N_i\mu _{NF}\left(T\right)+B_i)/T\right]`$
where $`f(T)`$ is the temperature distribution, $`V`$ represents the free volume of the system, $`\lambda _T=\sqrt{2\pi \mathrm{}^2/mT}`$, $`m`$ is the nucleon mass and $`\mu _{PF}`$ ($`\mu _{NF}`$) stands for the chemical potential associated with free protons (neutrons) at temperature $`T`$. The internal partition function of the fragment $`i`$ is given by:
$$\zeta _i(T)=\underset{j}{}g_i^j\mathrm{exp}\left[\frac{\mathrm{\Delta }E_j}{T}\right]$$
(12)
where $`\mathrm{\Delta }E_j`$ is the excitation energy of the state $`j`$ with respect to the ground state and $`g_i^j`$ stands for the spin degeneracy factor of this excited state.
Assuming that $`f(T)`$ is a gaussian centered at $`T`$ and with width $`\sigma _TT`$ (see Fig. 1 ), one may expand $`1/T`$, $`T^{3/2}`$, and the chemical potentials. By considering only fragments observed in the ground state, i.e. $`\zeta _i(T)=g_i^0`$, we obtain that
$`Y_i^{gs}`$ $`=`$ $`{\displaystyle \frac{g_i^0VA_i^{3/2}T^{3/2}}{\lambda _{}^3}}`$ (15)
$`\mathrm{exp}\left[{\displaystyle \frac{B_i}{T}}+{\displaystyle \frac{\mu _{PF}\left(T\right)Z_i+\mu _{NF}\left(T\right)N_i}{T}}\right]`$
$`{\displaystyle \frac{1}{\sqrt{2p}}}\mathrm{exp}\left[{\displaystyle \frac{q^2}{4p}}\right].`$
where $`\lambda _{}\sqrt{2\pi \mathrm{}^2/m}`$. In the above expression, the corrections to the grand canonical relationship are provided by the correction factor $`\frac{1}{\sqrt{2p}}\mathrm{exp}\left[\frac{q^2}{4p}\right]`$ which depends on assumed width of the temperature distribution, the binding energy of the i-th fragment, the neutron and proton chemical potentials and their derivatives through the parameters $`p`$ and $`q`$. These two parameters are defined by
$`p`$ $`=`$ $`{\displaystyle \frac{1}{2}}+\left[{\displaystyle \frac{\sigma _T}{T}}\right]^2\left[Z_i\alpha _{PF}+N_i\alpha _{NF}+{\displaystyle \frac{3}{4}}{\displaystyle \frac{B_i}{T}}\right]`$ (16)
$`q`$ $`=`$ $`{\displaystyle \frac{\sigma _T}{T}}\left(Z_i\beta _{PF}+N_i\beta _{NF}+{\displaystyle \frac{3}{2}}{\displaystyle \frac{B_i}{T}}\right),`$ (17)
where
$`\alpha _{PF}`$ $`=`$ $`\mu _{PF}^{^{}}\left(T\right){\displaystyle \frac{\mu _{PF}\left(T\right)}{T}}{\displaystyle \frac{1}{2}}\mu _{PF}^{^{\prime \prime }}\left(T\right)T`$ (18)
$`\beta _{PF}`$ $`=`$ $`\mu _{PF}^{^{}}\left(T\right){\displaystyle \frac{\mu _{PF}\left(T\right)}{T}}`$ (19)
$`\alpha _{NF}`$ $`=`$ $`\mu _{NF}^{^{}}\left(T\right){\displaystyle \frac{\mu _{NF}\left(T\right)}{T}}{\displaystyle \frac{1}{2}}\mu _{NF}^{^{\prime \prime }}\left(T\right)T`$ (20)
$`\beta _{NF}`$ $`=`$ $`\mu _{NF}^{^{}}\left(T\right){\displaystyle \frac{\mu _{NF}\left(T\right)}{T}}.`$ (21)
The isotopic temperature can be extracted from the above corrected yields. Replacing $`Y(A,Z)`$ in Eq. (1) by the right hand side of Eq. (15), one cancels out the spin and mass dependent term C and then obtains:
$$\mathrm{exp}\left[\mathrm{\Delta }B/T_{iso}^{cal}\right]=\frac{G(A_1,Z_1)/G(A_1+1,Z_1)}{G(A_2,Z_2)/G(A_2+1,Z_2)},$$
(22)
where
$`G(A,Z)`$ $`=`$ $`\mathrm{exp}\left[{\displaystyle \frac{B_i}{T}}+{\displaystyle \frac{\mu _{PF}\left(T\right)Z+\mu _{NF}\left(T\right)N}{T}}\right]`$ (24)
$`{\displaystyle \frac{1}{\sqrt{2p}}}\mathrm{exp}\left[{\displaystyle \frac{q^2}{4p}}\right].`$
In the above double ratio the terms involving the chemical potentials evaluated at the average temperature cancel; however, terms in the correction factor involving the derivatives of the chemical potentials remain.
Quantitative estimates of the correction factor require one to obtain estimates for the effective chemical potentials and their derivatives with respect to temperature. The proton and neutron chemical potentials at temperature $`T`$ may be calculated from the free proton and neutron multiplicities via the expression:
$`\mu _{PF}(T)`$ $`=`$ $`T\mathrm{log}\left[{\displaystyle \frac{\lambda _T^3Y_{PF}(T)}{g_{PF}V}}\right],`$ (25)
$`\mu _{NF}(T)`$ $`=`$ $`T\mathrm{log}\left[{\displaystyle \frac{\lambda _T^3Y_{NF}(T)}{g_{NF}V}}\right]`$ (26)
where $`g_{PF}(g_{NF})`$ represents the spin degeneracy factor of the proton(neutron). For the calculations presented in this work, it has proven advantageous and reasonably accurate to approximate the yields $`Y_{PF}(T)`$ and $`Y_{NF}(T)`$ over a modest range in temperature by power law expressions in the temperature. In this approximation,
$`Y_{PF}(T)`$ $`=`$ $`c_{PF}T^{\gamma _{PF}},`$ (27)
$`Y_{NF}(T)`$ $`=`$ $`c_{NF}T^{\gamma _{NF}}`$ (28)
For the decay of $`{}_{}{}^{112}Sn`$ nuclei at temperatures ranging over $`4T7MeV`$ , $`Y_{PF}`$ and $`Y_{NF}`$ are well described by $`\gamma _{PF}=4.5`$ and $`\gamma _{NF}=1.0`$ $`(c_{PF}=1.33\times 10^4`$ and $`c_{NF}=0.267)`$ according to the SMM; comparisons of this parameterization to yields calculated with the SMM model are shown Fig. (4). These values depend on the density, which has been chosen to be one third that of the saturation density of nuclear matter. Larger values of the free nucleon yields are obtained at lower density.
Using this approximation, the explicit forms of the correction factors in Eqs. (15)-(18) become 2$`\alpha _{PF}=\beta _{PF}=\left(\gamma _{PF}\frac{3}{2}\right)=3`$ and 2$`\alpha _{NF}=\beta _{NF}=\left(\gamma _{NF}\frac{3}{2}\right)=\frac{1}{2}`$. We note that the correction factor to the temperature $`T_{iso}^{cal}`$ in Eq. (22) depends on the power law exponents $`\gamma _{PF}(\gamma _{NF})`$ and not on the absolute values of the proton(neutron) yields.
Even though Eq. (11) has an exponent that appears to be strongly temperature dependent, there is a strong cancelation between the contributions from the chemical potentials and binding energy factors in the expressions for $`p`$ and $`q.`$ As a result, the correction factor is of order unity. Values in the range of $`\frac{1}{\sqrt{2p}}\mathrm{exp}\left[\frac{q^2}{4p}\right]12`$ are obtained, for example, in the decay of $`{}_{}{}^{112}Sn`$ nuclei at temperatures in the range of $`4T7MeV`$ .
The isotopic temperatures $`T_{iso}^{cal}`$ calculated from Eq. (22) for carbon thermometer are shown in Fig. (3) in comparisons with temperatures $`T_{MIC}`$, $`T_{IMF}`$ and $`T_{iso}^{smm}`$ derived from the SMM in the previous session. The very good agreement between $`T_{iso}^{cal}`$ , $`T_{iso}^{smm}`$ and $`T_{IMF}`$ indicates that the corrections to the isotopic temperatures associated with these temperature variations are small, although the yields can change by as much as a factor of two. This comparative insensitivity arises because the isotopic thermometers depend logarithmically on the yields.
This insensitivity depends on the nature and magnitude of the temperature variation. The corrections to the isotopic temperatures will be somewhat larger in other contexts or other models where the temperature variations are larger. The limited precision with which systems may be selected experimentally may also have a similar influence because the excitation energy and temperature varies experimentally from collision to collision due to variations in the impact parameter or in the energy removed by preequilibrium particle emission. The influence of this temperature variation, which may exceed the variation in temperature caused by the averaging over decay partitions, can also be estimated via techniques outlined in the present section. To illustrate how one can estimate the possible corrections due to an imprecision in the excitation energy definition, the circles in Fig. (3) show calculations using Eq. (22) for carbon thermometer assuming a width of $`\sigma _T0.8`$ MeV for the temperature distribution, which is twice as large as that predicted in Figs. 1 and 2. This width is not based upon a dynamical calculation; it is only to illustrate that larger isotopic temperatures can result if the excitation energy is poorly defined.
## V Chemical Potentials
The grand canonical limit has a great advantage of providing a simple analytic expression for the isotopic yields from which other useful expressions can be derived. However, the concept of uniform chemical potentials is not a prediction of microcanonical or canonical models and must be investigated to determine its applicability to finite systems. We do this by trying to comparing the grand canonical expression for the isotopic yields to the predictions of approximately microcanonical SMM calculations. We start by assuming that these isotopic distributions can be calculated within the grand canonical approximation and then test this assumption as follows.Using a pair of adjacent isotopes, we invert the grand canonical expression for the isotopic yields of two adjacent isotopes to obtain an equation for the effective neutron chemical potential:
$`\mu _n^{eff}(A,Z)`$ $`=`$ $`T\mathrm{log}[{\displaystyle \frac{g_{AZ}^{gs}}{g_{A+1Z}^{gs}}}\left({\displaystyle \frac{A}{A+1}}\right)^{3/2}`$ (30)
$`\mathrm{exp}((B_{AZ}B_{A+1Z})/T){\displaystyle \frac{Y_{A+1Z}^{gs}}{Y_{AZ}^{gs}}}]`$
where g$`{}_{}{}^{gs}{}_{AZ}{}^{}`$, $`B_{AZ}`$ and $`Y_{AZ}^{gs}`$ are the ground state spin degeneracy, the binding energy and the ground state primary yield for a fragment with (A,Z), respectively. If the $`Y_{AZ}^{gs}`$ taken to be the ground state yields predicted by the SMM, $`\mu _n^{eff}(A,Z)`$ becomes an effective ”SMM” chemical potential. By performing SMM calculations, we find the temperature- and isotopic- dependences of the effective neutron chemical potentials given in Fig. (5) for Carbon and Lithium isotopes from the decay of a <sup>112</sup>Sn nucleus at excitation energies of $`E_0^{}/A=3,6,9MeV`$.
These effective chemical potentials are essentially the same for the Carbon and Lithium isotope chains. This insensitivity to element number is consistent with the concept of a chemical potential and offers support for the use of the grand canonical expression to describe isotopic distributions. There is a dependence on the neutron number of the isotope, however, that lies outside of the grand canonical approximation. This variation in the neutron chemical potential basically comes as a result of mass, charge and energy conservation for a finite-size system. We can understand the influence of these conservation laws most easily at low excitation energies, where the two largest fragments in the final state are the IMF (Carbon or Lithium in this case) and a heavy residue which contains most of the remaining charge and mass. We estimate the influence of conservation laws at low excitation energy qualitatively by considering binary decay configurations. Assuming that a parent nucleus ($`A_0,Z_0`$) decays into a light fragment (A,Z) and a heavy residue $`(A_0A,Z_0Z)`$ , we can approximate the yield of fragment (A,Z) in its ground state by
$`Y_{AZ}^{gs}`$ $``$ $`\rho ^{gs}(A,Z)\rho ^{}(A_0A,Z_0Z)\overline{\rho }_{REL}`$ (31)
$``$ $`g_{AZ}^{gs}\mathrm{exp}\left[S^{}(A_0A,Z_0Z)\right]`$ (33)
$`\left[{\displaystyle \frac{A\left(A_0A\right)}{A_0}}\right]^{3/2}{\displaystyle \frac{1}{\lambda _T^3}}`$
where $`\rho ^{gs}=g_{AZ}^{gs},`$ $`\rho ^{}`$ and $`S^{}`$ are the density of states for the light nucleus in its ground state level, the density of states and entropy of the heavy residue in its excited state, respectively, The other factor, $`\overline{\rho }_{REL}\left[\frac{A\left(A_0A\right)}{A_0}\right]^{3/2}\lambda _T^3,`$ is the thermal average of the state density of relative motion.
Replacing the yields in Eq.(30) with Eq.(31) and assuming $`A<<A_0,`$ one finds that the effective chemical potentiail depends on the difference in residue entropies, $`S^{}(A_0A1,\overline{Z})S^{}(A_0A,Z_0Z)`$. Using an expansion for small changes in the nuclear entropy from ref. , this difference can be expressed in terms of the difference of binding energies,
$`S^{}(A_0A1,\overline{Z})S^{}(A_0A,Z_0Z)`$ (34)
$`=`$ $`(B_{A_0A,Z_0Z}B_{A_0A1,Z_0Z})/T`$ (36)
$`(B_{AZ}B_{A+1Z})/T+f^{}/T`$
plus a term depending on the free excitation energy per nucleon, $`f^{}=E^{}/A_0TS/A_0.`$ This difference in binding energies is further related to the neutron separation energy $`s_n(A_0A,Z_0Z)`$:
$$s_n(A_0A,Z_0Z)=B_{A_0A,Z_0Z}B_{A_0A1,Z_0Z}$$
(37)
One consequently obtains the following expression for the effective chemical potential:
$$\mu _n=s_n(A_0A,Z_0Z)+f^{}.$$
(38)
where the reduced free excitation energy has been approximated by its low energy limit,
$$f^{}=\frac{T^2}{\epsilon _0},\text{ }\epsilon _0=8MeV.$$
(39)
For the decay <sup>112</sup>Sn$`^{12}`$C$`+`$X , the chemical potential at $`T=0`$, i.e., $`s_n(A_0A,Z_0Z)`$, is plotted as the stars in Fig.(5); the binding energies for these calculations were calculated using the liquid-drop parametrization in ref.. The reduced free energy $`f^{}`$ gives a reasonable estimate for the trend with excitation energy. The dot-dashed line in Fig.(5) gives the chemical potential predicted from Eq. (38) for $`E_0^{}/A=3MeV`$ ($`T=4.58MeV`$). The predicted trend is close to that predicted by the SMM model (solid circles and squares) but has a somewhat stronger dependence on $`NZ.`$
In general, the slope of the effective neutron chemical potential is getting slightly flatter as the excitation energy or temperature increases. If we consider that the system undergoes a multiple fragment decay at higher temperatures, it is clear that approximating the entropy of the remaining system by that of a residue of comparable mass becomes rather inaccurate. The constraints imposed on the total system by the isospin asymmetry of one observed fragment should, in that case, be less significant. While there is a mass dependence to the effective chemical potential that is inconsistent with the grand canonical approach, it is useful to note that the mass dependence of the chemical potential (for these systems of more than 100 nucleons) is small if one is mainly concerned with nuclei near the valley of stability. If one cancels the chemical potential effects by constructing double ratios like that of the Algergo formula, the consequence of such finite size effects becomes negligible indeed.
## VI Influence of Secondary Decay <br>
As discussed in sect. II, fragments are formed in excited states as well as in their ground states, corresponding to the breakup temperature. Fragments in short lived excited states decay before they are detected and, therefore, the observed yields differ from that of the primary fragments. The effects of secondary decay on the isotopic yields and isotopic temperatures have already been reported by some authors (see for example ). Although the approaches employed in the description of the decay of hot primary fragments are different, all those works qualitatively agree on the point that the isotopic temperature is lower than the thermodynamical one.
At the quantitative level, details of the population and decay of the excited fragments are important. One issue concerns the importance of utilizing empirical binding energies, energy levels and decay branching ratios for the excited fragments. Fig. (6) shows the primary and secondary carbon isotopic distributions for the decay of a $`{}_{}{}^{112}Sn`$ nucleus at initial excitation energies of $`E_0^{}/A=4`$ and $`6MeV`$. The primary distribution (solid line) is calculated by considering empirical binding energies within the SMM for hot fragments. The simplified Weisskopf evaporative decay procedure of ref. is used for one final distribution (dotted line). The other final distribution (dashed line) is obtained by calculating the secondary decay for $`Z10`$ hot fragments, as in ref. , according to empirical nuclear structure information regarding the excitation energies, spins, isospins and decay branching ratios where available. For hot fragments with $`Z10`$ where such information is not available, the decay is calculated according to the Hauser-Feshbach formalism. The contributions to this latter calculation from the secondary decay of hot fragments with $`Z>10`$, are calculated, for simplicity, via the secondary evaporative decay procedure of ref. . Decays of fragments with $`Z>10`$ make a 15% contribution to the yields of $`{}_{}{}^{12}C`$ isotopes that may be altered when the decay of hot fragments with $`Z>10`$ is calculated more accurately.
Obviously, in Fig. (6), the final distribution after the empirical secondary decay is much wider than the final distribution obtained via the evaporative decay approach of ref. . This points out the importance of using the empirical information in such calculations. This also leads to the extraction of larger isotopic temperatures via Eq. (1) for the empirical approach. Temperatures for the Carbon isotope thermometer and He-Li thermometer calculated for the two secondary decay approaches are shown, for example, in Fig. (7) for the multifragmentation of a $`{}_{}{}^{112}Sn`$ nucleus at $`E_0^{}/A=410`$ MeV. For reference, the curves $`T_{MIC}`$ and $`T_{IMF}`$ from Fig. (3) are also shown as the dashed and solid lines in the figure. Clearly, incorporating empirical information in the decay makes a significant difference. Both calculations provide lower isotopic temperatures than have been obtained in recent experiments .
It should be noted, however, that the simplified Weisskopf evaporative decay, shown in Figs. (6) and (7), is only used in the SMM code of ref. to calculate the decay of fragments with $`A>16.`$ The decay of lighter fragments is calculated via a ”Fermi Breakup” multiparticle decay formalism. This latter decay mechanism makes the dominant contribution to the isotope temperatures calculated via the latter SMM code in ref. . Investigations of the experimental and theoretical basis for the ”Fermi Breakup” approach are needed, but are out of the scope of the present work.
Regardless of the decay formalism, memory of the breakup stage is lost via the secondary decay mechanism. The degree of memory loss depends on the details of the secondary decay correction and on the role of short-lived higher lying particle unbound states. A smaller degree of memory loss ensues in models such as those of refs. , where few, if any, particle unbound states are considered. The approach of ref. represents the other extreme, wherein all states are considered regardless of lifetime. This issue clearly needs further study to see whether the role of particle unstable nuclei can be constrained, for example, by direct measurements using techniques discussed in refs. or by other experimental observables.
## VII Concluding Remarks <br>
We discussed some of main aspects that could cause microcanonical predictions for isotopic distributions and isotopic temperatures to differ from grand canonical calculations and influence the determination of the breakup temperature and other experimental observables. We investigate this problem by checking the consistency of the grand canonical expression for the isotopic yields against the approximately microcanonical SMM predictions and explore the potential role which may be played by variations in the temperature and in the effective chemical potentials. These variations occur as a consequence of the finite size of the disintegrating system and are therefore present in all microcanonical calculations.
Concerning the temperature variation, we find that this causes the isotopic yields obtained with the approximately microcanonical SMM simulations for the primary distribution to differ from those of the grand-canonical ensemble by factors of order unity. One difference stems from the averaging over the temperatures corresponding to the different breakup partitions. These vary because the total binding, coulomb and translational kinetic energies vary from partition to partition and by subtraction, the thermal energy must vary as well. A simple and relatively accurate prescription that accounts for these temperature variations was given that may also prove useful for estimating the influence of thermal averaging over the variations in the actual excitation energy deposition within a data set that is constrained by an experimental cut on the estimated energy deposition.
We also extract effective chemical potentials by comparing approximate microcanonical and grand canonical expressions for the isotopic yields. These effective chemical potentials are approximately the same for isotopes of different elements that lie along the valley of beta stability, but vary as a function of (N-Z). For example, we observe for the neutron chemical potential a dependence upon (N-Z) that can be understood at low excitation energies to arise from the dependence of the neutron separation energy on the location of the accompanying residue relative to the line of beta stability.
Typically, these variations in temperature and effective chemical potential cause variations in the isotopic yields of order unity. The logarithmic relation between the isotopic temperature and the yields means that the latter may be wrongly predicted by a factor of two and one may still find a reasonable agreement between the approximate microcanonical and the isotopic temperatures provided the binding energy difference $`\mathrm{\Delta }B`$ is significantly larger than the temperature. When the effects of secondary decay is taken into account, however, the yields can change by more than an order of magnitude and the temperature values can decrease appreciably. While the magnitude of this change is not yet unambiguously established, it was shown that the incorporation of empirical information about the decay is essential for quantitative comparisons to experimental data. Measurements that quantify the role of higher lying particle unstable states are essential for determining the magnitude of these secondary decay corrections.
## ACKNOWLEDGMENTS
We would like to acknowledge important discussions with Alexander Botvina and HongFei Xi during the early stages of this work. R.D. and S.R.S. thank the NSCL at MSU for support during visits on which part of this work was performed, and also the MCT/FINEP/CNPq (PRONEX) program, under contract #41.96.0886.00, CNPq, FAPERJ, and FUJB for partial financial support. This work was supported in part by the National Science Foundation under Grant No. PHY-95-28844.
|
warning/0005/astro-ph0005294.html
|
ar5iv
|
text
|
# Exploring the evolution of spiral galaxies
## 1 Introduction
The chemical evolutionary histories of spiral galaxies provide considerable insight into many of the important processes involved in galaxy formation and evolution. For example, we can study star formation laws (SFLs; e.g. Wyse & Silk 1989; Phillipps & Edmunds 1991), the interactions between newly-formed stars and the interstellar medium (e.g. Dekel & Silk 1986; MacLow & Ferrera 1999), the importance and effects of gas flows (e.g. Lacey & Fall 1985; Edmunds 1990; Edmunds & Greenhow 1995) and the infall that must accompany disc formation (e.g. Tinsley & Larson 1978; Lacey & Fall 1983; Steinmetz & Müller 1994). The main challenge is obtaining unambiguous insight into particular physical processes. Some of the ambiguity can be circumvented by studying both the ages and the metallicities of galaxies: in this paper we use the ages and metallicities of a sample of face-on spiral galaxies to constrain which processes are the most important in affecting their observational properties.
Despite these difficulties, considerable progress has been made in understanding some important aspects of galaxy formation and evolution. A local density dependence in the SFL is strongly favoured, although other factors may affect the star formation rate (SFR) over galactic scales (e.g. Schmidt 1959; Dopita 1985; Kennicutt 1989; Wyse & Silk 1989; Dopita & Ryder 1994; Prantzos & Aubert 1995; Kennicutt 1998). Infall may be important in determining the metallicity distribution of stars in the solar neighbourhood (e.g. Tinsley 1980; Prantzos & Aubert 1995; Pagel 1998). Other processes are more controversial: e.g. metal-enriched outflows (e.g. MacLow & Ferrera 1999) or radial gas flows (e.g. Edmunds & Greenhow 1995; Lacey & Fall 1985).
However, recent observational advances, coupled with the development of multiple metallicity stellar population synthesis codes has allowed comparison of galaxy evolution models with both the gas metallicities and colours of spiral galaxies (Contardo, Steinmetz & Fritze-von Alvensleben 1998; Jimenez et al. 1998; Boissier & Prantzos 2000; Prantzos & Boissier 2000; Cole et al. 2000). The colours of spiral galaxies depend on both their ages and metallicities, therefore study of their colours offers fresh insight into galaxy formation and evolution, although inevitably degeneracies remain.
In Bell & de Jong (2000; BdJ hereafter), we analysed the optical–near-infrared (near-IR) colours of a sample of 121 low-inclination spiral galaxies in conjunction with up-to-date stellar population synthesis models to explore trends in age and metallicity with galaxy parameters, such as magnitude or surface brightness. In particular, we found that there are significant trends between the age and $`K`$ band surface brightness of a galaxy, and between the metallicity and both the $`K`$ band magnitude and surface brightness of a galaxy. In that paper, we argued that these correlations could be the result of a surface density-dependent SFL, coupled with galaxy mass-dependent chemically-enriched gas outflows.
In this paper, we investigate these ideas in more detail. We use a family of simple models to explore the effects of infall, outflows, age differences and SFLs on the colour-based ages and metallicities of spiral galaxies. Our aim is not to construct a self-consistent, realistic model of galaxy formation and evolution. This work is intended to guide future, more detailed explorations of the star formation histories (SFHs) of spiral galaxies: this simple modelling isolates which physical processes affect which observables, to allow more realistic models to concentrate on formulating self-consistent prescriptions for the most important physical phenomena.
The plan of this paper is as follows. In section 2, we outline the data and its main limitations. We describe the chemical evolution model, the basic assumptions and equations and outline how we translate the model output into observables which we can readily compare with the data. In section 3, we describe the properties of the closed box model. In section 4 we explore the effects of infall, outflow and systematic trends in galaxy formation epoch. In section 5, we investigate the effects of changing the SFL on our results. In section 6 we discuss the results further, checking the plausibility of these models with other observational constraints. There, we also compare our models to a comparable, but more detailed model by Boissier & Prantzos. Finally, in section 7, we summarise our results.
## 2 The Method
### 2.1 The data
For this paper, we use the sample of 121 low-inclination spiral galaxies from BdJ. The sample is described in more detail in BdJ and in the sample’s source papers \[Bell et al. 2000, de Jong & van der Kruit 1994, Tully et al. 1996\]. The sample galaxies were chosen to have radially-resolved surface brightnesses in at least one near-IR and two optical passbands. The sample galaxies have a wide range of surface brightnesses, magnitudes, scale lengths and gas fractions, but are not complete in any statistical sense (at least as a unit).
In BdJ, we used a combination of at least one near-IR and two optical passbands to split (to some degree) the age-metallicity degeneracy. We fit simplified SFHs to the optical–near-IR colours using a maximum likelihood technique to derive crude age (reflecting the amount of recent to past star formation; c.f. a birthrate parameter) and metallicity estimates. These estimates are not accurate in an absolute sense: they are subject to uncertainties from a number of sources including model uncertainties, the effects of small bursts of star formation, the assumption of a single epoch of galaxy formation 12 Gyr ago and dust, to name a few. However, we argue that the estimates are robust in a relative sense: these uncertainties compromise the absolute ages and metallicities but leave the relative ranking of galaxies by age or metallicity relatively unaffected.
The ages and metallicities of one scale length wide annuli in our sample galaxies were estimated using the above method. We also constructed estimates of the age and metallicity at one disc half-light radius and their gradients per $`K`$ band scale length using a weighted linear fit to the ages and metallicities as a function of radius. More description of the method, its caveats and limitations can be found in BdJ.
### 2.2 Basic assumptions and equations
In order to make the investigation of the trends in age and metallicity a tractable problem, we adopt highly simplified ad hoc prescriptions describing star formation and galaxy evolution. These simple approximations allow us to investigate which effects play an important role in e.g. imprinting mass dependence in the SFH. We do not include gas flows in these models: assuming that the final total baryonic mass distribution is no different from a model without gas flows, the primary difference between models with and without gas flows will be the metallicity gradients (e.g. Lacey & Fall 1985; Edmunds 1990; Edmunds & Greenhow 1995). Therefore, the metallicity gradients are not iron-clad constraints on the models (in any case, the metallicity gradients are relatively unaffected by many of the changes explored in this paper, so the metallicity gradients were not particularly strong model constraints anyway).
For consistency with our age and metallicity estimation procedure, we assume that our model galaxy forms 12 Gyr ago as an exponential disc of gas with surface density $`\mathrm{\Sigma }_0(r)=\mathrm{\Sigma }_0(r=0)e^r`$, where $`\mathrm{\Sigma }_0`$ is the initial surface density of gas in M pc<sup>-2</sup>, and $`r`$ is the radius in units of the scale length of the gas (denoted by $`h`$). For the infall case (see section 4.1), $`\mathrm{\Sigma }_0(r)=0`$ initially, and the gas mass is gradually built up over time assuming an exponentially declining infall rate with $`e`$-folding time $`\tau _{\mathrm{infall}}`$. For the case in which we allow the galaxy formation epoch to vary as a function of its mass (see section 4.3), we change the galaxy formation epoch from 12 Gyr to between 4 and 12 Gyr, depending on galaxy mass.
The gas forms stars according to a prescribed SFL: in much of this paper we adopt a Schmidt (1959) SFL in terms of the gas surface density $`\mathrm{\Sigma }_{\mathrm{gas}}`$:
$$\psi =k\mathrm{\Sigma }_{\mathrm{gas}}^n,$$
(1)
where $`\psi `$ is the SFR in M pc<sup>-2</sup> Gyr<sup>-1</sup>, $`k`$ is the efficiency of star formation at a gas surface density of 1 M pc<sup>-2</sup> and $`n`$ is the exponent specifying how sensitively the SFR depends on gas surface density.
This star formation produces heavy elements; here we adopt the instantaneous recycling approximation (IRA; e.g. Tinsley 1980; Pagel 1998). A fraction $`R`$ of the mass of newly formed stars is instantaneously returned to the gas. We adopt a Salpeter (1955) initial mass function (IMF) with lower and upper mass limits of 0.1 M and 125 M respectively to describe the chemical and photometric evolution of our stellar populations: for this IMF the returned fraction $`R0.3`$. This gas is returned along with a mass $`p\psi (1R)`$ of heavy elements, where $`p`$ is the true yield, and is defined as the mass of freshly produced heavy elements per unit mass locked up in long-lived stars. The true yield $`p`$ is taken to be 0.02 (solar metallicity) hereafter unless explicitly stated otherwise. Note that, for simplicity, we assume a metallicity-independent yield. Note that our use of the IRA should not lead to significant inaccuracies, as the metallicity of spiral galaxies is typically measured via their oxygen content: for oxygen, the IRA is a fairly accurate approximation as it is produced primarily by Type II supernovae (c.f. Pagel 1998; although this assumption breaks down at late stages of galactic evolution near gas exhaustion; Portinari & Chiosi 1999; Prantzos & Boissier 2000).
Once the IRA is adopted, the following three equations specify the evolution of the galaxy completely:
$`{\displaystyle \frac{d\mathrm{\Sigma }_{\mathrm{gas}}}{dt}}`$ $`=`$ $`FE\psi (1R)`$ (2)
$`{\displaystyle \frac{d\mathrm{\Sigma }_{\mathrm{stars}}}{dt}}`$ $`=`$ $`\psi (1R)`$ (3)
$`{\displaystyle \frac{d(\mathrm{\Sigma }_{\mathrm{gas}}Z)}{dt}}`$ $`=`$ $`p\psi (1R)Z\psi (1R)Z_EE+Z_FF,`$ (4)
where $`\mathrm{\Sigma }_{\mathrm{stars}}`$ is the surface density of the stars in M pc<sup>-2</sup>, $`F`$ is the gas surface density infall rate (with an initial metallicity $`Z_F`$; we assume $`Z_F=0`$ hereafter) in M pc<sup>-2</sup> Gyr<sup>-1</sup>, $`E`$ is the surface density of gas ejected in outflows (with metallicity $`Z_E`$) in M pc<sup>-2</sup> Gyr<sup>-1</sup> and $`Z`$ is the gas metallicity (Tinsley 1980; Pagel 1998).
### 2.3 Determining ages and metallicities
We follow the evolution of the galaxy using a numerical scheme with a 20 Myr timestep. We split our model galaxies into 20 radial zones between $`r=0`$ and $`r=4`$ gas disc scale lengths, to allow study of both global and radial trends in age and metallicity. While some of the cases we study here have analytical solutions, our use of a numerical scheme allows us to use more complex SFLs and e.g. infall or outflow histories.
In order to properly compare the models with the data, we use the colour-based maximum-likelihood technique from BdJ to determine the ages and metallicities of our model galaxies. In order to use this technique, we must have a set of optical and near-IR colours for our model galaxies. Therefore, in each zone at each timestep, we use the total mass of newly-formed stars (both short- and long-lived) to compute the contribution of those stars to the total flux at the present day in $`U`$, $`B`$, $`V`$, $`R`$, $`I`$, $`J`$, $`H`$ and $`K`$ bands using interpolations between the multi-metallicity gissel98 models of Bruzual & Charlot (in preparation).
We then use these local colours as input to the maximum-likelihood age and metallicity estimator developed for and presented in BdJ. In this way, we obtain model ages and metallicities determined in exactly the same way as the observations we compare with. This can be quite important: especially so for older stellar populations. Mass-weighted average ages of older stellar populations can differ considerably from the luminosity-weighted ages derived using the colour-based technique because of only relatively modest amounts of recent star formation.
Using these local age and metallicity estimates, we construct estimates of the global age and metallicity gradients and intercepts (at the $`K`$ band disc half-light radius), using an unweighted least-squares fit (c.f. BdJ). $`K`$ band disc central surface brightnesses and scale lengths are determined by fitting the first 3 (gas) disc scale lengths of the surface brightness profile. Global gas fractions are determined by direct summation of the model gas and stellar masses.
The sample galaxies from BdJ cover a broad range of magnitudes and surface brightnesses. Therefore, to provide a fair comparison, the model galaxies must cover a similarly broad range of magnitudes and surface brightnesses. We adopt an empirical approach: we run a grid of 357 models with total (baryonic and dark) masses between $`10^9`$ M and $`10^{14}`$ M (we assume a baryon fraction of 0.05 hereafter), and central baryonic surface densities between $`10^{0.5}`$ M pc<sup>-2</sup> and $`10^{4.5}`$ M pc<sup>-2</sup>. The step size is 0.25 dex. This range is sufficient to cover the full observed range of parameter space probed in BdJ, assuming that the baryonic content of a galaxy turns entirely into solar-type stars.
However, this situation is complicated by the (broad) correlation between surface brightness and magnitude (c.f. Fig. 12 from BdJ, or the grey dots in panel a of Fig. 1). Because of this correlation between surface brightness and magnitude, any correlation between e.g. age and surface brightness will automatically translate into a correlation between age and magnitude. However, our model grid does not incorporate this correlation. Therefore, to provide a fair comparison with the data, we select galaxies from the model grid that fall within the region inhabited by the sample galaxies in the $`K`$ band surface brightness–absolute magnitude plane (Fig. 1) using the following criteria:
$`\mu _{K,0}`$ $`>`$ $`13+0.6(M_K+26)`$ (5)
$`\mu _{K,0}`$ $`<`$ $`20+0.6(M_K+26)`$ (6)
$`\mu _{K,0}`$ $`>`$ $`191.25(M_K+28)`$ (7)
$`\mu _{K,0}`$ $`<`$ $`231.25(M_K+21),`$ (8)
where $`\mu _{K,0}`$ is the $`K`$ band disc central surface brightness and $`M_K`$ is the $`K`$ band absolute magnitude of the galaxy. In this way, we can empirically select galaxies with a range of physical parameters consistent with those taken from BdJ. This approach ensures that the galaxies we produce automatically roughly satisfy the selection criteria of BdJ’s sample. We make no attempt to derive the allowed range of surface brightnesses and scale lengths on the basis of the angular momentum of infalling gas (e.g. Dalcanton, Spergel & Summers 1997; Mo, Mao & White 1998).
Note that galaxies with different sets of surface densities and masses may be chosen, depending on the model details (especially on the efficiency with which stars are turned into gas). We show an example of the selection box in panel a of Fig. 1 for the fiducial closed box model discussed in the next section. We show the $`K`$ band central surface brightness against the $`K`$ band absolute magnitude of the data (in grey) and the fiducial model (in black). The model points show clearly the selection criteria that are applied on the model central surface brightnesses and magnitudes: these selection limits are applied to preserve the broad correlation between surface brightness and magnitude in the dataset, and to make sure we do not compare the data with models of galaxies that are drastically different from those in the data. Panel b of Fig. 1 shows the $`K`$ band central surface brightness against gas fraction relation for the same fiducial model. All of the models presented in this paper are constrained to reproduce this correlation (although, in the case of e.g. infall models, there is some scatter in this correlation).
Now we use this model grid to investigate trends in age and metallicity with local and global structural parameters in sections 3 through 5, where we vary the galaxy evolution and SFL prescriptions. A summary of the models presented in the following sections is given in Table 1. The age, stellar metallicity and gas metallicity gradients are also given for these models in Table 2: note that there may be some slight mismatch in the properties of the observed and model galaxies (e.g. in panel a of Fig. 1 the model galaxies are regularly distributed in a rectangle, whereas the observed galaxies are clustered primarily towards the centre with a fairly significant contingent of galaxies with more extreme properties) which may affect the comparison of the average model and observed gradients slightly.
### 2.4 How we compare the models and the data
In the subsequent sections, we compare the colour-based ages and metallicities of a sample of face-on spiral galaxies with colour-based ages and metallicities from a suite of simple galaxy evolution models. However, earlier we stated that the colour-based ages and metallicities were only robust in a relative sense. This leaves open an important issue: how should we approach the comparison of the model ages and metallicities and the data?
To understand how we should approach these model comparisons, we need to go back to understanding the nature of the data. In essence, the trends in the data of e.g. Fig. 2 are describing trends in colour with galaxy properties. Therefore, if the black model points describe well the trends in age or metallicity with galaxy properties, then what the model really does is adequately describe the trend in optical–near-IR colour with galaxy properties. The model, of course, has one significant limitation: it interprets the trends in colour solely in terms of smoothly varying SFHs under the assumption of a constant IMF. The data, in contrast, has contributions to the colours from bursts of star formation, dust, and possibly from variations in the IMF.
Low-level bursts of star formation (e.g. variations of a factor of two in SFR over 0.5 Gyr timescales; Rocha-Pinto et al. 2000b) are relatively unimportant: these bursts only contribute modest extra scatter. IMF variations between different types of galaxy are disfavoured (see e.g. Kennicutt 1998; Bell & de Jong, in preparation). The overall IMF normalisation produces subtle effects, as the colours will change only very little for plausible changes inm the IMF, but the galaxy evolution will be affected through e.g. the fraction of mass locked up in long-lived stars ($`1R`$). Therefore, IMF uncertainties are unlikely to affect the qualitative behaviour of our models but may require the adoption of e.g. slightly different SFL parameters such as $`n`$ or $`k`$. Similar effects are expected for some types of model uncertainty or model age differences.
Dust is one major remaining uncertainty. The amounts and effects of dust are still hotly debated \[Disney, Davies & Phillipps 1989, Peletier & Willner 1992, Huizinga 1994, Tully et al. 1998, Kuchinski et al. 1998\], so being quantitative about the effects of dust is difficult. The zeroth order expectation is that dust is unlikely to be important for e.g. low surface brightness or luminosity galaxies, and is much more important for higher luminosity or surface brightness galaxies (e.g. Tully et al. 1998). In this scenario, young and metal poor galaxies would be little affected by dust; however, the older and more metal-rich galaxies would be likely to appear even older and more metal-rich, if dust was properly accounted for.
For these reasons, it is not fair to take the details of the model comparisons with the data too seriously. However, the relative trends shown by the comparison should be reasonably robust and it is certainly fair to compare the performance of different models. Also, along the same vein, we compare the models on a qualitative, visual level because no simple statistical approach (e.g. a two dimensional Kolmogorov-Smirnov test or least-squares line fitting) can take into account both the modelling uncertainties that we have discussed above and the role of selection effects in limiting the area of parameter space that is observed. We show all of the relevant plots for each model so that the reader can make their own comparisons and assessments of the different models.
## 3 Closed box model
Our fiducial model is a closed box model (i.e. $`E=F=0`$ in equations 2 through 4) with a Schmidt SFL, with efficiency $`k=0.012`$ M pc<sup>-2</sup> Gyr<sup>-1</sup> at a surface density of 1 M pc<sup>-2</sup> and $`n=1.6`$ (see Table 1; note that modest increases in $`n`$ are possible if $`k`$ decreases, and vice versa). The primary prediction of the closed box Schmidt SFL model is that the SFH and metallicity of a given area in any galaxy depend only on the initial local gas surface density. Note that the star formation and chemical enrichment history of a galaxy are independent of the initial density of gas if $`n=1`$, i.e. if the star formation rate is directly proportional to the gas density.
When we evaluate the effects of different galaxy evolution prescriptions, we will evaluate them with respect to this model. As such, it is worth spending some time on understanding what this model can and cannot reproduce, in terms of the trends observed in BdJ.
In Fig. 1 we show the fiducial model selection box and $`K`$ band central surface brightness–gas fraction correlation. As discussed in the last section, these two plots are used to select the galaxy models (panel a) and help to constrain the SFL parameters (panel b). In Fig. 2 we show the local age (panel a) and local metallicity (panel b) against the local $`K`$ band surface brightness. These local ages, metallicities and surface brightnesses are for individual one scale length wide annuli in the sample and model galaxies: in the limit of a SFL that depends on gas density only (such as the Schmidt law), these correlations should be unique, well-defined lines with no intrinsic width (e.g. the black model points).
In panels a and b of Fig. 2 we see that the fiducial model does a good job of reproducing the observed trends in age and metallicity with local $`K`$ band surface brightness. Although age and metallicity increase monotonically with surface brightness in the model, the ages (or metallicities) derived from the colours show more irregular trends with surface brightness. This is primarily due to degeneracies and irregularities in the colour-colour grid for stellar populations with near-constant star formation rates (i.e. average ages $`6`$ Gyr), which cause irregularities in the age–surface brightness and metallicity–surface brightness plots.
However, the observational data have significant scatter: while this is partly due to observational errors, there is a component of the scatter which is real intrinsic scatter in annulus colour at a given surface brightness. This indicates at least departures from a smooth SFH (e.g. Rocha-Pinto et al. 2000b), and potentially indicates dependence on some factor other than the local $`K`$ band surface brightness (BdJ). Thus, while the model gives a good match to the overall trend, additional factors need to be introduced to account for the scatter between annuli with the same $`K`$ band surface brightnesses.
Recall that we carried out linear fits to trends in age and metallicity with radius within a galaxy (yielding an intercept at the half-light radius and a gradient per $`K`$ band disc scale length). In panels c and d of Fig. 2 we explore the correlations between the age intercept at the half-light radius and the $`K`$ band central surface brightness (panel c) and absolute magnitude (panel d). We see that the fiducial model produces a correlation between the age at the half-light radius and the $`K`$ band central surface brightness (panel c: also between age and gas fraction; not shown). Note however that the slope of the model correlation is too shallow: real galaxies show a steeper correlation between age at the half-light radius and $`K`$ band central surface brightness than the model galaxies.
The selection limits imposed on the model galaxies produce a small residual correlation between age at the half-light radius and $`K`$ band absolute magnitude (panel d): this correlation is fictitious (age does not depend on mass in this model) and is the result of the magnitude–surface brightness correlation shown in panel a of Fig. 1. The correlation imposed by the magnitude–surface brightness correlation on panel d of Fig. 2 is too shallow compared to the data, however.
In panels e and f of Fig. 2 we explore correlations between the metallicity intercept at the disc half-light radius and the $`K`$ band central surface brightness (panel e) and absolute magnitude (panel f). Note that the model grid becomes quite uncertain at low metallicities: many of the grey data points at metallicities lower than $`\mathrm{log}_{10}(Z_{eff}/Z_{\mathrm{}})=1`$ may not have metallicities as low as those plotted (BdJ). The metallicity–surface brightness correlation (panel e: and metallicity–gas fraction correlation; not shown) is well-described by the fiducual model. We can also see from panel f of Fig. 2 that there is a small residual correlation between metallicity and magnitude imposed by the selection limits in panel a of Fig. 1, but that the slope of the correlation is too small: this indicates that metallicity has both a surface brightness and magnitude dependence (see e.g. BdJ; Skillman, Kennicutt & Hodge 1989; Garnett et al. 1997).
In panels g and h of Fig. 2 and Table 2, we explore the age (panel g) and metallicity (panel h) gradients per $`K`$ band disc scale length. Note that these gradients are considerably noisier than the intercepts that were discussed above: in particular, individual metallicity gradients are often only marginally detected (see BdJ for an indication of the uncertainties, and versions of these data plots with error bars).
Both the amplitude of and trends in age gradient with $`K`$ band central surface brightness are poorly reproduced by the fiducial model (panel g of Fig. 2): there is little if any correlation between age gradient and surface brightness in the model, and the model underpredicts the average age gradient in the data by around a factor of two (see Table 2). This problem may be linked with the inability of the model to reproduce the steepness of the global age–$`K`$ band central surface brightness correlation in panel c of Fig. 2, as the slope of the model correlation is too shallow, indicating that the rate of change of age with surface brightness (i.e. the age gradient) is underpredicted by the model. In contrast, the fiducial model reproduces both the average metallicity gradient and the trends in metallicity gradient with $`K`$ band central surface brightness (albeit with no scatter in the model metallicity gradients; Table 2 and panel h of Fig. 2).
To summarise, the fiducial model does a reasonable job of describing the correlations between the ages and metallicities of galaxies and their physical parameters. The main shortcomings of the fiducial model are a lack of magnitude dependence in both the age and metallicity (panels d and f of Fig. 2) and the underprediction of the rate of change of age with $`K`$ band surface brightness (Table 2 and panels c and g of Fig. 2). An additional shortcoming which we address more explicitly later in section 6 is that closed box models predict too many low metallicity stars in both the solar neighbourhood and in external galaxies (the G dwarf problem; e.g. Rocha-Pinto & Maciel 1996; Worthey, Dorman & Jones 1996; Pagel 1998). In the next section, we see how modifying the fiducial model by introducing infall, outflow or systematic trends in galaxy formation epoch can improve the match between the models and the observed correlations.
## 4 Galaxy evolution prescriptions
Can any of these shortcomings of the fiducial model be alleviated by invoking infall, outflow or variations in formation epoch between galaxies? Infall is a natural part of galaxy formation: at some level the disc of a spiral galaxy must be built up by infall, and a plausible explanation for the local G dwarf problem is an extended infall history at the solar cylinder (e.g. Tinsley & Larson 1978; Lacey & Fall 1983; Pagel 1998; Boissier & Prantzos 1999). Outflow is somewhat more speculative: successful heirarchical models of galaxy formation rely on negative feedback to supress star formation in low-mass systems at early times. Without feedback, these models generically produce too many faint galaxies to be compatible with the observed luminosity function (e.g. Kauffmann & Charlot 1998; Somerville & Primack 1999; Cole et al. 2000). The most likely source of feedback is the energy released by Type II supernovae. The feedback may take several forms: e.g. it may result in the preferential ejection of high metallicity gas or it may result in a ‘super-wind’ which completely removes the gas content of the galaxy (e.g. Dekel & Silk 1986; Arimoto & Yoshii 1987; MacLow & Ferrera 1999; Martin 1999). Significant differences in formation epoch (i.e. significant differences in the time at which a galaxy’s gas supply becomes available for star formation) between galaxies of e.g. different masses or halo spin parameters are quite possible. Semi-analytic, linear collapse and gas-dynamical cosmological simulations all suggest systematic trends in galaxy formation epoch with halo properties (e.g. Cole et al. 2000; Mo, Mao & White 1998; Steinmetz & Müller 1995). We explore the effects of infall, outflow and variations in formation epoch below in the next three subsections.
### 4.1 Infall
In order to test the effects of infall, we adopt a simple parameterisation. We assume that the infall timescale varies with both galaxy mass and radius within a galaxy: a radial dependence, and to a certain extent, a mass dependence in infall timescale are expected theoretically (e.g. Eggen, Lynden-Bell & Sandage 1962; Larson 1976; Contardo et al. 1998; Mo, Mao & White 1998; Boissier & Prantzos 2000; Cole et al. 2000). We have also run cases with only mass-dependent infall timescales, and even a mass-independent infall timescale; however, these cases are not discussed further as the effects of each of these individual cases are included (and are easily distinguished) in the infall model we consider.
We assume that the galaxy initially has a surface density of zero at all radii: we then build up the exponential disc of the galaxy from a reservoir of metal-free gas with an exponential timescale $`\tau _{\mathrm{infall}}`$. The infall timescale depends on both mass and radius. The timescale $`\tau _{\mathrm{infall}}`$ ranges from $`\tau _{\mathrm{infall}}=8`$ Gyr for galaxies with a total mass of $`10^{10}`$ M to $`\tau _{\mathrm{infall}}=0.5`$ Gyr for galaxies with a total mass of $`10^{13}`$ M according to the formula: $`\tau _{\mathrm{infall}}=82.5\mathrm{log}_{10}(M_{galaxy}/10^{10}\mathrm{M}_{\mathrm{}})`$ for galaxies with masses lower than $`M_{galaxy}=10^{13}`$ M; for galaxies more massive than $`10^{13}`$ M the infall timescale is fixed at 0.5 Gyr. Thus the most massive galaxies build up their gas mass quickly, and then evolve almost as a closed box, while low-mass galaxies have only recently reached their maximum gas mass and star formation rate. In this model, the infall starts at the same epoch in every galaxy and is always a decaying exponential: we investigate the effects of varying the formation epoch in Section 4.3.
The above infall timescales are defined at the half-light radius $`R_e`$. To impose radial variation, we reduce the central infall timescale by a factor of two and the infall timescale at $`2R_e`$ is doubled. As an example, a $`10^{12}`$ M galaxy in this model has infall timescales at $`(0,1,2)R_e`$ of $`(1.5,3,6)`$ Gyr.
All the model galaxies with infall have a Schmidt SFL with exponent $`n=1.8`$ and efficiency $`k=0.012`$. Note that these models have a different Schmidt law exponent from the fiducial model: this is because infall delays star formation compared to the closed box model, therefore star formation must be more efficient than in a closed box to provide a satisfactory match to the age observations.
From Fig. 3 (comparing it to Fig. 2 for the fiducial model) we can see that infall mainly affects the ages of the model galaxies. The reduction of the average age depends (of course) on the infall timescale: the maximum average age of a stellar population with an exponential infall timescale $`\tau _{\mathrm{infall}}`$ decreases as the infall timescale increases (and is roughly $`12\tau _{\mathrm{infall}}`$ Gyr for $`\tau _{\mathrm{infall}}4`$ Gyr). Metallicity is affected less: the difference between the mean stellar metallicity of a closed box and an exponential infall timescale model at a given gas fraction (therefore roughly surface brightness) is $`0.1`$ dex for any plausible SFL. Infall does ‘narrow’ the metallicity distribution of a stellar population, where there are relatively less low and high metallicity stars than the closed box model: this in fact was one of the original motivations for the infall model (e.g. Larson 1972; Tinsley 1980; Prantzos & Aubert 1995; Pagel 1998; Prantzos & Silk 1998).
The mass dependence in the infall timescale introduces a slope in the magnitude–age intercept relation (panel d in Fig. 3). The slope is too shallow to adequately describe the trends in the data; however, this is closely linked to the underprediction of the slope of the age intercept–central surface brightness correlation (panel c in Fig. 3) which feeds into a shallow slope for the age intercept–magnitude correlation. Tuning a better match with an infall model is difficult: the real problem is achieving a stronger variation of age with surface brightness which is still consistent with the data in panels a and c of Fig. 2. Perhaps this could be linked to a central surface density dependent infall history (e.g. through a spin parameter dependence in halo formation timescale); however, adding yet another parameter to the infall model is unappealing both in terms of the simplicity of the modelling technique and the quality of the data at this stage.
The radial dependence in the infall timescale increases the age gradient: this is illustrated in panel g of Fig. 3 and Table 2, where the metallicity gradients (both gas and stellar) are relatively unchanged from the closed box case, but the age gradient is increased by a factor of two. Both the radial and mass dependence increases the scatter in the local age–local $`K`$ band surface brightness relation, but not towards unacceptable levels.
To summarise: infall affects primarily the age of a region within a galaxy, but does not significantly affect the mean metallicity of the stellar population. Therefore, mass (or radial) variation of the infall timescale introduces mass dependence (or a radial gradient) in the age of a galaxy.
### 4.2 Outflow
In order to test the possible importance of outflow, we adopt a basic parameterisation of its effects. Outflow caused by supernovae winds is predicted to be much more effective for low-mass galaxies (Dekel & Silk 1986; Arimoto & Yoshii 1987; MacLow & Ferrera 1999; Martin 1999; Cole et al. 2000). Furthermore, for galaxies with total masses in excess of $`10^9`$ M, there may be little gas mass loss; however, many of the freshly-synthesized metals can be lost reasonably easily (MacLow & Ferrera 1999). Accordingly, we adopt this simple outflow recipe: galaxies with masses greater than $`10^{13}`$ M lose no metals in an outflow; galaxies with masses lower than $`10^{13}`$ M lose increasingly more and more metals with decreasing mass, down to a limit of $`10^9`$ M, where a galaxy loses 90 per cent of its freshly-synthesized metal content. We assume negligible gas mass loss in the outflow in order to separate the effects of outflow from those discussed in the previous section: the fraction $`f`$ of metal mass is given by $`f=0.90.225\mathrm{log}_{10}(M_{galaxy}/10^9\mathrm{M}_{\mathrm{}})`$ below $`M_{galaxy}=10^{13}`$ M; for $`M_{galaxy}10^{13}`$ M we assume no metal-enriched outflow. This approximation glosses over all the physics behind outflow; however, it does allow us to explore the possible effects of outflow in a simple, well-defined way. Note that there is no infall in this model.
Results from the outflow model (model O) are shown in Fig. 4. Note that we have increased the true yield $`p`$ from solar metallicity for the fiducial model to 1.5 solar metallicity for the outflow case: as outflow involves the loss of metals from the model galaxies, a larger yield must be used to allow the models to reproduce the mean observed metallicities. In contrast with the infall case, outflow leaves the colour-based age of a stellar population relatively unaffected. This is to be expected since the SFL only depends on gas density and since there is no gas mass loss in the outflow, the SFHs of the outflow model are the same as those from the closed box fiducial model. However, outflow profoundly affects the metallicity of a galaxy, causing a great deal of scatter in the metallicity–local $`K`$ band surface brightness diagram. The effects of outflow are visible in panels e and f of Fig. 4, where we plot the metallicity at the disc half-light radius against the galaxy parameters. Comparison of panels e and f of Fig. 4 for the outflow model with Fig. 2 for the fiducial (closed box) model clearly shows that outflow of this type produces a strong mass dependence in the galaxy metallicity; furthermore, a simple model of this type reproduces the metallicity scatter fairly accurately.
To summarise: outflow affects the metallicity of a galaxy. If outflow is mass dependent, it can imprint a metallicity-mass correlation without significantly affecting the age of the model galaxies.
### 4.3 Variation in Formation Epoch
In the previous three models we have assumed that all galaxies start forming stars at a common epoch. Heirarchical models of galaxy formation predict that more massive galaxies, despite being assembled later (from a number of smaller subunits), have the bulk of their stellar population forming earlier than less massive galaxies (e.g. Somerville & Primack 1999; Cole et al. 2000). Furthermore, stripped-down linear collapse models (e.g. Mo, Mao & White 1998; Dalcanton et al. 1997) as well as the more comprehensive gas-dynamical simulations (e.g. Steinmetz & Müller 1995; Contardo et al. 1998) predict that disc formation must happen relatively late to produce disc galaxies that have anywhere near enough angular momentum to match the observations (although note that gas dynamical cosmological simulations produce discs that are typically too small to match the observations, e.g. Navarro & Steinmetz 1997; 2000). However, we have assumed in the previous sections that all galaxies are formed at a look-back time of 12 Gyr.
To investigate the impact of variations in the formation epoch <sup>1</sup><sup>1</sup>1We reserve the term ‘formation epoch’ to indicate the look-back time at which stars start to from. We use the term ‘colour age’ (or simply ‘age’) to mean the average age of the stellar population as recovered from our colour inversion algorithm. on spiral galaxy colours, we allow the formation epoch of a galaxy to depend on its halo mass, where galaxies with masses greater than $`10^{13}`$ M have form at a look-back time of 12 Gyr, and galaxies with masses lower than $`10^{13}`$ M have formation epochs, $`E`$, smoothly decreasing from 12 Gyr at $`M_{galaxy}=10^{13}`$ M to 4 Gyr at $`10^9`$ M according to the formula $`E(\mathrm{Gyr})=4+2\mathrm{log}_{10}(M_{galaxy}/10^9`$ M$`{}_{\mathrm{}}{}^{})`$. This approximation, in the spirit of this semi-empirical modelling, glosses over the important physics behind these variations in formation epoch in an attempt to gain insight into their most important observable consequences. We always assume a galaxy formation epoch of 12 Gyr when interpreting the broad band colours in the age and metallicity fitting routine. This assumption is necessary: it was the assumption used to derive the observational ages and metallicities, and in the fitting routine we have no a priori knowledge of the formation epoch of galaxies. An additional benefit of this modelling is to demonstrate how much effect different galaxy formation epochs would have on the observational ages and metallicities that we derive using this fitting routine.
The variable formation epoch galaxies are assumed to be closed box systems, with a yield of 0.03 (1.5 times solar) and Schmidt SFL exponent of $`n=1.7`$ and efficiency at 1 M$`{}_{\mathrm{}}{}^{}\mathrm{pc}_{}^{2}`$ of $`k=0.012`$ M$`{}_{\mathrm{}}{}^{}\mathrm{pc}_{}^{2}\mathrm{Gyr}^1`$. The Schmidt SFL exponent was raised to 1.7 and the yield was increased to 0.03 to allow the rather younger galaxies reach gas fractions and metallicities at a given surface brightness which are in rough agreement with the observations. If the same set of Schmidt SFL model parameters were used as for the fiducial model the conclusions would be unchanged but the match to the observations would apprear slightly poorer. No infall or outflow is included in this set of models.
One important caveat: we parameterise the formation epoch model by simply starting the evolution of the galactic disc at a particular look-back time. In practice, it might be more suitable to think of this model as allowing large amounts of very late infall. For example, if a small fraction of stars were formed at early times but the bulk of stars were formed only a few Gyr ago or less, we still parameterise that galaxy by a single, relatively recent formation epoch. The basic point is that the formation epoch really parameterises when the bulk of the gas supply became available for star formation: how that gas supply became available is relatively unimportant from this perspective.
We show the effects of the mass-dependent formation epoch in Fig. 5. One important first result is that because our colour-based age estimator is essentially a birthrate parameter estimator (estimating the proportion of young to old stars), it is possible to have large apparent galaxy ages for galaxies which are, in fact, relatively young. As an example, it is possible to get colour-based ages in excess of 9 Gyr in a galaxy which is only 4 Gyr old. The important message is that the age derived using this technique is not to be taken as a literal stellar population age; it is simply a way of parameterising how the recent SFR compares to the SFR several Gyr ago.
We can see from Fig. 5, by comparison with the analagous figures for the closed box, infall and outflow models (Figs. 2, 3 and 4 respectively) that mass-dependent formation epoch variations between galaxies affects their ages and metallicities in ways similar to mass-dependent infall and outflow. In particular, formation epoch differences generate magnitude–age and magnitude–metallicity correlations that the fiducial model lacks.
### 4.4 Summary
To summarise:
* A local gas surface density-dependent Schmidt SFL describes many of the age and metallicity trends from BdJ surprisingly well. However, the slope of the age–central surface brightness correlation and the age gradients are underpredicted by the fiducial model.
* If the decay timescale of infall is mass-dependent, it can ‘imprint’ mass-dependency on the galaxy ages without significantly affecting their metallicities.
* If the infall timescale depends on radius, an age gradient can be generated without significantly affecting the metallicity gradient.
* If outflow is mass-dependent, it can ‘imprint’ a metallicity–mass correlation without significantly affecting the ages of the model galaxies.
* Variations in galaxy formation epoch do not invalidate our conclusion that a local surface density-dependent SFL describes the data reasonably well. However, mass-dependent variations imprint an age–magnitude and metallicity–magnitude relation even if no infall or outflow occurs.
In this way, the lack of mass dependence in the fiducial model may be provided either by mass dependence in the formation epochs of galaxies (where less massive galaxies are younger), or through a combination of mass dependent infall (where less massive galaxies have longer infall timescales) and outflow (where less massive galaxies lose a larger fraction of their newly-synthesized metal content). Furthermore, an ‘inside-out’ formation scenario may be able to reproduce the rather steep observed age gradients within individual galaxies.
We are therefore left in an uncomfortable situation: is there any way that we can differentiate between the effects of formation epoch variations versus infall and outflow? One possible approach is to study the effective yield of galaxies (i.e., the yield of a closed box model which reproduces the observed metallicity of a galaxy at its observed gas fraction). If the mass–metallicity correlation is due to mass-dependent variations in formation epoch then the effective yield should be constant. If, however, the mass–metallicity correlation is generated by mass-dependent metal-enriched outflows then the effective yield should strongly vary with galaxy mass, providing an observational test between the two possibilities.
We show the result of this experiment in Fig. 6 for the outflow and formation epoch models respectively. One can see that our naive expectation outlined above is confirmed: the outflow model (panel a of Fig. 6) has a strongly mass-dependent effective yield, whereas the formation epoch model has an almost constant colour-based effective yield (panel b of Fig. 6). The data do not clearly support either option: the effective yield of the faint, metal-poor galaxies denoted by crosses in Fig. 6 may be substantially underestimated due to stellar population model uncertainties at metallicities below 1/10 solar. However, if trend indicated by these points is correct, the colour-based effective yields marginally support the outflow model, although discounting the formation epoch model on data with this much uncertainty and scatter is clearly premature.
## 5 Star formation laws
We have not yet considered changes in the SFL: as we found that our results were quite sensitively affected by the SFL in the Schmidt law model, it is worth investigating if some of the shortcomings of the fiducial model can be alleviated by the use of an alternative SFL.
### 5.1 Density threshold
One possible modification to the star formation law is the imposition of a cutoff critical density $`\mathrm{\Sigma }_c`$ below which star formation cannot occur. Kennicutt (1989) found that the radially-averaged star formation rate (SFR) in a sample of 15 well-studied spiral galaxies was well-described by a critical density model. One physically motivated choice for that critical density is the maximum stable surface density of a thin isothermal gas disc, given by (Toomre 1964; Kennicutt 1989):
$$\mathrm{\Sigma }_c=\alpha \frac{\kappa c}{3.36G},$$
(9)
where $`\alpha `$ is a dimensionless constant of order unity, $`c`$ is the velocity dispersion of the gas (which we take, as Kennicutt did, to be a constant $`c=6`$ km s<sup>-1</sup> for all galaxies) and $`\kappa `$ is the epicyclic frequency (Kennicutt 1989). Kennicutt found that $`\alpha 0.67`$ was a good fit to his data: stars typically did not form in regions where the density was lower than the critical density, and formed according to a Schmidt law with index $`n1.3\pm 0.3`$ above the critical density threshold.
In order to estimate $`\mathrm{\Sigma }_c`$ for our model galaxies, we must assume a rotation curve: for simplicity, we adopt an adaptation of the mass-dependent ‘universal rotation curve’ from Persic & Salucci (1991) (where we adopt a baryonic mass of $`1.5\times 10^{11}\mathrm{M}_{\mathrm{}}`$ for a $`L^{}`$ galaxy with $`B`$ band absolute magnitude of $`20.6`$ assuming $`H_0=65`$ km s<sup>-1</sup> Mpc<sup>-1</sup>). We use this rotation curve to determine the critical density in our model galaxies: we allow stars to form according to a Schmidt law with $`n=1.8`$ and $`k=0.012`$ in gas denser than the critical density and do not allow stars to form at densities lower than the critical density. We do not use a critical density threshold in the central half scale length of the galaxy: apart from the undoubted influence of bars and bulges in the central regions of galaxies (which we do not include in this model), the universal rotation curve is undefined in the inner regions of a galaxy, and the critical density becomes very large in the innermost regions of spiral galaxies due to strong differential rotation. Thus, the behaviour of this star formation law at small radii is ill-constrained in this model: we, therefore, neglect the existence of a critical density at small radii and form stars according to a Schmidt law.
In Fig. 7 we show the ages and metallicities for a model with a critical density for star formation in which we adopt infall (see section 4.1). Comparison with Fig. 3 (and inspection of Table 2) shows that the imposition of a critical density does little to affect the colour-based ages or metallicities of spiral galaxies. Only regions with relatively low gas densities are affected significantly: regions near the centre of galaxies where the gas is depleted by star formation may be affected (although note that we do not impose the critical density within the central half scale length of our model galaxies), and the outermost regions where infall only recently brought the gas density above the critical density. In these outer regions the ages can be somewhat younger than those of the infall model without a critical density (compare the colour-based ages of those regions with the lowest $`K`$ band surface brightnesses in panel a of Figs. 7 and 3).
To summarise: the critical density threshold proposed by Kennicutt (1989) does little to affect the global correlations between SFH and physical parameters predicted by models with a Schmidt SFL. This is not to say that the existence of a critical density threshold has no effects at all: it is merely to state that the existence of a critical density threshold for star formation mainly affects the spatial distribution of present day star formation and does little to affect the star formation or chemical enrichment histories as probed by our colour-based technique.
### 5.2 Dynamical time dependence
An alternative SFL was proposed by Kennicutt (1998; Ke98 hereafter) based on his analysis of the global SFRs of a large sample of spiral and starburst galaxies. The SFR in this case scales with the ratio of the gas density to the local dynamical timescale:
$$\psi =k\mathrm{\Sigma }_{\mathrm{gas}}/\tau _{\mathrm{dyn}},$$
(10)
where $`\tau _{\mathrm{dyn}}=6.16R(\mathrm{kpc})/\mathrm{V}(\mathrm{km}\mathrm{s}^1)`$ Gyr is the dynamical timescale, here defined as the time taken to orbit the galaxy at a distance $`R`$. In this picture, the SFR is related to both the gas supply and the timescale in which the gas can be brought together. As we did for the critical density model, we use the mass-dependent universal rotation curve of Persic & Salucci (1991) to estimate the dynamical time as a function of radius in our model galaxies. This SFL is similar in many ways to the radially-dependent SFLs proposed by Wyse & Silk (1989) and allows us to explore how explicit radial dependence in the SFL affects how we interpret the trends in SFH with galaxy parameters presented in BdJ.
In Fig. 8 we show the ages and metallicities given by the dynamical time model (Model D) overplotted on the data from BdJ. The dynamical time dependence does three main things: it introduces significant (but reasonable) scatter in the local ages and metallicities, it produces anticorrelations between age/metallicity and magnitude, and it produces steeper age and gas metallicity gradients. The production of relatively steep age and metallicity gradients is a success of this type of model. Steep age gradients are expected from this model: in the limit of a flat rotation curve the Ke98 SFL depends linearly on gas density, and the star formation efficiency varies as $`1/R`$, leading to larger star formation timescales at larger radii. However, the observed trend in age gradient with surface brightness is not reproduced by the dynamical time model (or, indeed, the infall model): in both the dynamical time and infall model the age gradients are steeper for higher surface brightness galaxies, which is the opposite of the observed trend (Fig. 2).
Another more serious shortcoming of this model is the production of a ‘backwards’ age–magnitude and metallicity–magnitude correlation. This is at first sight counter-intuitive: brighter galaxies have larger rotation velocities, which would increase the SFR at a given gas mass and physical radius. However, brighter galaxies are also typically larger, therefore there is little change in the average ratio of radius to velocity, that is, there is little change in the typical dynamical time as a function of magnitude. There is obviously a scatter in these properties, leading to a scatter in SFHs. The ‘backwards’ age/metallicity–magnitude correlation is generated by the correlation between magnitude and surface brightness: small galaxies have older stellar populations in this model, and the smallest galaxies are also the faintest galaxies (see Fig. 1).
To summarise: the dynamical time model, due to its explicitly radius-dependent SFR, generates steeper age and metallicity gradients, compared to the fiducial closed box model. However, the dynamical time, on average, does not significantly depend on magnitude. Coupled with the surface brightness–magnitude correlation, this implies that the faintest galaxies appear, on average, older than the brightest galaxies, which is clearly at odds with the observations. Therefore a dynamical time dependence, alone, is not favoured by our dataset.
## 6 Discussion
In contrast to the previous sections, where we tuned the models to improve the match with the observational data, in this section we discuss three ‘blind’ applications of the models to literature data: the models were not tuned to fit these observations (solar cylinder data, global SFRs and gas metallicities), and they represent a powerful test of the model’s validity. We also present a comparison of our models and the model of Boissier & Prantzos (2000).
### 6.1 Properties of the Milky Way
Many chemical evolution models focus (sometimes almost exclusively) on the Milky Way (e.g. Wyse & Silk 1989; Steinmetz & Müller 1994; Prantzos & Aubert 1995; Boissier & Prantzos 1999; Portinari & Chiosi 1999): only after the models have been normalised to the Milky Way are the models extended to external galaxies, if at all (e.g. Wyse & Silk 1989; Boissier & Prantzos 2000; Prantzos & Boissier 2000). We have approached the problem from the other direction: we use the trends in galaxy properties with e.g. surface brightness or magnitude to learn what processes might be at play in determining the star formation and chemical evolution histories of galaxies. Nonetheless, it is interesting to check our models against observations of the ages and metallicities of stars in the solar cylinder, to provide a consistency check for our models.
In order to compare our models with the solar cylinder age and metallicity distributions, it is necessary to choose one model galaxy as a ‘Milky Way’. We choose to adopt a galaxy with a magnitude of around $`M_K^{}24`$ and a $`K`$ band disc scale length of around 2.5 kpc in the closed box model (this model galaxy has total mass of $`10^{12}`$ M and a baryonic central surface density of 1000 M pc<sup>-2</sup>). In Fig. 9 (panel a) we show the solar cylinder age–metallicity relation (points and error bars; Rocha-Pinto et al. 2000a) against the model predictions at a galactocentric radius of 8.5 kpc for the fiducial model (dotted lines) and the infall model I2 (solid lines). Note that stars from Rocha-Pinto et al. (2000a) older than 10 Gyr and younger than 15 Gyr have been put in a single bin centred at 11$`\pm `$1 Gyr. In panel b of Fig. 9 we show the metallicity (taken to be the oxygen abundance; determined from their iron abundance by applying the trend in \[O/Fe\] with iron abundance from Edvardsson et al. 1993) distribution of solar cylinder G stars taken from Rocha-Pinto & Maciel (1996) against the model distributions convolved with a gaussian of width 0.13 dex (to simulate the intrinsic metallicity spread of stars at a given age; Twarog 1980; Rocha-Pinto et al. 2000a).
From Fig. 9, we see that the models have some difficulties in reproducing the properties of the solar cylinder. Panel b of Fig. 9 clearly demonstrates that both our closed box and infall models have a significant G dwarf problem. This was expected: infall models require large amounts of late infall to solve the G dwarf problem (e.g. Pagel 1998). Possible solutions of this problem include pre-enrichment of the infalling gas from bulge and/or population III stars, more late infall, or some mixing of metals produced in outflows with the infalling gas. These additions are more complex processes that clearly need to be added to fine-tune the models. However, since any of the models could be tuned (in a few ways) to solve the galactic G dwarf problem, this problem does not in itself help us choose between models, and so these processes are not considered here for simplicity. The model age–metallicity relations are also not a perfect match to the data: our model age–metallicity relations are offset to lower metallicities, and show a steeper slope than the observed age–metallicity relation. While some of the slope mismatch may be due to the effects of observational errors (Rocha-Pinto et al. 2000a), the mismatches in the age–metallicity relation are related to the G dwarf problem: the model metallicity is always lower than the observations, especially so at early times.
The Milky Way analogue model gas metallicity gradient is $`0.055`$ dex kpc<sup>-1</sup>, which is broadly consistent with observational estimates of between $`0.05`$ and $`0.1`$ dex kpc<sup>-1</sup> (e.g. Vilchez & Esteban 1996; Smartt & Rolleston 1997; Pagel 1998). We therefore conclude that the models are a reasonable match to the observed ages and metallicities of solar cylinder stars, modulo a G dwarf problem, and that our approach of normalising the models on external galaxies has not seriously compromised the comparison with the Milky Way’s properties.
### 6.2 Comparison with the global star formation laws
Another important test is the comparison of the SFLs that we used in these models against observed SFRs. Here, we compare our fiducial Schmidt law model and the dynamical time model with observations of the globally averaged SFRs, gas densities and dynamical times of Ke98’s subsample of spiral galaxies. To make the comparison fair, we construct gas densities and SFRs averaged within the model galaxy’s $`R_{25}`$ (i.e. interior to the 25 $`B`$ mag arcsec<sup>-2</sup> isophote), and we take the dynamical time at $`R_{25}`$.
This comparison is presented in Figs. 10 and 11 for the fiducial model and dynamical time model respectively. The left-hand panels show the globally averaged SFR as a function of the gas density, and the right-hand panels show the globally averaged SFR as a function of the average gas density divided by the dynamical time. The data are denoted by grey circles and the models by black circles. We can see that both models provide a fair match to the slope of Ke98’s observations, and encompass the range of SFRs and gas densities typical of spiral galaxies. However, there is a significant zero-point offset between the model and the observations. A possible source of the zero-point offset is the efficiency of the SFL $`k`$: our model efficiencies are somewhat lower than e.g. those efficiencies used by Boissier & Prantzos (1999,2000). (Somewhat higher star formation efficiencies would be required to match the observations if we assumed younger galaxy ages or postulated large amounts of late gas infall as Boissier & Prantzos assume.) A higher efficiency raises the SFR at a given gas density, providing a better match to the observations. However, there are other possibilities: it is possible that there are aperture mismatches or discrepancies between the observational and model SFR calibrations (e.g. in the conversion of the H$`\alpha `$ flux to SFR).
### 6.3 Comparison with gas metallicities
Earlier, we compared the models to the colour-based stellar ages and metallicities of the sample of galaxies from BdJ. However, those ages and metallicities are subject to systematic errors from e.g. model uncertainties and dust reddening. For this reason, it is important to check the models against the gas metallicities derived from H ii region spectroscopy of spiral galaxies: these metallicities are subject to a completely independent set of uncertainties, and as such offer a chance to check our models.
We have chosen to compare our model gas metallicities with data from Garnett et al. (1997) in Fig. 12, where we show the surface density–local gas metallicity correlation for M33 and NGC 2403 from their Fig. 12. Data from Garnett et al. (1997) are denoted by grey dots and the fiducial model output is shown as black circles. The model metallicity–surface density correlation has the right slope but an offset zero point.
That the closed box fiducial model reproduces the right slope of the metallicity–surface density correlation disagrees with the conclusions of Phillipps & Edmunds (1991), who concluded that observed metallicity gradients exceed those achievable with a Schmidt SFL and closed box chemical evolution. However, inspection of their Figs. 1 and 2 suggests a good agreement between their simple Schmidt law model and the observations, indicating that they may have overestimated the size of the observational trend between metallicity and surface density when compared to the Schmidt law model.
The offset zero point is a well-known problem: when simple models are used to infer a yield from observations of the oxygen abundances of spiral galaxies, the yield is estimated to be around 1/3 solar (e.g. Garnett et al. 1997; Pagel 1998). However, we require a yield of around or somewhat above solar to match the observations: this discrepancy amounts to an offset of around 0.5 dex, which is of the order of the offset between the average metallicity of the models and the observations. This discrepancy between the yield estimates indicates a mismatch between the colour-based and gas metallicities, although it is as yet uncertain whether this is a deficiency in the calibration of the colour-based metallicities, gas metallicities, or even both. However, uncertainties in the stellar mass-to-light ratios used to derive the surface densities will propagate into this comparison. In addition, M33 and NGC 2403 are relatively faint (M$`{}_{B}{}^{}18.5`$), and because of the metallicity–magnitude correlation they are expected to have a somewhat lower metallicity than that of the fiducial model.
We do not show the comparison with their global gas metallicities here: the results are consistent with the overall yield offset between the gas and stellar metallicities outlined above and, again, the well-known gas metallicity–magnitude correlation (e.g. Skillman et al. 1989; Garnett et al. 1997) can only be adequately reproduced by the outflow or formation epoch model (as we found for the stellar metallicity–magnitude correlation). The average gas metallicity gradients of the model galaxies are presented in Table 2: the gas metallicity gradient per disc scale length of most models (with the exception of the dynamical time model) is within 10 per cent of the observational average, indicating that gas metallicity gradients have relatively little power to discriminate between the models.
### 6.4 Comparison with Boissier & Prantzos
Recently, Boissier & Prantzos discussed the properties of a specific, comprehensive disc galaxy spectro-photometric chemical evolution model in a trio of papers (Boissier & Prantzos 1999,2000; Prantzos & Boissier 2000). They constructed a specific spectro-photometric chemical evolution model that reproduced many observational constraints, and explored why these observational constraints were reproduced by their model. We have taken a complementary approach, where we have explored a wide range of physical processes to understand the effects of each process on the ages and metallicities of spiral galaxies. In this section, we compare the results of our modelling and theirs with a two-fold aim: to check that our studies give consistent results, and to gain some insight into how robust the conclusions of Boissier & Prantzos are likely to be if any of the modelling details were changed.
Their model used a combined Schmidt and dynamical time SFL: $`\psi =k\mathrm{\Sigma }_{\mathrm{gas}}^{1.5}/\tau _{dyn}`$, including surface density and mass-dependent infall, but, like our models, they did not include gas flows. They used a more sophisticated approach in dealing with the chemical evolution than adopted in this work (they did not use the IRA, but treated the full chemical evolution of the galaxy explicitly). They used halo circular velocities and spin parameters to parameterise their models but did not use the information about halo formation to fully specify the ages and infall histories of the discs they constructed (c.f. Dalcanton et al. 1997; Mo, Mao & White 1998). Instead, they assumed a constant galaxy formation epoch of 13.5 Gyr and tuned the surface density and mass dependence in the infall history to provide as good a match to the observations as possible.
Here, we focus on two issues: the origin of age and metallicity gradients in spirals, and the nature and origin of the mass–metallicity correlation.
One of the main conclusions of Prantzos & Boissier (2000) is that they required radial variation in both the infall timescale and the SFL to produce colour and metallicity gradients that were large enough to agree with the observations. However, we found earlier that explicit radial dependence of both the SFL and infall history is not required to reproduce stellar and gas metallicity gradients, and that radial dependence in either one can produce a sufficient effect.
Another of the significant successes of the model from Boissier & Prantzos (2000) is the reproduction of the metallicity–magnitude correlation. However, at first sight, this is somewhat of a puzzle as they do not include mass-dependent outflows or formation epoch differences in their model: their metallicity–magnitude correlation is reproduced entirely by a mass-dependent infall timescale. This puzzle is resolved by inspection of Fig. 3 of Boissier & Prantzos (2000): it shows that at surface densities typical of spiral discs, the infall prescription that they adopt increases with time for galaxies with halo circular velocities $`150`$ km s<sup>-1</sup>. This increasing infall rate mimics a variation in formation epoch. Therefore, we must modify our earlier conclusions: the infall history at early and intermediate times modifies primarily the colour-based ages of spirals; however, large amounts of late infall affects both the colour-based ages and metallicities of spirals. Furthermore, if the amount of late gas infall depends on galaxy mass, then a metallicity–magnitude relation can be generated. In their model, the metallicity–magnitude relation is generated by differences in galaxy formation epoch (essentially) and so the prediction of a mass-independent effective yield will hold for their model, just as it holds for our mass-dependent formation epoch model.
## 7 Conclusions and prospects
We have constructed a simple family of chemical evolution models with the aim of gaining some insight into the origins of many of the trends in SFH with galaxy parameters presented in BdJ. The model is used to generate colour-based ages and metallicities, which are directly comparable with those derived in BdJ. We generated a grid of model galaxies and selected only those which lie in a pre-defined region of the $`K`$ band absolute magnitude–central surface brightness plane.
Using this model, we have found the following:
* A local gas surface density-dependent Schmidt SFL describes many of the colour-based age and metallicity trends from BdJ surprisingly well. A model of this type does not explain the mass dependence in SFH required by BdJ and significantly underpredicts the age gradient in spiral galaxies and the slope of the age–central surface brightness correlation.
* The global properties of the fiducial model can be improved in either of two ways.
+ A combination of mass-dependent infall and metal enriched outflow imprint independent mass correlations on the galaxy colour-based ages and metallicities. Smaller galaxies have a more extended period of inflow in this model, and lose a greater fraction of their freshly-synthesized metals.
+ Galaxy formation epoch varies systematically with galaxy mass. If less massive galaxies are younger (i.e. if they assembled the bulk of their gas content at very late times) we explain the mass–metallicity and mass–age correlation without resorting to outflow.
* Regarding the radial variations within galaxies:
+ If the infall timescale varies with radius, an age gradient can be generated. This has little or no effect on the metallicity gradient.
+ Alternatively, Kennicutt’s (1998) SFL (which involves both gas density and the dynamical time) produces both strong age and gas metallicity gradients and a reasonable scatter in the local age/metallicity–surface brightness correlation. The main shortcoming of this model is a ‘backwards’ age/metallicity–magnitude correlation.
One deliberate limitation of our empirical approach is that we do not incorporate our model galaxies into a detailed cosmological context. For example, cold dark matter cosmologies make well-defined predictions about the formation mechanisms and infall histories of galactic discs that we ignore (e.g. Steinmetz & Müller 1994; Mo, Mao & White 1998; Somerville & Primack 1999; Cole et al. 2000). A more realistic treatment of the formation of our initial discs would be desirable, but is sensitive to the poorly understood details of angular momentum evolution in forming disc galaxies (e.g. Navarro & Steinmetz 1997; 2000). In this paper our philosophy has been to determine which of our conclusions seem the most robust to model details, and which may change if our initial disc formation was made more realistic. Our study gives an indication of which of our conclusions (and those of Boissier & Prantzos) are robust: i.e. which physical processes must operate in any galaxy formation scenario to reproduce the observations. From the above discussion, we find the following.
* Gas surface density determines the SFR, although other factors (such as dynamical time or a critical density for star formation) may also influence the SFR.
* Radial dependence in the SFL and/or the infall history is favoured.
* The infall of gas either varies strongly with galaxy mass, peaking at late times in low-mass galaxies (as parameterised by our formation epoch model, E, or the model of Boissier & Prantzos), or infall varies more weakly with galaxy mass but operates in conjunction with a higher efficiency of metal-enriched outflows in low-mass systems (Models I and O). The data (as it stands) marginally favours the latter option (Fig. 6).
However, the last conclusion is strongly dependent on the treatment of very low metallicity galaxies for which our colour-based ages and metallicities are highly uncertain. A combination of models I and O receives further support from studies of resolved stellar populations: there is ample evidence for older stellar populations in local faint, gas-poor and metal-poor dwarf Spheroidal galaxies. These cannot fit into the mass-dependent age scheme postulated above (e.g. Grebel 1998; Hurley-Keller, Mateo & Nemec 1998). Of course, it is quite possible that very low-mass dwarf galaxies have a metallicity–magnitude correlation driven primarily by outflows, and more massive galaxies (such as spirals, which have managed to keep the bulk of their gas content) have a metallicity–mass correlation driven primarily by differences in galaxy formation epoch. The key observable is the effective yield of galaxies: the effective yield gives insight into whether a galaxy is simply under-evolved and metal-poor (like, perhaps, low surface brightness disc galaxies; de Blok et al. 1996; Bell et al. 2000) or has had the bulk of its metals removed in gas outflows (as appears likely for dwarf Spheroidals: e.g. Dekel & Silk 1986). What is clear is that more work, both on observational and theoretical fronts, is required to fully elucidate the origin of the mass–metallicity correlation.
## Acknowledgements
We thank the anonymous referee for their helpful suggestions and comments, and Helio Rocha-Pinto for communicating results in advance of their publication. We would like to thank Don Garnett for providing data in electronic form and for his comments on the manuscript. We acknowledge useful discussions with and input from Matthias Steinmetz, Roelof de Jong, Rob Kennicutt and Andrew Benson. EFB acknowledges financial support from the Isle of Man Education Department, NASA grant NAG58426 and NSF grant AST9900789.
|
warning/0005/cond-mat0005223.html
|
ar5iv
|
text
|
# On the Theory of a New Maximum Observed in dc Transport in Modulated Quantum Hall Systems Near 𝜈=1/2
## Abstract
We propose a theory for the new maximum recently observed by Willett et al \[Phys. Rev. Lett. 83, 2624 (1999)\] in the longitudinal magnetoresistance of a weakly modulated two-dimensional electron gas (2DEG) near filling factor $`\nu =1/2`$ for the current driven along the modulation lines. The maximum is superimposed upon a new resonance structure. This occurs due to the geometric resonance of composite fermion cyclotron orbits with the period of modulation of the effective magnetic field. We propose here a semiquantitative theory of the dc magnetotransport in a modulated 2DEG near one half filling. Our analysis is based on the Boltzmann transport equation and it enables us to obtain this maximum which was neither observed nor predicted previously.
The present work is motivated by the new experimental results of dc transport experiments of Willett et al 1 in a two-dimensional electron gas (2DEG) in the fractional quantum Hall regime near half filling of the lowest Landau level ($`\nu =1/2`$). The 2DEG was modulated with a small period density modulation applied in one direction. A minimum in the magnetic field dependence of the dc resistivity component for a current driven across the modulation lines, and a maximum in the magnetoresistance corresponding to a current driven along the modulation lines, was observed at about $`\nu =1/2`$. Both minimum and maximum are superimposed upon a new resonance structure produced by the modulation in the region immediately adjacent to $`\nu =1/2`$.
Similar structures corresponding to a special kind of geometric resonance (so called Weiss oscillations) were observed in dc transport experiments in both electrostatic and magnetic field modulated 2DEG in low magnetic fields 2 ; 3 ; 4 ; 5 . A theory of these oscillations in modulated 2DEG systems was first developed within the framework of a quantum mechanical approach 6 ; 7 ; 8 . An equivalent semiclassical approach to the analysis of these phenomena was first proposed by Beenakker 9 who pointed that Weiss oscillations could be explained by means of the guiding center drift of cyclotron orbits of the electrons in the presence of the modulating electric field. The most complete semiclassical consideration of magneto-transport in a modulated 2DEG is presented in Ref. 10 (See also Ref. 11 ).
The minimum in the magnetoresistivity corresponding to a current driven across the modulation lines ($`\rho _{}`$) at $`\nu =1/2`$ was previously observed in the experiments 12 . A theoretical description of this feature of the magnetoresistivity $`\rho _{}`$ was presented in Refs. 13 ; 14 ; 15 . Here we focus on the maximum in the magnetic field dependence of the magnetoresistivity for a current driven along the modulation lines ($`\rho _{}`$). Such a maximum was never observed before and it was never predicted in theoretical studies. We develop a semiquantitative theory of the dc magnetotransport which gives a new explanation of this maximum observed in the experiments 1 .
Our work is based on the theory of the quantum Hall system at and near $`\nu =1/2`$ proposed by Halperin, Lee and Read (HLR) 16 ; 17 , which corresponds to the physical picture of the electrons decorated by attached quantum flux tubes. These are the relevant quasiparticles of the system – so called composite fermions (CF). The CFs are charged spinless fermionic quasiparticles which move in the reduced effective magnetic field $`B_{eff}=B4\pi \mathrm{}cn/e(n`$ is the electron density) 23 . At $`\nu =1/2`$ the CFs form a Fermi sea and exhibit a Fermi surface (FS). For the unmodulated 2DEG the CF-FS can be supposed to be a circle in quasimomenta space. Its radius $`p_F`$ equals $`\sqrt{4\pi n\mathrm{}^2}.`$
Density modulation influences the CF system through the direct effect of the modulating potential which can deform the CF–FS 22 and through the effect of an additional inhomogeneous magnetic field $`\mathrm{\Delta }B(𝐫)`$ proportional to the density modulation $`\mathrm{\Delta }n(𝐫)\left(\mathrm{\Delta }B(𝐫)=4\pi \mathrm{}c\mathrm{\Delta }n(𝐫)/e\right)`$ 14 . To analyse the effect systematically we have to solve the Boltzmann transport equation for the CF distribution function in the presence of a spatially inhomogeneous disturbance due to the density modulation, similarly to Refs. 10 ; 15 . When, however, the CFs mean free path $`l`$ is larger than the radius of their cyclotron orbit at the effective magnetic field $`R`$ and the period of modulation $`\lambda `$ we can obtain the desired response functions by means of simplified considerations based on the works of Beenakker 9 and Gerhardts 11 .
We start from the Lorentz force equations describing the CF motion along the orbit:
$$\frac{dp_x}{dt}=\frac{e}{c}B(𝐫)v_y;\frac{dp_y}{dt}=\frac{e}{c}B(𝐫)v_x,$$
(1)
where $`p_{x,y}`$ and $`v_{x,y}`$ are the components of the CF quasimomentum and velocity; $`B(𝐫)=B_{eff}4\pi \mathrm{}c\mathrm{\Delta }n(𝐫)/e.`$
We will consider a single-harmonic sinusoidal density modulation of period $`\lambda =2\pi /g`$ along the $`{}_{}{}^{\prime \prime }y_{}^{\prime \prime }`$ direction: $`\mathrm{\Delta }n(𝐫)\mathrm{\Delta }n(y)=\mathrm{\Delta }n\mathrm{sin}(gy).`$ We assume that the correction term $`\mathrm{\Delta }B(𝐫)`$ is small compared to $`B_{eff}.`$ Under this assumption we can write the CF velocity $`𝐯`$ in the form $`𝐯=𝐯_0+\delta 𝐯,`$ where $`𝐯_0`$ is the uniform-field velocity and the correction $`\delta 𝐯`$ arises due to the inhomogeneity of the magnetic field. For a circular CF-FS we have: $`v_{x0}=v_F\mathrm{cos}\mathrm{\Omega }t;v_{y0}=v_F\mathrm{sin}\mathrm{\Omega }t;\mathrm{\Delta }n(y)\mathrm{\Delta }n\mathrm{sin}(gYgR\mathrm{cos}\mathrm{\Omega }t),`$ where $`v_F`$ is the CFs Fermi velocity, $`\mathrm{\Omega }`$ is their cyclotron frequency in the effective magnetic field and $`Y`$ is the $`{}_{}{}^{\prime \prime }y_{}^{\prime \prime }`$ coordinate of the guiding center. Substituting these expressions for $`𝐯`$ and $`\mathrm{\Delta }n(y)`$ into Eq. (1) and keeping only first-order terms we obtain:
$$\frac{d(\delta v_x)}{dt}=\mathrm{\Omega }\delta v_y\frac{\mathrm{\Delta }B}{B_{eff}}\mathrm{\Omega }v_F\mathrm{sin}\mathrm{\Omega }t\mathrm{sin}(gYgR\mathrm{cos}\mathrm{\Omega }t);$$
$$\frac{d(\delta v_y)}{dt}=\mathrm{\Omega }\delta v_x+\frac{\mathrm{\Delta }B}{B_{eff}}\mathrm{\Omega }v_F\mathrm{cos}\mathrm{\Omega }t\mathrm{sin}(gYgR\mathrm{cos}\mathrm{\Omega }t).$$
(2)
We have to remark here that in the presence of the density modulation the CF’s Fermi velocity gets a correction due to the modulation. To evaluate this correction we calculate the average of $`\mathrm{\Delta }n(y)`$ over the cyclotron orbit. Expanding the functions $`\mathrm{cos}(gR\mathrm{cos}\mathrm{\Omega }t)`$ and $`\mathrm{sin}(gR\mathrm{cos}\mathrm{\Omega }t)`$ in Bessel functions we arrive at the following result for the averaged correction to the inhomogeneous density modulation:
$`<\mathrm{\Delta }n(y)>`$ $``$ $`{\displaystyle \frac{\mathrm{\Delta }n}{2\pi }}{\displaystyle _0^{2\pi }}\mathrm{sin}(gYgR\mathrm{cos}\psi )𝑑\psi `$ (3)
$`=`$ $`\mathrm{\Delta }n\mathrm{sin}(gY)J_0(gR).`$
Here $`\psi =\mathrm{\Omega }t.`$
The result (3) gives a spatially inhomogeneous correction to the chemical potential of the CFs and to their Fermi velocity:
$$\stackrel{~}{v}_F=v_F\sqrt{1+\frac{\mathrm{\Delta }n}{n}\mathrm{sin}(gY)J_0(gR)}.$$
The same holds for the scattering rate of CFs. In periodically modulated samples the scattering rate can be a function of a coordinates. Within the chosen geometry (the modulation is applied along the $`\mathrm{"}y\mathrm{"}`$ direction) we can describe a periodic modulation of the relaxation time following Ref. 10 :
$$\frac{1}{\tau (y)}=\frac{1}{\tau }+r(y).$$
The correction $`r(y)`$ here is proportional to $`\mathrm{\Delta }n(y)/n`$. However we can neglect these corrections to the CF Fermi velocity and their relaxation time in further calculations because the parameter $`(\mathrm{\Delta }n/n)=(\mathrm{\Delta }B/B)`$ is much smaller than the parameter $`\mathrm{\Delta }B/B_{eff}`$. Thus omitted terms are in order of magnitude smaller than those which are kept. It is natural to suppose that to the first order in the modulating field the corrections $`\delta v_x`$ and $`\delta v_y`$ are periodic over the unperturbed cyclotron orbit. This assumption is equivalent to that used in Ref. 11 . Under this assumption we can calculate averages of Eq. (2) over the cyclotron orbit. This gives us the following expressions for the components of the velocity of the guiding center drift $`V_x`$ and $`V_y`$ defined below:
$`V_x(Y)`$ $`=`$ $`<\delta v_x>={\displaystyle \frac{v_F}{2\pi }}{\displaystyle \frac{\mathrm{\Delta }B}{B_{eff}}}{\displaystyle _0^{2\pi }}\mathrm{cos}\psi \mathrm{sin}(gY`$
$``$ $`gR\mathrm{cos}\psi )d\psi =v_F{\displaystyle \frac{\mathrm{\Delta }B}{B_{eff}}}\mathrm{cos}gYJ_1(gR);`$
$`V_y(Y)`$ $`=`$ $`<\delta v_y>={\displaystyle \frac{v_F}{2\pi }}{\displaystyle \frac{\mathrm{\Delta }B}{B_{eff}}}{\displaystyle _0^{2\pi }}\mathrm{sin}\psi \mathrm{sin}(gY`$ (4)
$``$ $`gR\mathrm{cos}\psi )d\psi =0.`$
To evaluate the CF conductivity semiquantitatively we assume that the $`x`$ component of the CF velocity can be written in the form $`v_x(Y)=v_{x0}+V_x(Y).`$ We also assume that the cyclotron frequency $`\mathrm{\Omega }`$ can be replaced by the quantity $`\mathrm{\Omega }(Y)=\mathrm{\Omega }+<\mathrm{\Delta }\mathrm{\Omega }(y)>`$ where $`<\mathrm{\Delta }\mathrm{\Omega }(y)>`$ is the correction to the cyclotron frequency arising due to the inhomogeneity of the effective magnetic field averaged over the cyclotron orbit:
$`\mathrm{\Omega }(Y)`$ $`=`$ $`\mathrm{\Omega }\left\{1+{\displaystyle \frac{\mathrm{\Delta }B}{B_{eff}}}{\displaystyle \frac{1}{2\pi }}{\displaystyle _0^{2\pi }}\mathrm{sin}(gYgR\mathrm{cos}\psi )𝑑\psi \right\}`$ (5)
$`=`$ $`\mathrm{\Omega }\left\{1+{\displaystyle \frac{\mathrm{\Delta }B}{B_{eff}}}\mathrm{sin}(gY)J_0(gR)\right\}.`$
We showed before 22 that in the semiquantitative analysis of the magnetotransport in modulated the 2DEG we can use the following approximation for the CF conductivity:
$$\sigma _{\alpha \beta }^{cf}\frac{g}{2\pi }_{\pi /g}^{\pi /g}\sigma _{\alpha \beta }^{cf}(Y)𝑑Y,$$
(6)
where
$$\sigma _{\alpha \beta }^{cf}(Y)=\frac{e^2m_c\tau }{2\pi \mathrm{}^2}\underset{k}{}\frac{v_{k\beta }(Y)v_{k\beta }(Y)}{1+ik\mathrm{\Omega }(Y)\tau }.$$
(7)
Here $`m_c`$ is the CF cyclotron mass; $`\tau `$ is the relaxation time; $`v_{k\beta }(Y)`$ are the Fourier transforms for the CF velocity components:
$`v_{kx}={\displaystyle \frac{v_F}{2}}(\delta _{k,1}+\delta _{k,1})+V_x(Y)\delta _{k,0};`$
$`v_{ky}={\displaystyle \frac{iv_F}{2}}(\delta _{k,1}\delta _{k,1}).`$
Keeping terms of the order of $`(\mathrm{\Delta }B/B_{eff})^2`$ or larger we obtain the following approximations for the CF conductivity components $`\left({\displaystyle \frac{\mathrm{\Delta }B}{B_{eff}}}\mathrm{\Omega }\tau ={\displaystyle \frac{\mathrm{\Delta }n}{n}}{\displaystyle \frac{p_F}{\mathrm{}}}l={\displaystyle \frac{\mathrm{\Delta }n}{n}}k_Fl\right):`$
$`\sigma _{xx}^{cf}`$ $``$ $`{\displaystyle \frac{\sigma _0}{1+(\mathrm{\Omega }\tau )^2}}\left\{1+{\displaystyle \frac{3}{2}}\left({\displaystyle \frac{\mathrm{\Delta }n}{n}}k_Fl\right)^2{\displaystyle \frac{J_0^2(gR)}{1+(\mathrm{\Omega }\tau )^2}}\right\}`$ (8)
$`+`$ $`\sigma _0\left({\displaystyle \frac{\mathrm{\Delta }B}{B_{eff}}}\right)^2J_1^2(gR);`$
$`\sigma _{yy}^{cf}`$ $``$ $`{\displaystyle \frac{\sigma _0}{1+(\mathrm{\Omega }\tau )^2}}\left\{1+{\displaystyle \frac{3}{2}}\left({\displaystyle \frac{\mathrm{\Delta }n}{n}}k_Fl\right)^2{\displaystyle \frac{J_0^2(gR)}{1+(\mathrm{\Omega }\tau )^2}}\right\};`$
$`\sigma _{xy}^{cf}`$ $`=`$ $`\sigma _{yx}^{cf}`$
$``$ $`{\displaystyle \frac{\sigma _0\mathrm{\Omega }\tau }{1+(\mathrm{\Omega }\tau )^2}}\left\{1+{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{\mathrm{\Delta }n}{n}}k_Fl\right)^2{\displaystyle \frac{J_0^2(gR)}{1+(\mathrm{\Omega }\tau )^2}}\right\}.`$
where $`\sigma _0=ne^2l/p_F`$ is the CF conductivity in a homogeneous magnetic field.
The last term in the expression for $`\sigma _{xx}^{cf}`$ describes the contribution from CFs diffusing along the $`{}_{}{}^{\prime \prime }x_{}^{\prime \prime }`$ direction which arises due to the guiding center drift. To show it we can calculate the corresponding contribution to the diffusion coefficient $`\delta D.`$ Following Refs. 9 ; 11 we write:
$$\delta D=\tau \frac{g}{2\pi }_{\pi /g}^{\pi /g}V_x^2(Y)𝑑Y.$$
(11)
This term $`\delta D`$ gives the additional contribution to the $`{}_{}{}^{\prime \prime }x_{}^{\prime \prime }`$ component of the diffusion tensor $`D.`$ The latter is connected with the CF conductivity through the Einstein relation $`\sigma _{\alpha \beta }^{cf}=Ne^2D_{\alpha \beta }(N`$ is the CF density of states). Substituting Eq. (11) into this relation we obtain the expression for this diffusion correction to $`\sigma _{xx}^{cf}`$ which coincides with the last term in Eq. (8).
According to the HLR theory, the 2DEG resistivity tensor $`\rho `$ equals: $`\rho =\rho ^{cf}+\rho ^{cs}`$ where $`\rho ^{cf}`$ is the CF resistivity tensor $`\left(\rho ^{cf}=(\sigma ^{cf})^1\right)`$ and the contribution $`\rho ^{cs}`$ arises due to a fictitious magnetic field which originates from the Chern-Simons formulation of the theory. The latter has only off diagonal elements. Hence the diagonal components of the 2 DEG resistivity tensor coincide with the corresponding components of the CF resistivity tensor $`\rho ^{cf}.`$ After straightforward calculations we arrive at the result:
$`\rho _{xx}`$ $``$ $`{\displaystyle \frac{1}{\sigma _0}}\left\{1+\left({\displaystyle \frac{\mathrm{\Delta }B}{B_{eff}}}\right)^2\chi _1(gR)\right\}^1;`$ (12)
$`\rho _{yy}`$ $``$ $`{\displaystyle \frac{1}{\sigma _0}}\left\{1+\left({\displaystyle \frac{\mathrm{\Delta }n}{n}}k_Fl\right)^2{\displaystyle \frac{\chi _2(gR)}{1+(\mathrm{\Delta }B/B_{eff})^2\chi _3(gR)}}\right\}.`$
Here $`\chi _i(gR)=\alpha _iJ_0^2(gR)+J_1^2(gR)(i=1,2,3)`$ and the coefficients $`\alpha _i`$ are given by the expressions:
$$\alpha _2=0;\alpha _1=\frac{1}{2}\frac{(\mathrm{\Omega }\tau )^2}{1+(\mathrm{\Omega }\tau )^2};\alpha _3=\frac{(\mathrm{\Omega }\tau )^2}{1+(\mathrm{\Omega }\tau )^2}.$$
(14)
When the density modulation is very weak $`\left[(\mathrm{\Delta }n/n)k_Fl1\right]`$ the corrections to the magnetoresistivity are small and we can neglect them. Under this condition the inhomogeneity of the effective magnetic field does not significantly affect dc transport. For stronger modulation $`\left[(\mathrm{\Delta }n/n)k_Fl1\right]`$ the resistivity component $`\rho _{}`$ can be significantly changed due to the modulation. Keeping only the greatest correction (in the small parameter $`\mathrm{\Delta }B/B_{eff}`$) we can write for this component of the magnetoresistivity the following approximate expression:
$$\rho _{}\frac{1}{\sigma _0}\left\{1+\left(\frac{\mathrm{\Delta }n}{n}k_Fl\right)^2J_1^2(gR)\right\}.$$
We also see that the resistivity component $`\rho _{}`$ can be considerably changed due to the modulation:
$$\rho _{}\frac{1}{\sigma _0}\left[1+\left(\frac{\mathrm{\Delta }n}{n}k_Fl\right)^2\frac{1}{(\mathrm{\Omega }\tau )^2}\left(\frac{1}{2}J_0^2(gR)J_1^2(gR)\right)\right].$$
(15)
Our result for $`\rho _{}`$ is in general agreement with the previous theory (See e.g. Ref. 10 ) For $`gR1`$ our formula coincides with the corresponding results of 10 ; 11 for a simple harmonic modulation. Apart from $`\nu =1/2`$ when the parameter $`gR`$ is of the order of or smaller than unity (i.e. the radius of the CF cyclotron orbit is of the order of or smaller than the modulation wavelength), we have to modify our results for the magnetoresistivity component $`\rho _{}`$. The reason is that our approximation for $`\rho _{}`$ is valid within the relaxation time approximation in the Boltzmann equation with relaxation towards the total instead of the local equilibrium distribution function (See Ref. 10 ). For $`gR1`$ the difference between the total and local distribution functions becomes significant. It follows from the results of a detailed and systematic analysis that for $`gR1`$ we have to modify the correction term arising due to the modulations in the expression for $`\rho _{}`$. The function $`J_1^2(gR)`$ has to be replaced by $`J_1^2(gR)/(1J_0^2(gR))`$. This denominator originates from the back scattering term in the collision operator in the Boltzmann equation which describes relaxation towards the local equilibrium. Thus it is the consequence of the continuity equation for CFs.
It was shown in Ref. 10 that the expression for the magnetoresistivity component $`\rho _{}`$ is not changed if we assume the relaxation is towards either local or total equilibrium. Thus we can use our semiquantitative expression (15) for $`\rho _{}`$ within the range of strong effective magnetic fields when $`gR<1`$. We remark that our result (15) for $`\rho _{}`$ disagrees with the previous theory presented in Refs. 10 ; 13 ; 14 ; 15 . It was concluded in these papers that magnetic modulations themselves cannot influence this component of the magnetoresistivity. This discrepancy originates from the difference in the procedure of calculation of the response functions averaged over the period of modulations. Here we first introduce the average conductivity \[See Eqs. (6) and (7)\] and after that we convert it to the magnetoresistivity tensor, in contrast to Refs. 10 ; 13 ; 14 ; 15 where the average is taken last. It enables us to keep corrections proportional to $`\left(\mathrm{\Delta }\mathrm{\Omega }/\mathrm{\Omega }\right)^2J_0^2(gR)`$.
These corrections are missed when we first calculate the resistivity tensor and then average it over the period of modulation as in Ref. 10 . Another source of the discrepancy is the diffusion correction included to $`\sigma _{xx}`$. However this term has a clear physical sense and originates from the extra current along the $`\mathrm{"}x\mathrm{"}`$ direction due to the guiding center drift.
For $`gR<1`$ the correction to the magnetoresistivity component $`\rho _{}`$ takes on values of the order of unity and increases upon increase of the effective magnetic field. This corresponds to a minimum in the magnetic field dependence of $`\rho _{}`$ around $`\nu =1/2`$. Such a minimum was observed in experiments 1 ; 12 . The magnetic field dependence of the resistivity component $`\rho _{}`$ for $`gR1`$ is more complicated. Within a certain range of $`gR`$ we may observe a maximum of $`\rho _{}`$ about $`\nu =1/2`$. The maximum can be developed when the CF mean free path and the modulation period are large enough to obey the inequalities $`l>R`$ and $`gR<1`$ for sufficiently small $`B_{eff}.`$ Within another interval of $`gR`$ this maximum can be converted to a minimum and for some values of $`gR`$ this component of the magnetoresistivity will exhibit only a very weak dependence on the effective magnetic field. The character of this dependence is determined by the relation between Bessel functions $`J_0^2(gR)`$ and $`J_1^2(gR)`$ included in the second term of our expression (15).
We conjecture that by changing the modulation wavelength $`\lambda `$ we can observe all three kinds of the dependence of $`\rho _{}`$ upon the magnetic field for the same experimental sample. Actually a very weak dependence of the resistivity component $`\rho _{}`$ on the effective magnetic field near $`\nu =1/2`$ was observed in the experiments of Ref. 12 and a maximum was observed in the experiments of Ref. 1 . In both cases the resistivity component $`\rho _{}`$ exhibits a large maximum about $`\nu =1/2`$.
We tried to describe the maximum in the magnetic field dependence of the longitudinal magnetoresistivity about $`\nu =1/2`$ using our result (15) for the parameters close to the parameters of the corresponding experiments of 1 : $`n1.1\times 10^{11}`$ cm$`{}_{}{}^{1},p_F1.2\times 10^{21}`$g cm/s$`{}_{}{}^{2},\lambda 1.5\mu `$m,$`l10^4`$ cm and we obtained a maximum in the longitudinal magnetoresistivity near $`\nu =1/2`$. The magnitude of the this maximum strongly depends on the magnitude of the density modulations \[see Fig. 1(a)\]. However for a circular CF-FS the maximum about $`\nu =1/2`$ is noticeably smaller in magnitude than the experimental results of 1 for reasonable values of $`\mathrm{\Delta }n/n.`$ We can reach much better agreement with this experiment for a noncircular CF-FS. In principle, a noncircular shape of the CF-FS may originate from the effect of the crystalline field of adjacent layers of GaAS/AlGaAs. For clean samples like those used in the experiments of 1 this effect can be noticeable. Suppose that the CF-FS is not a circle but an ellipse. In this case the parameter $`gR=gp_1c/|e|B_{eff}`$ includes the maximum value of the CF quasimomentum component $`p_x`$ which does not coincide with $`p_F`$. This can give us a noticeable enhancement in the magnitude of the maximum as is shown in Fig. 1(b). Plotting the curves in this figure we assumed that $`p_1`$ is $`1.0\times 10^{21}`$ g cm/s$`^2.`$ Using this value of $`p_1`$ we get reasonable agreement with the experimental data of Willett et al as is shown on Fig. 2.
Our simplified formula (20) cannot be applied to the region immediately adjacent to $`\nu =1/2`$ where the condition $`\mathrm{\Delta }B/B_{eff}1`$ is not satisfied. So we cannot analyze the effect of channeled orbits of CFs which occur immediately adjacent to $`\nu =1/2`$ where the effective magnetic field is of the same order of magnitude as the oscillating correction $`\mathrm{\Delta }B.`$ To analyze the dc response of the modulated 2DEG for $`\mathrm{\Delta }BB_{eff}`$ we have to solve the CF transport problem consistently for example following the way treated in Ref. 10 . Nevertheless our semiquantitative approach captures the essential physics of the considered effect and gives simple analytical results applicable for the comparison with experimental data.
In summary, we show that within the single relaxation time approximation we obtain good semiquantitative agreement with experimental observation of a maximum in parallel resistivity $`\rho _{}`$, observed in the experiments of Ref. 1 . We predict that the magnetic field dependence of this component of the magnetoresistivity can be of different character for different $`gR`$ i.e by changing the wavelength of modulation the maximum in $`\rho _{}`$ can be converted to a minimum or to a negligeably weak dependence of this resistivity component upon the effective magnetic field. This maximum as well as the minimum in the magnetic field dependence of the resistivity $`\rho _{}`$ about $`\nu =1/2`$ arises due to the inhomogeneity of the effective magnetic field in a modulated 2DEG near one half filling.
We thank W.R. Willett for kindly giving us a preprint of reference 1 and G.M. Zimbovsky for help with the manuscript. We also thank S. Das Sarma for discussions of possible asymmetry effects on the CF-FS due to the crystal field present on the ultra clear surfaces of AlGaAs, Used in the magnetotransport experiments. Support from a PSC-CUNY FRAP ”In – Service” Award is acknowledged.
|
warning/0005/astro-ph0005493.html
|
ar5iv
|
text
|
# Anomalous Radio-Wave Scattering from Interstellar Plasma Structures
## 1 Introduction
Images of scattered radio sources and distorted pulses from pulsars provide some of the most-used observables for probing microstructure in the electron density of interstellar gas. Over the last decade, interstellar scattering measurements have revealed asymmetries in the scattered images of radio sources. These are most often interpreted in terms of underlying anisotropy of the very small irregularities that diffract the radiation. That anisotropy, in turn, most likely reflects the orientation of magnetic fields in the H II gas that contains the microstructure. Angular broadening of compact sources and pulse distortions due to multi-path propagation are used to probe the amplitude of scattering and also, through the frequency scaling, to constrain the shape of the wavenumber spectrum for the microstructure. Inversion of scattering observables into information about the microstructure almost invariably relies on the assumption that the scattering strength is uniform in directions transverse to the line of sight.
We reconsider the assumptions used to analyze angular and temporal broadening, in particular the assumption of uniformity of the scattering medium transverse to the line of sight. One reason is that the interstellar medium (ISM) shows structures on a wide variety of scales and so it is reasonable to expect manifestations of nonuniformities, at least in some directions. Secondly, the physics that underlies asymmetric images is quite different if the asymmetry occurs on scales much larger than diffractive scales, as they would if the asymmetry is caused by the large-scale distribution of diffracting irregularities. Thirdly, observations of the Crab pulsar show anomalous scalings of pulse broadening with frequency. These are interpreted by some as indicating that scattering occurs within the pulsar magnetosphere rather than in a cold plasma (J. Eilek 1997, private communication; Hankins & Moffett 1998; Lyutikov & Parikh 2000). As we show, anomalous scalings occur quite naturally from cold plasma extrinsic to the pulsar if the scattering region is bounded in the transverse direction. Additionally, future observations at low radio frequencies of a variety of sources – including high redshift sources – are expected to reveal further anomalous scattering that most likely will be the result of confined scattering structures. Finally, the scaling with frequency of angular and pulse broadening is often used to constrain the shape of the wavenumber spectrum of scattering irregularities (e.g., Cordes, Weisberg, & Boriakoff 1985; Fey et al. 1991). Weakening of the frequency dependence by confined plasma structures would be interpreted as a steeper wavenumber spectrum. Thus it is important to assess the role of confined structures in the observations of scattered radio sources.
In §2 we discuss previous treatments of angular and pulse broadening. In §3 we derive a general formalism for scattering that takes into account arbitrary variations of scattering strength transverse to the line of sight. Examples are given in §4. Applications to the Crab pulsar and other Galactic sources are given in §5. Future observations of extragalactic sources and at low frequencies are considered in §6. In §7 we discuss possible implications for the interstellar medium. Finally, in §8, we summarize our results.
## 2 Past Treatments of Angular and Pulse Broadening
It is well known that the shape of a scattered impulse, viewed through a thin (along the line of sight), infinitely-extended (transverse to the line of sight) screen with a circularly symmetric, Gaussian angle distribution, is a one-sided exponential function (e.g., Rickett 1990 and references therein). Thick scattering screens produce slower rise times, while screens containing Kolmogorov irregularities (e.g., Rickett 1990; Lambert & Rickett 1999) produce decays that are slower than exponential. Results along these lines have been presented by Scheuer (1968), Williamson (1972, 1973, 1975), Lee & Jokipii (1975), and Isaacman & Rankin (1977). Williamson (1975) has shown that the pulse broadening function from multiple, discrete screens or from a continuous medium consists of an $`n`$-fold convolution of (one-sided) exponential functions. Williamson’s result applies to the case where the phase structure function is square-law in form. Media with Kolmorogov wavenumber spectra produce different shapes, though the differences are small compared to effects we consider in this paper. The key, implicit assumption in Williamson’s analysis (and essentially all other published results on interstellar pulse broadening; however, see Lyne & Thorne 1975) is that the transverse extent of the scattering screen is arbitrarily large and that the strength of scattering is uniform across the screen. If these assumptions are relaxed, quite different results emerge.
We show that several “anomalous” phenomena occur when scattering structures have finite transverse extents, including:
1. Angular broadening from scattering or refraction that scales anomalously with $`\nu `$. In the case of radio scattering in cold plasmas, anomalous scaling is defined as a significant departure from a $`\nu ^2`$ scaling, which is determined by the microphysics of the plasma.<sup>1</sup><sup>1</sup>1 Scattering from irregularities with a Kolmogorov wavenumber spectrum shows $`\nu ^{11/5}`$ scaling under the circumstance of moderate scattering (cf. Cordes & Lazio 1991). We do not consider this scaling anomalous. When a confined structure contains Kolmogorov fluctuations, the scaling of angular size may be considerably shallower than $`\nu ^{11/5}`$. Departures will always consist of a weaker dependence on frequency.
2. Elongated (or otherwise distorted) images of point sources that are due to scattering but do not scale as $`\nu ^2`$.
3. Multiple imaging by multiple, discrete screens with image intensities influenced by dilution from scattering.
4. Temporal broadening of pulsar pulses which shows a weaker dependence on frequency than $`\nu ^4`$, in accord with the scaling of angular broadening.<sup>2</sup><sup>2</sup>2 A Kolmogorov spectrum can show a $`\nu ^{22/5}`$ scaling for pulse broadening. As with angular broadening, we do not consider this anomalous.
5. Replication of pulses by multiple imaging from an ensemble of screens.
The goal of this paper is to discuss the impact of confined or heavily modulated scattering screens on some of the basic observables, primarily pulse broadening and angular broadening. Therefore we do not explicitly consider refraction from large scale features in the ISM. We do so for two reasons. First, purely diffractive effects are rich enough in variety that we need to isolate the discussion to those effects. Secondly, our results can also be applied to some cases where refraction is important by considering a “renormalized” version of the Kirchoff diffraction integral (e.g., Cordes, Pidwerbetsky & Lovelace 1986). Renormalization of the large scale gradient and phase curvature caused by refraction at a given screen location can be cast as an increased or decreased image intensity and also as a change in the ellipticity of the angular distribution of scattered radiation. In the following we use a probability density, $`f_𝒂`$, to describe the scattering angles $`𝒂`$ from a screen. The renormalization approach allows some of the effects of refraction — image shifts and shapes and intensity changes — to be absorbed into $`f_𝒂`$. Our approach considers only geometrical path length contributions to arrival times and excludes dispersive delays, which are associated with the screen itself. For some situations, dispersive delays can be important. In this paper, the points that we wish to make concern the geometrical phase and a complete discussion that includes dispersive delays would distract our discussion of these points. Consequently, we defer to another paper a complete treatment that includes all contributions to arrival times.
## 3 Probability Densities for Angle of Arrival and Time of Arrival
We derive the image of a scattered point source and the scattered pulse shape of an impulse by calculating, respectively, the probability density functions (PDFs) for the angle of arrival (AOA) and the time of arrival (TOA). In the following, we calculate the effects of scattering while using some of the language and mathematics of ray theory. Williamson (1975) has shown the equivalence of wave and ray optics for some contexts, as have Cordes & Rickett (1998). In this paper, we make the simplifying assumption that the TOA is related geometrically to the AOA. This relationship applies when only the geometrical path length of a given ray path contributes significantly to the TOA. In general, the TOA includes another term related to the integrated refractive index (the integrated electron density in the case of a cold plasma). We ignore the non-geometrical term because there are astrophysical contexts where its contribution is negligible.
Consider a series of diffracting screens at distances $`D_{\mathrm{s}_\mathrm{j}}`$ from a source that is at distance $`D`$ from the observer. Letting $`𝒂_j`$ be the (two-dimensional) scattering angle from the $`j`$th screen, the angular deviation $`𝜽(s)`$ of a ray path and its transverse offset $`𝐱(s)`$ from the direct ray path at distance $`s`$ from the source are
$`𝜽(s)`$ $`=`$ $`𝜽_𝒊+{\displaystyle \underset{j}{}}𝒂_jU(sD_{\mathrm{s}_\mathrm{j}})`$ (1)
$`𝐱(s)`$ $`=`$ $`s𝜽_𝒊+{\displaystyle \underset{j}{}}(sD_{\mathrm{s}_\mathrm{j}})𝒂_jU(sD_{\mathrm{s}_\mathrm{j}}),`$ (2)
where $`U(x)`$ is the unit step function. The first equation relates the observed ray angle ($`𝜽`$) to the the initial ray angle ($`𝜽_𝒊`$) and the scattering angles ($`𝒂_j`$). The relation $`𝐱(D)=0`$ stipulates that rays must reach the observer. We assume all angles are small ($`|𝜽_𝒊|,|𝒂_j|,|𝜽|1`$) , though it is not difficult to extend our results to large angles. Note that $`𝒂_j`$ is a random variable described by a distribution of angles that is determined by diffraction (and, as mentioned above, can also include refraction).
Including only the geometric path-length difference, the corresponding time delay relative to the direct ray path is
$$t=\frac{1}{2c}_0^D𝑑s|𝜽(s)|^2.$$
(3)
The overall time delay also includes dispersive components which, as stated above, we choose to ignore because our main points concern the effects of truncated screens on the arrival times.
To calculate the probability density function (PDF) of the observed angle of arrival, $`𝜽`$, and the time of arrival, $`t`$, we use Dirac delta functions to enforce Eq. 1 and $`𝐱(D)=0`$ for those rays that reach the observer. We use conditional probabilities to include these relations and to integrate over the PDFs for the scattering angles in each screen and over the PDF of $`𝜽_𝒊`$. The result is simple for an isotropic source or, less restrictively, where the PDF of $`𝜽_𝒊`$ is constant over the relevant range of initial ray angles, $`𝜽_𝒊`$, as we assume. Note that the equation $`𝐱(D)=0`$ allows us to eliminate $`𝜽_𝒊`$ as an independent variable.
It is standard to assume the scattering strength is invariant across a scattering screen. Here we specify a more general description. Consider each screen to scatter or refract rays according to a PDF $`f_{𝒂}^{}{}_{j}{}^{}`$ whose width varies arbitrarily across the screen. Accordingly we write each screen’s PDF as $`f_{𝒂}^{}{}_{j}{}^{}(𝒂_j;𝐱(D_{\mathrm{s}_\mathrm{j}}))`$, where $`𝐱(D_{\mathrm{s}_\mathrm{j}})`$ is Eq. 2 evaluated at the location of each screen, $`s=D_{\mathrm{s}_\mathrm{j}}`$.
Let $`Q`$ be an observed quantity such as the AOA or TOA and let $`Q^{}(𝜽_𝒊,𝒂_j,D_{\mathrm{s}_\mathrm{j}},D)`$ be its value given $`𝜽_𝒊`$, $`𝒂_j`$ and $`D_{\mathrm{s}_\mathrm{j}}`$ (and it is implicit that we consider all $`j`$ when there are multiple screens). The PDF of $`Q`$ is
$$f_Q(Q)=\frac{{\displaystyle \underset{j}{}}{\displaystyle 𝑑𝒂_jf_{𝒂}^{}{}_{j}{}^{}(𝒂_j;𝐱(D_{\mathrm{s}_\mathrm{j}}))𝑑𝜽_𝒊f_{𝜽_𝒊}(𝜽_𝒊)\delta (𝐱(D))\delta (QQ^{}(𝜽_𝒊,𝒂_j,D_{\mathrm{s}_\mathrm{j}},D))}}{{\displaystyle \underset{j}{}}{\displaystyle }d𝒂_jf_{𝒂}^{}{}_{j}{}^{}(𝒂_j;𝐱(D_{\mathrm{s}_\mathrm{j}})){\displaystyle }d𝜽_𝒊f_{𝜽_𝒊}(𝜽_𝒊)\delta (𝐱(D)),}$$
(4)
where the numerator is the joint PDF of Q and $`𝐱(D)=0`$ (that rays reach the observer), while the denominator is the PDF that rays reach the observer. Using Eq. 2, we transform $`\delta (𝐱(D))`$ to a delta function involving $`𝜽_𝒊`$, perform the integral over $`𝜽_𝒊`$, and assume that the PDF for $`𝜽_𝒊`$, $`f_{𝜽_𝒊}`$, is constant for angles of interest. Then $`Q^{}`$ becomes independent of $`𝜽_𝒊`$ and the PDF of $`Q`$ becomes
$$f_Q(Q)=\frac{{\displaystyle \underset{j}{}}{\displaystyle 𝑑𝒂_jf_{𝒂}^{}{}_{j}{}^{}(𝒂_j;𝐱(D_{\mathrm{s}_\mathrm{j}}))\delta (QQ^{}(𝒂_j,D_{\mathrm{s}_\mathrm{j}},D))}}{{\displaystyle \underset{j}{}}{\displaystyle 𝑑𝒂_jf_{𝒂}^{}{}_{j}{}^{}(𝒂_j;𝐱(D_{\mathrm{s}_\mathrm{j}}))}}.$$
(5)
### 3.1 The General Single Screen (N=1)
For the simple case of a single scattering screen, Eq. 5 becomes
$`f_Q(Q)`$ $`=`$ $`{\displaystyle \frac{{\displaystyle 𝑑𝒂f_𝒂(𝒂;𝒂D_s^{})\delta (QQ^{}(𝒂,D_s,D))}}{{\displaystyle 𝑑𝒂f_𝒂(𝒂;𝒂D_s^{})}}},`$ (6)
where
$`D_s^{}`$ $`=`$ $`D_s(1D_s/D).`$ (7)
For the angle of arrival, $`Q=𝜽`$, $`Q^{}=𝒂(D_s/D)`$ and
$$f_𝜽(𝜽)=\left(\frac{D}{D_s}\right)^2\frac{f_𝒂({\displaystyle \frac{D}{D_s}}𝜽;𝜽(DD_s))}{{\displaystyle 𝑑𝒂f_𝒂(𝒂;𝒂D_s^{})}}.$$
(8)
For the TOA, $`Q=t`$ and $`Q^{}=D_s(1D_s/D)|𝒂|^2/2c`$ so only the magnitude, $`|𝒂|`$, is constrained. Transformation of $`\delta (QQ^{})`$ to $`\delta (|𝒂|a_t)`$ yields
$`f_t(t)`$ $`=`$ $`\left({\displaystyle \frac{c}{D_s^{}}}\right){\displaystyle \frac{{\displaystyle _0^{2\pi }}𝑑\varphi f_𝒂(a_t\widehat{𝐚}_\varphi ;a_t\widehat{𝐚}_\varphi D_s^{})}{{\displaystyle 𝑑𝒂f_𝒂(𝒂;𝒂D_s^{})}}},`$ (9)
where $`\widehat{𝐚}_\varphi `$ is a unit vector,
$`\widehat{𝐚}_\varphi `$ $`=`$ $`\mathrm{cos}\varphi \widehat{𝐱}+\mathrm{sin}\varphi \widehat{𝐲},`$ (10)
and
$`a_t`$ $`=`$ $`\left({\displaystyle \frac{2ct}{D_s^{}}}\right)^{1/2}.`$ (11)
The flux density of a source is conserved only for an infinite screen with homogeneous statistics because only in that case is as much flux scattered toward the observer as is scattered away. We define the flux dilution factor as the ratio of the probability that rays reach the observer to the probability for a uniform, infinite screen:
$`\eta _F={\displaystyle 𝑑𝒂f_𝒂(𝒂;𝒂D_s^{})},`$ (12)
equal to the denominator of Eq. 6. For a uniform, infinite screen, $`\eta _F=1`$. In general, $`\eta _F1`$.
We illustrate these expressions by considering specific cases.
#### 3.1.1 Infinitely Wide Screen with Homogeneous Statistics
For an infinite screen with homogeneous statistics, the denominator of Eq. 6 is unity. Specializing to circularly symmetric $`f_𝒂`$, we find the normalized 1D PDF for the magnitude $`\theta |𝜽|`$:
$$f_\theta (\theta )=2\pi \theta \left(\frac{D}{D_s}\right)^2f_𝒂(D𝜽/D_s).$$
(13)
If $`f_𝒂`$ is a Gaussian function with rms angle $`\sigma _a`$ in each coordinate direction, then
$`f_\theta (\theta )`$ $``$ $`\sigma _{\theta }^{}{}_{}{}^{2}\theta e^{\theta ^2/2\sigma _{\theta }^{}{}_{}{}^{2}}`$ (14)
$`\sigma _\theta `$ $``$ $`(D_s/D)\sigma _a`$ (15)
$`f_t(t)`$ $`=`$ $`\tau _0^1e^{t/\tau _0}U(t)`$ (16)
$`\tau _0`$ $`=`$ $`D_s^{}\sigma _a^2/c.`$ (17)
The 1D rms angle
$$\theta _{\mathrm{rms}}\frac{1}{\sqrt{2}}|𝜽|^2^{1/2}$$
(18)
characterizes the observed range of angles and in this case is identical to $`\sigma _\theta `$, which is the scaled range of scattering angles. In general, $`\theta _{\mathrm{rms}}\sigma _\theta `$.
If $`\sigma _a\nu ^2`$ as for a plasma, then the AOA PDF has a scale $`\sigma _\theta \nu ^2`$ and the TOA PDF has scale $`\tau _0\nu ^4`$. These scaling laws for observable quantities rely on the assumption that the screen is infinitely wide.
#### 3.1.2 Circular Screen with Finite Radius
Consider a circular screen with radius $`X_{\mathrm{max}}`$ centered on the line of sight. Now the PDF for $`\theta `$ is truncated for $`\theta >\theta _{\mathrm{max}}X_{\mathrm{max}}/(DD_s)`$. Again adopting circularly symmetric, Gaussian statistics for $`𝒂`$, we find
$`f_\theta (\theta )`$ $`=`$ $`[\sigma _{\theta }^{}{}_{}{}^{2}(1e^\zeta )]^1\theta e^{\theta ^2/2\sigma _{\theta }^{}{}_{}{}^{2}}U(\theta _{\mathrm{max}}\theta ),`$ (19)
$`\zeta `$ $``$ $`{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{\theta _{\mathrm{max}}}{\sigma _\theta }}\right)^2,`$ (20)
where the unit step function enforces truncation of the PDF for $`\theta >\theta _{\mathrm{max}}`$. If the rms scattering angle is small, $`\sigma _aX_{\mathrm{max}}D/D_s(DD_s)`$, the scaling of the observed size with frequency is according to that of $`\sigma _a`$. For larger scattering angles, the physical size of the screen becomes important. To see this, we calculate the rms angular size, which is, for the circularly symmetric Gaussian and a frequency scaling $`\sigma _a=\sigma _{a}^{}{}_{0}{}^{}(\nu /\nu _0)^2`$,
$$\theta _{\mathrm{rms}}=\sigma _{a}^{}{}_{0}{}^{}\left(\frac{D_s}{D}\right)\left(\frac{\nu }{\nu _0}\right)^2\left[\frac{1(1+\zeta )e^\zeta }{1e^\zeta }\right]^{1/2}.$$
(21)
The frequency scaling is $`\zeta \nu ^4`$. At large $`\nu `$, $`\zeta 1`$ and $`\theta _{\mathrm{rms}}\nu ^2`$. At small $`\nu `$, $`\zeta 0`$ and $`\theta _{\mathrm{rms}}`$ becomes frequency independent.
The TOA has a PDF and mean value
$`f_t(t)`$ $`=`$ $`{\displaystyle \frac{e^{t/\tau _0}}{\tau _0\left(1e^\zeta \right)}}U(t_{\mathrm{max}}t)`$ (22)
$`t`$ $`=`$ $`\tau _0\left[1\zeta \left({\displaystyle \frac{e^\zeta }{1e^\zeta }}\right)\right]`$ (23)
$`t_{\mathrm{max}}`$ $`=`$ $`\zeta \tau _0.`$ (24)
For a wide screen, $`\zeta \mathrm{}`$, $`t_{\mathrm{max}}\mathrm{}`$ and $`t=\tau _0`$, as before. However, a narrow screen with $`\zeta 1`$ gives $`t\zeta \tau _0`$. The TOA scaling is $`t\nu ^4`$ for wide screens but becomes frequency independent for very narrow screens.
The observed flux of a source viewed through the truncated screen is attenuated by the scattering. The flux dilution factor is
$$\eta _\mathrm{F}=1e^\zeta .$$
(25)
For $`\zeta 1`$, $`\eta _\mathrm{F}\zeta `$ while $`\zeta 1`$ yields $`\eta _\mathrm{F}=1`$. When the angular diameter and pulse broadening of a source are observed to have anomalous frequency dependence, i.e., when scattering is dominated by a truncated screen, the flux is diminished. Inspection of Eq. 21 and Eq. 25 indicates that there can be a correlation of rms angular size and flux density.
### 3.2 A Screen with Arbitrary Variations of Scattering Strength
Above, we considered screens with extreme variations of scattering strength: discontinuous or truncated to zero. Here we consider other cases that may have relevance to sources that are viewed through scattering regions with structure on scales $`(DD_s)\theta `$. As in Eq. 5, we specify the scattering angle $`𝒂_j`$ from the $`j`$th screen by a PDF that depends on location along the screen, $`𝐱(D_{\mathrm{s}_\mathrm{j}})`$: $`f_{𝒂}^{}{}_{j}{}^{}(𝒂_j;𝐱(D_{\mathrm{s}_\mathrm{j}}))`$. For simplicity, we drop the $`j`$ subscripts and discuss scattering from a single screen. Also, for ease of discussion, we consider $`f_𝒂`$ to be a 2D Gaussian with angular variance $`𝝈_𝒂(𝐱)`$ that varies across the screen.
In some circumstances, the angle-of-arrival distribution $`f_𝜽`$ is determined purely by the shape and width of $`f_𝒂`$, while in others it is determined by the variations in $`𝝈_𝒂`$ across the screen. If $`𝝈_𝒂`$ is constant across the screen, $`f_𝜽`$ is a scaled version of $`f_𝒂`$ and the frequency scaling of the observed angular size is identical to that of the scattering angles.
However, if $`𝝈_𝒂`$ varies across the screen, the observed angular size may reflect variations of $`𝝈_𝒂`$ in addition to or rather than the width of $`f_𝒂`$ itself. Let $`\mathrm{}_𝒂`$ be the characteristic length scale on which $`𝝈_𝒂`$ varies across the screen. We compare this with the observed angular diameter projected back to the screen, yielding a length scale $`\mathrm{}_{}(DD_s)\theta _{\mathrm{rms}}`$. We consider three cases:
1. If $`\mathrm{}_𝒂\mathrm{}_{}`$ and variations in $`𝝈_𝒂`$ are statistically homogeneous, the variations average out. The scaling with frequency of the observed AOA width is identical to that of $`𝒂`$, which is due to the microphysics. Note that our examples of truncated screens in previous sections satisfy the inequality but are not statistically homogeneous.
2. If $`\mathrm{}_𝒂\mathrm{}_{}`$, then $`𝝈_𝒂`$ constant over the part of the screen sampled and the angle-of-arrival PDF is determined by $`f_𝒂`$. The frequency scaling is again determined by the microphysics. Depending on how large $`\mathrm{}_𝒂`$ is, eventually time variations in the image are expected on a time scale $`t_a\mathrm{}_𝒂/V_{\mathrm{eff}}`$, where $`V_{\mathrm{eff}}`$ is an effective velocity determined by the velocities of the source, medium and observer (e.g., Cordes & Rickett 1998).
3. If $`\mathrm{}_𝒂\mathrm{}_{}`$, the AOA PDF is determined by a combination of $`f_𝒂`$ and the spatial variation of $`𝝈_𝒂`$. The truncated screen of §3.1.2 is an extreme example of this case. The frequency scaling is likely to depart significantly from that of the microphysics.
## 4 Examples
### 4.1 Illustration of Anomalous Frequency Scaling
Figure 1 shows the “image” $`f_𝜽`$, the pulse broadening function $`f_t`$, and the scaling with frequency of the pulse broadening time for two cases: (a) an infinite screen (bold solid lines) and (b) a truncated circular screen centered on the direct line of sight (light and dashed lines). The underlying scattering function ($`f_𝒂`$) is a circular Gaussian PDF and the rms scattering angle scales as $`\nu ^2`$. For the infinite screen, the pulse broadening scales as $`\nu ^4`$, as expected. However, the truncated screen yields truncated images and truncated pulse broadening functions if the rms scattering angle is large enough that rays from the screen’s edges reach the observer. Thus, truncation occurs at low frequencies and not at high frequencies. For the example given, the break frequency $`0.5`$ GHz. Actual break frequencies will depend on particular sizes and scattering strengths of screens.
As a second illustration of anomalous frequency scaling, we calculate the pulse broadening for a pulse scattered by a two component screen. The first, circular component with radius $`X_1`$ is centered on the line of sight and scatters radiation much more strongly than the remainder of the screen, which is of infinite extent. The distributions of AOA and TOA follow from the master equations, Eq. 8,9. The rms angular size, the mean TOA, and related quantities are:
$`\theta _{\mathrm{rms}}`$ $`=`$ $`\sqrt{2}\sigma _{\mathrm{d}}^{}{}_{1}{}^{}\left({\displaystyle \frac{D_s}{D}}\right)\left[{\displaystyle \frac{1+(\tau _2/\tau _1)(1+\zeta _2)e^{\zeta _2}(1+\zeta _1)e^{\zeta _1}}{1+e^{\zeta _2}e^{\zeta _1}}}\right]^{1/2}`$ (26)
$`t`$ $`=`$ $`\tau _1\left[{\displaystyle \frac{1+(\zeta _1+\tau _2/\tau _1)e^{\zeta _2}(1+\zeta _1)e^{\zeta _1}}{1+e^{\zeta _2}e^{\zeta _1}}}\right]`$ (27)
$`\tau _{1,2}`$ $`=`$ $`c^1D_s^{}\sigma _{\mathrm{d}}^{}{}_{1,2}{}^{2}`$ (28)
$`\zeta _1`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left[{\displaystyle \frac{X_1}{D_s^{}\sigma _{\mathrm{d}}^{}{}_{1}{}^{}}}\right]^2`$ (29)
$`\zeta _2`$ $`=`$ $`{\displaystyle \frac{\tau _1}{\tau _2}}\zeta _1,`$ (30)
where $`\sigma _{\mathrm{d}}^{}{}_{1,2}{}^{}`$ is the rms scattering angle produced by each screen.
Figure 2 shows $`\theta _{\mathrm{rms}}`$ plotted against frequency for different ratios, $`\sigma _{\mathrm{d}}^{}{}_{1}{}^{}/\sigma _{\mathrm{d}}^{}{}_{2}{}^{}`$ and assuming that $`\sigma _{\mathrm{d}}^{}{}_{1,2}{}^{}\nu ^2`$. Note that we vary $`\sigma _{\mathrm{d}}^{}{}_{1}{}^{}/\sigma _{\mathrm{d}}^{}{}_{2}{}^{}`$ while keeping $`\sigma _{\mathrm{d}}^{}{}_{1}{}^{}`$ constant. The figure demonstrates how the stronger central component dominates the apparent source size at high frequencies and the weaker, distributed component dominates at low frequencies. At intermediate frequencies, there is a plateau where the angular size is nearly independent of frequency.
Figure 3 shows a similar plot, now for the mean pulse broadening time, $`t`$, plotted against frequency for different ratios, $`\tau _1/\tau _2`$ and assuming that $`\sigma _{\mathrm{d}}^{}{}_{1,2}{}^{}\nu ^2`$. The roles of the central and distributed components are the same as for the angular scattering shown in Figure 2.
### 4.2 Scattering from Filaments
Figure 4 shows scattering from a filament located along the direct ray path for three values of rms scattering angle in the filament. Small rms scattering yields a circular image and an exponential pulse broadening function. For sufficiently large scattering angles, the image becomes elongated and tends toward a $`t^{1/2}e^t`$ pulse broadening function
Figure 5 shows scattering from a filament at different locations relative to the direct ray path but for identical rms scattering angles in the filament. When the filament is near enough to the image center, the pulse broadening function is bimodal. When the filament is far, the pulse broadening function is dominated by the much weaker scattering from outside the filament. Clearly, a superposition of filaments near the direct ray would produce a multiplicity of pulses.
Figure 6 shows scattering from an ensemble of filaments at different frequencies. As can be seen, the pulse broadening function shows multiple peaks that align at different frequencies. The number of filaments that are ‘lit up’ by the scattering decreases in going to higher frequency. The scattering screen consists of a very strong, extended component which has embedded “gaps” where the scattering is weaker but still strong enough to scatter radiation toward the observer. The extended component scatters radiation to such wide angles that it produces negligible contributions to the pulse broadening function and to the image. Thus the strongest contributions to measured quantities come from the filamentary gaps. We will explore this result further in a separate paper to discuss giant pulses from the Crab pulsar.
## 5 Application to Galactic Scattering
Galactic sources show a wide range of scattering levels, indicative of the concentration of intense scattering into a Population I type Galactic distribution (Cordes, Weisberg, & Boriakoff 1985 ; Taylor & Cordes (1993)). Here we discuss particular objects whose scattering may be interpreted in the context of this paper’s formalism. Our discussion is brief. We defer to separate articles any detailed treatment on particular sources.
### 5.1 The Crab Pulsar
The Crab pulsar shows enhanced pulse broadening from nebular material that has been recognized since shortly after the pulsar’s discovery in 1968 (e.g., Vandenburg 1976 and references therein). The nebular contribution is highly episodic, with dramatic increases of the pulse broadening time by a factor of 100 (Lyne & Thorne 1975; Isaacman & Rankin 1977). Recently, multiple images have been inferred from the presence of echoes of the pulse shape (Graham Smith & Lyne (2000); Backer (2000); Backer, Wong, & Valanju 2000).
Giant pulses from the Crab pulsar show additional evidence for nebular contributions to the scattering that are probably from discrete filaments. At relatively high frequencies (1.4 to 5 GHz), giant pulses show multiple pulse components that tend to have exponential scattering tails with time constants that often differ, even within the same spin period of the pulsar (Hankins & Moffett (1998); Sallmen et al. (1999); Hankins (2000)). Also, the widths of the pulse components appear to scale less strongly with frequency than $`\nu ^4`$. These characteristics suggest consistency with the overall picture developed in this paper. A detailed analysis of the Crab pulsar’s pulses is deferred to another paper.
### 5.2 NGC 6334B
The largest angular broadening measured is for the extragalactic source NGC 6334B viewed through the H II complex NGC 6334 (Trotter, Moran, & Rodríguez 1998 and references therein), $`\theta _d3\mathrm{}`$ at 1.4 GHz. The image’s position angle rotates in going from low to high frequency, and the image axial ratio may increase from 1.2 at 1.4 and 5 GHz to 1.5 at 15 GHz. Trotter et al. (1998) interpret this variation as signifying an outer scale for the wavenumber spectrum of *anisotropically-scattering* density irregularities, $`\mathrm{}_{\mathrm{out}}10^{16}`$ cm. This proposed outer scale is comparable to the thickness of the H II shell in this region (Rodríguez, Canto, & Moran 1988; Kahn & Breitschwerdt 1990). An alternate possibility is that the density irregularities scatter isotropically, but that the anisotropic images reflect density irregularities that are confined to the H II shell.
If the latter were the case, we would expect that the axial ratio would increase as a function of decreasing frequency, opposite to what is observed. At lower frequencies, the size of the minor axis of the scattering diameter would be constrained by the width of the H II shell while the size of the major axis would be essentially unconstrained, unless the scattering is so intense that the size of the major axis is also limited by the scale of the H II shell. For NGC 6334B, the axial ratio appears to be constant with frequency or increasing with increasing frequency. The frequency behavior of the axial ratio indicates that any relevant length scales in the H II shell must be smaller than $`10^{14}`$ cm (the smallest length scale probed by the highest frequency observations) or larger than $`10^{17}`$ cm (the largest length scale probed by the lowest frequency observations).
### 5.3 Cygnus X-3
The compact source Cyg X-3 is heavily scattered (0$`\stackrel{}{\mathrm{.}}`$5 at 1 GHz) and has an anisotropic image (axial ratio $`1.2`$ and increasing with increasing frequency) whose position angle varies with frequency (Wilkinson, Narayan, & Spencer 1994; Molnar et al. 1995). The variation of position angle has been interpreted by Wilkinson et al. (1994) as due to a changing orientation of (anisotropic) diffracting irregularities on a length scale of order $`\theta (DD_s)0.01`$ pc. In their picture, the image asymmetry is due to anisotropic diffracting irregularities, and the orientation change with frequency is attributed to the spatial variation of those irregularities.
An alternative explanation is that the diffracting irregularities are isotropic and that the image anisotropy reflects spatial variations of the strength of the diffracting irregularities on the scale $`\theta (DD_s)`$. Molnar et al. (1995) have proposed that the H II region DR 11 is responsible for the bulk of the scattering along this line of sight. A key difficulty with this explanation is that the observed axial ratio increases with increasing frequency. This is inconsistent with the notion of scattering from a single filament but may be consistent with scattering from a group of filaments which are individually smaller than $`10^{16}`$ cm.
### 5.4 Sgr A\* & Galactic-Center OH/IR Masers
Galactic center sources show large scattering diameters ($`1\mathrm{}`$ at 1 GHz) and significant image asymmetries that vary across the roughly 30′ size of the scattering region (Lazio & Cordes 1998). The major axis of the image of Sgr A\* shows a $`\nu ^2`$ frequency dependence from 1.4 to 22 GHz (Yusef-Zadeh et al. 1994; Lo et al. 1998), and the image itself displays no change in its major axis, axial ratio, or orientation on time scales as long as a decade (Lo et al. 1998; Marcaide et al. 1999).
Deviations from a $`\nu ^2`$ dependence for the major axis, minor axis, or both are claimed at a variety of frequencies from as low as 43 GHz (Lo et al. 1998) to as high as 215 GHz (Krichbaum et al. 1998). These deviations from the $`\nu ^2`$ dependence observed at lower frequencies are commonly interpreted as an effect of the intrinsic source diameter becoming important at the various frequencies.
The deviations could also be symptomatic of the size scale of the scattering medium. If this is the case, then the frequencies at which the breaks occur correspond to spatial scales $`\mathrm{}5`$ AU (for a break occurring at 43 GHz) to as small as 0.17 AU (at 215 GHz). These spatial scales are comparable to the outer scale inferred by Lazio & Cordes (1998), on the basis of a comparison of scattering strength and thermal free-free emission. However, the velocities in the Galactic center ($`50`$ km s<sup>-1</sup> suggest that variations in the orientation of the image of Sgr A\* should be seen on timescales of order $`0.1\mathrm{yr}(\mathrm{}/1\mathrm{AU})(v/50\text{km s}\text{-1})^1`$. As mentioned above, these are not seen. This implies either that all striations in the medium are oriented in the same direction or that the intrinsic source size is in fact important at high frequencies.
The lack of variations in the image orientation indicates that there are not likely to be striations or other structure in the scattering screen on scales smaller than about 1 AU. Variations do occur, though, on much larger scales, of order 15′ corresponding to spatial scales of roughly 25 pc, the separations between Sgr A\*, the various OH masers, and other scattered sources (Lazio et al. 1999). Effects from the spatially-limited scattering described in this paper are unlikely to be seen with the current census of Galactic center sources. The narrowband nature of OH masers means that the frequency dependence of their scattering diameters cannot be measured, and Sgr A\* itself is obscured below 1 GHz due to free-free absorption by Sgr A. Detection of additional radio transients (e.g., Zhao et al. 1992) or radio pulsars (Cordes & Lazio 1997) may allow such effects to be detected at frequencies below 1 GHz.
### 5.5 Extreme Scattering Events
“Extreme scattering events” (ESE’s) are events identified in the light curves of several AGN’s (Fiedler et al. 1987; 1994) and two pulsars (Cognard et al. 1993; Maitia, Lestrade, & Cognard 1999). They are roughly consistent with refractive defocusing and caustic formation from discrete, small scale plasma structures (Fiedler et al. 1987; Romani et al. 1987; Clegg et al. 1998). However, alternative explanations invoke the outer ionized regions of predominantly neutral, primordial HI clouds (Walker & Wardle 1998) or distributed fluctuations much like those that account for the diffractive scintillations of pulsars (Fiedler et al. 1994). The fundamental difference between these models is the implied gas pressure. As discussed further below, a discrete structure is necessarily overpressured compared to the general ISM, so any such structures must exist either transiently or in regions of small volume filling factor that can support such pressure.
If ESEs result from discrete ionized structures, then effects described in this paper should be present in the scattered image and pulse shape (for pulsars). This notion is little explored because few ESEs have been identified and, given that most are seen from AGN’s, intrinsic source size effects can also diminish the appearance of diffraction effects. VLBI observations of the source 1741$``$038 undergoing an ESE have shown no indication of a truncated image (Lazio et al. 2000). However, those observations were at relatively high frequencies ($`1.7`$ GHz) and had limited dynamic range. Pulse timing observations of PSR B1937$`+`$21 (Cognard et al. 1997) show no change in the pulse shape, though, again these observations were obtained at 1.4 GHz. Future observations of a source undergoing an ESE at lower frequencies (e.g., 0.33, 0.41, or 0.61 GHz) would place much more stringent constraints on the notion that ESEs arise from discrete ionized structures.
## 6 Future Observations
### 6.1 Application to Extragalactic Scattering
Scattering from extragalactic plasma can arise from the distributed intergalactic medium (IGM), most of which is expected to be ionized, from intervening Ly$`\alpha `$ clouds, and from intervening galaxies. Of greatest relevance to this paper are the last two cases and, of these, intervening galaxies are likely to be the more important because of their greater column densities. A face-on galaxy like the Milky way will scatter radiation from a background source into an apparent size of at least $`\theta _d1(\nu /0.33\mathrm{GHz})^{2.2}`$ mas (Cordes & Lazio 1991; Taylor & Cordes (1993)). Scattering by H II regions yields even larger angles, so some background sources, albeit at low-probability alignments, will display images that reflect the sizes of H II regions and, in some instances, spiral arms that contain them.
Scattering from an edge-on galaxy will be about $`10^3`$$`10^4`$ times larger, or 1″–10″ at 0.3 GHz.<sup>3</sup><sup>3</sup>3 This large increase in the scattering diameter occurs because of the presence of enhanced scattering regions, such as those described in §5. In the Milky Way Galaxy, the approximate mean free path between enhanced scattering regions is 8 kpc (Taylor & Cordes (1993)). A line of sight through the disk of a galaxy like the Milky Way is quite likely to encounter one or more enhanced scattering regions leading to the large increase cited. The lateral scale is $`\theta _dD15`$$`150\times D_{3000}`$ kpc for $`D=3000D_{3000}`$ Mpc. Thus, near edge-on galaxies will produce scattered images that, in part, display the shapes of the galaxies. At even lower frequencies, scattering diameters from Ly$`\alpha `$ clouds and galaxies with significantly smaller scattering strength will produce similar effects. Imaging radio observations at $`0.1`$ GHz will thus probe intergalactic structures.
We defer to another paper a thorough discussion of intergalactic scattering, taking into account cosmological expansion and evolution. Scattering may be able to probe the intergalactic medium at redshifts near the reionization epoch.
### 6.2 Low-Frequency Galactic Observations
The (nominally) strong frequency dependence of interstellar scattering observables suggest that the anomalous scattering described here will most likely occur at low frequencies. High-resolution, low-frequency instruments such as the Giant Metrewave Radio Telescope (GMRT, Ananthakrishnan (1995)) and the proposed Low Frequency Array (LOFAR, Kassim et al. (2000)) and the low-frequency Square Kilometer Array (SKA, Butcher (2000)) have or will have sub-arcminute resolution at frequencies below 150 MHz. Consequently, they may detect anomalous scattering along lines of sight less heavily scattered than those described in §5. Here we consider relevant lines of sight and frequencies for which anomalous scattering is a possibility.
The relevant length scale in regions of less intense scattering may be the outer scale of the density fluctuation spectrum $`\mathrm{}_0`$. (This may also be the relevant length scale in intense regions, though its value could be quite different and potentially much smaller.) Near the Sun (within $`1`$ kpc), $`l_01`$ pc (Armstrong, Rickett, & Spangler 1995). It is unlikely that scattering diameters will probe this spatial scale (i.e., $`\theta _dD\mathrm{}_0`$) unless $`\nu <10`$ MHz. As the ionosphere becomes increasingly opaque at frequencies $`\nu <10`$ MHz, ground-based interferometric arrays will likely not be affected by anomalous scattering in the solar neighborhood.
Toward the inner Galaxy, Galactic latitudes $`|\mathrm{}|<50\mathrm{°}`$, stronger scattering than that in the solar neighborhood (but still weaker than the intense scattering described in §5) will obtain. In this case, observations at meter wavelengths may display anomalous scattering.
A competing effect for detecting anomalous scattering is free-free absorption. The density fluctuations responsible for interstellar scattering also contribute to free-free absorption. Sources seen along heavily scattered lines of sight at shorter wavelengths may be free-free absorbed at longer wavelengths. For instance, free-free absorption renders the Galactic center increasingly opaque for frequencies $`\nu <1`$ GHz (Anantharamaiah et al. (1991)).
## 7 Implications for the Interstellar Medium
As alluded to before, the existence of compact, turbulence-containing ionized structures is directly related to their longevity and rarity, or filling factor, in the Galaxy. Except for chance fluctuations from distributed turbulence, many observed phenomena suggest the existence of compact structures with densities that imply they are overpressured compared to most of the ISM. This is not overly surprising because the ISM is highly dynamic and is in pressure equilibrium only in some average sense. It is not known which kinds of locales (H II regions, supernova shocks, etc.) provide the largest scattering strengths. We suggest, simply, that the observable effects described in this paper might be used to better probe the physical sizes of regions that produce the largest levels of scattering.
Striations in interstellar gas densities on sub-parsec scales are most likely associated with magnetic fields. On diffraction scales $`10^{11}`$ cm, a compelling idea is that turbulence is essentially two dimensional and that irregularities are elongated along the field lines (Higdon 1984, 1986; Goldreich & Sridhar 1995). Larger scale filaments, such as those seen near the Galactic center perpendicular to the plane of the Galaxy (Yusef-Zadeh & Morris (1987)), are also thought to be along the local field direction. If anisotropically diffracting irregularities are contained in screens that are themselves elongated in the same direction, it may be difficult to separate (and thus identify) the two possible contributions to image elongation.
The frequency dependence of anomalous scattering may offer a means for identifying the cause of image elongation for a particular source if it is heavily scattered by a single (or few) filament(s) (cf. Figure 2 and §5). If the image elongation arises from anisotropic scattering by small-scale density irregularities, increased spatial resolution of the scattering material (e.g., by making observations at higher frequencies) may yield increased axial ratios as less spatial averaging is done over the small-scale irregularities (e.g., as argued by Wilkinson et al. (1994) for the image of Cyg X-3). This change in axial ratio is expected only if there are spatial variations of the orientation of scattering turbules. Conversely, if image elongation is produced by the boundaries of filaments, a different frequency dependence may be seen. A more complicated frequency dependence may result if the scattering results from a number of smaller filaments (e.g., Figure 6).
The wavenumber spectrum of electron density irregularities is often constrained by the scaling law of angular size and pulse broadening (and its reciprocal, the scintillation bandwidth) with frequency. For moderate scattering, where the dominant length scales are between the inner and outer scales, the pulse broadening time scales as $`\nu ^x`$ with $`x=2\beta /(\beta 2)`$, where $`\beta `$ is the exponent of the three dimensional, isotropic wavenumber spectrum (CWB85; Rickett 1990). For a Kolmogorov spectrum, $`\beta =11/3`$ and $`x=4.4`$. If irregularities are isotropic, so that $`\beta =11/3`$, but the medium is confined in the transverse direction, the actual value of $`x`$ is lessened. The value of $`\beta `$ inferred would be greater than $`11/3`$ in this instance. A similar trend occurs when angular broadening is used to infer $`\beta `$. It is not clear which, if any, of the published constraints on $`\beta `$ are affected by the influence of scattering-region confinement. A detailed study of the wavenumber spectrum is deferred to another paper.
## 8 Summary
In this paper we have shown that radio scattering observables such as image shapes and pulse broadening functions can be strongly influenced by structure in the scattering medium on length scales substantially larger than those that cause the scattering. As such, careful multi-frequency observations can be used to constrain properties of the interstellar medium on scales that are typically $`1`$–10 AU. Intergalactic scattering has not been identified but is certainly expected from intervening spiral galaxies, probably expected for some Lyman-$`\alpha `$ clouds, and may occur from distributed ionized gas. For intergalactic scattering, relevant length scales can be comparable to the sizes of galaxies. A low-frequency VLBI survey of extragalactic sources may thus probe the level of scattering in other galaxies and in the general intergalactic medium. It is also expected that scattering of radiation from gamma-ray burst afterglows will be influenced in some cases by intervening ionized gas in the IGM as well as in the Milky Way’s ISM. These issues will be explored in separate articles.
We thank B. Rickett for helpful discussions. This work was supported by NSF Grant AST 9819931 to Cornell University. Basic research in radio astronomy at the NRL is supported by the Office of Naval Research.
|
warning/0005/hep-ph0005301.html
|
ar5iv
|
text
|
# TTP/00–11hep-ph/0005301May 2000 Subleading Sudakov Logarithms in Electroweak Processes Talk presented by J.H.Kühn
## 1 Introduction
Four fermion processes are generally considered as benchmark processes at high energy colliders, with electron positron annihilation into muon or quark pairs at LEP and the Drell Yan process at hadron colliders as characteristic examples. Within the presently accessible energy region, typically up to 200 GeV, radiative corrections are dominated by the shift in the $`W`$ and $`Z`$ masses as parametrized by the $`\rho `$ parameter and by the running of the coupling constant. Vertex corrections and box diagrams involving gauge bosons are generally of minor importance. In the TeV region, accessible at future colliders like the LHC or TESLA, this picture changes drastically. A new class of effects starts to become relevant and rapidly dominant which are generally denoted as double logarithmic corrections and which were first observed by Sudakov in the context of quantum electrodynamics for reactions with a tight cut on the radiated energy of the photons. For electroweak interactions large negative corrections arise from the exchange of gauge bosons which remain uncompensated if one restricts the analysis to exclusive final states, consisting e. g. of a fermion antifermion pair only. The discussion of double logarithmic corrections is fairly straightforward for a theory with massive gauge bosons only. An important complication arises from the presence of massless photons in the final state. Events with soft and hard photon radiation are normally included in the sample – whence a “semiinclusive” definition of the cross section is closest to the the actual experimental analysis.
One loop corrections to the four fermion process are available since long (see e. g. ). In this case the separation of photonic and weak corrections is still possible and, employing axial gauge, the leading logarithms can be trivially attributed to fermion self energy diagrams with virtual $`W`$ or $`Z`$ boson emission .
Results for the leading logarithmic corrections in higher orders have been presented in (for a more recent discussion see also ). The first three results disagree, a consequence of different requirements on the inclusion or exclusion of hard photon emmission and a different treatment of the virtual corrections. In it was demonstrated that the approach developed in , where axial gauge was adopted and both real and virtual photon emmission neglected, is numerically nearly identical to the results of where virtual photon corrections and real photon emmission are included — a consequence of the smallness of the weak mixing angle, $`\mathrm{sin}^2\theta _W=0.23`$. Furthermore, an analysis of subleading logarithms for the form factor, for the four fermion scattering in a spontaneously broken $`SU(2)`$ gauge theory and, last not least, for the Standard Model with $`W`$, $`Z`$ and the massless photon was performed in .
Before presenting a brief review of the techniques and some results from let us emphasize the importance of subleading corrections. In the energy region relevant at LHC or TESLA, i.e. around 1 to 2 TEV, this is already evident from the one loop vertex correction. In the timelike region
$$|F_B+\delta F|^2F_B(1+2\frac{\alpha _W}{\pi }\rho (s/M^2)),$$
(1)
$$\rho (u)=\frac{1}{4}\mathrm{ln}^2u+\frac{3}{4}\mathrm{ln}u\frac{\pi ^2}{12}+\frac{7}{8}.$$
For characteristic values of the coupling, energy and mass of $`\alpha _W/\pi =10^2`$, $`C_F=3/4`$ and $`s/M^2=10^2`$ one finds
$$2\frac{\alpha _W}{\pi }\rho =1.5\times 10^2(5.30+3.451.70)=0.053.$$
Large compensations between leading and subleading terms are observed and, in fact, this pattern will reappear for the Standard Model as discussed below.
The treatment of is based on evolution equations that govern the dynamics of the amplitudes in the Sudakov limit as obtained in . In this approach was applied to the next-to-leading analysis of the Abelian form factor and the four fermion amplitude in the $`SU(N)`$ gauge theory. Functions that enter the evolution equations in the next-to-leading logarithmic approximation were evaluated by using, as an input, asymptotic expansions of one-loop diagrams. The solution of these equations lead to a summation of the leading and subleading Sudakov logarithms. The expansion of one-loop diagrams through the so-called generalized strategy of regions (see also ) identifies in a systematic way the nature of various contributions and the origin of logarithms. This strategy is based on expanding integrands of Feynman integrals in typical regions and extending the integration domains to the whole space of the loop momenta so that a crucial difference with respect to the standard approach is the absence of cut-offs that specify the regions in individual terms of the expansions.
## 2 The Abelian form factor in the Sudakov limit
Let us first analyse the (vector) form factor which determines the amplitude of the fermion scattering in the external Abelian field. In Born approximation
$$F_B=\overline{\psi }(p_2)\gamma _\mu \psi (p_1),$$
(2)
We consider the limit $`s=(p_1p_2)^2\mathrm{}`$ with on-shell massless fermions, $`p_1^2=p_2^2=0`$, and gauge bosons with a small non-zero mass $`M^2s`$. For convenience $`p_{1,2}=(Q/2,0,0,Q/2)`$ so that $`2p_1p_2=Q^2=s`$.
The asymptotic behaviour can be found by solving the corresponding evolution equation
$$\frac{}{\mathrm{ln}Q^2}F=\text{ }$$
(3)
$$\text{ }\left[_{M^2}^{Q^2}\frac{\text{d}x}{x}\gamma (\alpha (x))+\zeta (\alpha (Q^2))+\xi (\alpha (M^2))\right]F.$$
For the non-Abelian gauge theory, this equation was first derived in by factorizing collinear logarithms in the axial gauge. Its solution is
$$F=F_0(\alpha (M^2))\mathrm{exp}\{_{M^2}^{Q^2}\frac{\text{d}x}{x}\text{ }$$
(4)
$$\times [_{M^2}^x\frac{\text{d}x^{}}{x^{}}\gamma (\alpha (x^{}))+\zeta (\alpha (x))+\xi (\alpha (M^2))]\}.$$
For a proper treatment of the next-to-leading logarithms one must keep renormalization group corrections to the leading logarithmic approximation as well as single infrared and renormalization group logarithms. In this approximation
$$F=F_0(\alpha )\mathrm{exp}[_{M^2}^{Q^2}\frac{\text{d}x}{x}_{M^2}^x\frac{\text{d}x^{}}{x^{}}\gamma (\alpha (x^{}))\text{ }$$
(5)
$$\text{ }+(\zeta (\alpha )+\xi (\alpha ))\mathrm{ln}(Q^2/M^2)]$$
The leading terms of the functions $`\gamma `$, $`\zeta `$ and $`\xi `$ are obtained from the one loop analysis and the one loop running of $`\alpha `$ in the argument of the function $`\gamma `$ should be taken into account.
In the covariant gauge, the self energy insertions to the external fermion lines do not give $`Q`$-dependent contributions. The one loop calculation of the vertex correction gives
$$F=\frac{\alpha }{2\pi }C_F\left(V_0+2V_1+2(12ϵ)V_2V_2^{}\right)F_B,$$
(6)
where $`C_F`$ is the quadratic Casimir operator of the fundamental representation and the functions $`V_i`$ are obtained from
$$\frac{\text{d}^dk}{(k^22p_1k)(k^22p_2k)(k^2M^2)}=\text{ }$$
(7)
$$\text{ }i\pi ^{d/2}e^{\gamma _\mathrm{E}ϵ}s^1V_0,$$
$$\frac{\text{d}^dkk_\mu }{(k^22p_1k)(k^22p_2k)(k^2M^2)}=\text{ }$$
$$\text{ }i\pi ^{d/2}e^{\gamma _\mathrm{E}ϵ}s^1(p_1+p_2)_\mu V_1,$$
$$\frac{\text{d}^dkk_\mu k_\nu }{(k^22p_1k)(k^22p_2k)(k^2M^2)}=\text{ }$$
$$\text{ }i\pi ^{d/2}e^{\gamma _\mathrm{E}ϵ}\left[g_{\mu \nu }V_2+\frac{p_{1}^{}{}_{\mu }{}^{}p_{2}^{}{}_{\nu }{}^{}+(\mu \nu )}{s}V_2^{}\right]$$
To expand these integrals in the limit $`Q^2M^2`$ we apply a generalized strategy of regions formulated in and discussed using characteristic two-loop examples in :
* Consider various regions of the loop momenta and expand, in every region, the integrand in Taylor series with respect to the parameters that are there considered small;
* Integrate the expanded integrand over the whole integration domain of the loop momenta;
* Put to zero any scaleless integral.
The following “typical” regions arise in the Sudakov limit :
hard (h): $`kQ`$
1-collinear (1c): $`k_0+k_3Q`$
$`k_0k_3M^2/Q,k_{1,2}M`$
2-collinear (2c): $`k_0k_3Q`$
$`k_0+k_3M^2/Q,k_{1,2}M`$
soft (s): $`kM`$
ultrasoft (us): $`kM^2/Q.`$
Keeping the leading power in the expansion in the limit $`Q^2/M^2\mathrm{}`$ one observes that the leading double logarithm results from the hard region, whereas the single logarithm receives contributions from the hard and collinear regions as well. Soft regions do not contribute, at least in the leading power. Combining the vertex correction with the $`Q^2`$ independent fermion self energies one arrives at a finite result.
From the one-loop result one derives
$$\gamma (\alpha )=C_F\frac{\alpha }{2\pi }.$$
(8)
The total double logarithms originate from the hard region. This determines the scale of the coupling constant in the second order logarithmic derivative of the form factor in $`Q`$. At the same time we cannot distinguish, in the one loop approximation, the contribution to the functions $`\zeta `$ and $`\xi `$ coming from the collinear region because this region includes both $`Q`$ and $`M`$ scales. For the sum of these functions we find
$$\zeta (\alpha )+\xi (\alpha )=3C_F\frac{\alpha }{4\pi }.$$
(9)
Finally, in the NLO logarithmic approximation
$$F=F_B(1C_F\frac{\alpha }{2\pi }(\frac{7}{2}+\frac{2\pi ^2}{3}))\mathrm{exp}\{\frac{C_F}{2\pi }[\text{ }$$
$$\text{ }_{M^2}^{Q^2}\frac{\text{d}x}{x}_{M^2}^x\frac{\text{d}x^{}}{x^{}}\alpha (x^{})+3\alpha \mathrm{ln}(Q^2/M^2)]\}$$
(10)
in agreement with the result of .
## 3 The four fermion amplitude
We study the limit of the fixed-angle scattering when all the invariant energy and momentum transfers of the process are much larger than the typical mass scale of internal particles $`|s||t||u|M^2`$. Besides the extra kinematical variable the analysis of the four fermion amplitude is more complicated by the presence of different “color” and Lorentz structures. The Born amplitude, for example, can be expanded in the basis of color/chiral amplitudes
$$A_B=\frac{ig^2}{s}A^\lambda =\frac{ig^2}{s}T_F(\frac{1}{N}(A_{LL}^d+A_{LR}^d)$$
(11)
$$\text{ }+A_{LL}^c+A_{LR}^c+(LR)),$$
where
$`A^\lambda `$ $`=`$ $`\overline{\psi }_2(p_2)t^a\gamma _\mu \psi _1(p_1)\overline{\psi }_4(p_4)t^a\gamma _\mu \psi _3(p_3),`$
$`A_{LL}^d`$ $`=`$ $`\overline{\psi _2}_L^i\gamma _\mu \psi _{1}^{}{}_{L}{}^{i}\overline{\psi _4}_L^j\gamma _\mu \psi _{3}^{}{}_{L}{}^{j},`$ (12)
$`A_{LR}^c`$ $`=`$ $`\overline{\psi _2}_L^j\gamma _\mu \psi _{1}^{}{}_{L}{}^{i}\overline{\psi _4}_R^i\gamma _\mu \psi _{3}^{}{}_{R}{}^{j}`$
and so on. Here $`t^a`$ is the $`SU(N)`$ generator, $`p_1`$, $`p_3`$ are incoming and $`p_2`$, $`p_4`$ outgoing momenta so that $`t=(p_1p_4)^2`$ and $`u=(p_1+p_3)^2=(s+t)`$. For the moment we consider a parity conserving theory. Hence only two chiral amplitudes are independent, for example, $`LL`$ and $`LR`$. Similarly only two color amplitudes are independent, for example, $`A^\lambda `$ and $`A^d`$.
Let us first compute the one loop corrections, employing again the strategy of regions. The total contribution from vertex and box diagrams in the logarithmic approximation is independent from chirality and the same both for the $`LL`$ and $`LR`$ amplitudes:
$$\frac{ig^2(Q^2)}{s}\frac{1}{2}[\{C_FL(s)+(3C_FC_A\mathrm{ln}\left(\frac{u}{s}\right)$$
(13)
$$\text{ }+2(C_F\frac{T_F}{N})\mathrm{ln}\left(\frac{u}{t}\right))l(s)\}A^\lambda $$
$$\text{ }+\left\{2\frac{C_FT_F}{N}\mathrm{ln}\left(\frac{u}{t}\right)\mathrm{ln}\left(\frac{s}{M^2}\right)\right\}A^d]$$
with
$$L(s)=\frac{g^2}{16\pi ^2}\mathrm{ln}^2\left(\frac{s}{M^2}\right);l(s)=\frac{g^2}{16\pi ^2}\mathrm{ln}\left(\frac{s}{M^2}\right)$$
and the same both for the $`LL`$ and $`LR`$ amplitudes.
Now the collinear logarithms can be separated from the total one-loop correction. For each fermion-antifermion pair, they form the exponential factor found in the previous section (eq. (10)). This factor in addition incorporates the renormalization group logarithms which are not absorbed by changing the normalization scale of the gauge coupling. The rest of the single logarithms in eq. (13) is of the soft nature. Let us denote by $`\stackrel{~}{A}`$ the amplitude with the collinear logarithms factored out. It can be represented as a vector in the basis $`A^\lambda `$, $`A^d`$ and satisfies the following evolution equation
$$\frac{}{\mathrm{ln}Q^2}\stackrel{~}{A}=\chi (\alpha (Q^2))\stackrel{~}{A},$$
(14)
where $`\chi `$ is the matrix of the “soft” anomalous dimensions. From eq. (14) we find the elements of this matrix to be, in units of $`\alpha /(4\pi )`$,
$`\chi _{\lambda \lambda }`$ $`=`$ $`2C_A\mathrm{ln}\left({\displaystyle \frac{u}{s}}\right)+4\left(C_F{\displaystyle \frac{T_F}{N}}\right)\mathrm{ln}\left({\displaystyle \frac{u}{t}}\right)`$
$`\chi _{\lambda d}`$ $`=`$ $`4{\displaystyle \frac{C_FT_F}{N}}\mathrm{ln}\left({\displaystyle \frac{u}{t}}\right)`$
$`\chi _{d\lambda }`$ $`=`$ $`4\mathrm{ln}\left({\displaystyle \frac{u}{t}}\right)`$ (15)
$`\chi _{dd}`$ $`=`$ $`0.`$
The solution of eq. (14) reads
$$\stackrel{~}{A}=A_1^0(\alpha (M^2))\mathrm{exp}\left[_{M^2}^{Q^2}\frac{\text{d}x}{x}\chi _1(\alpha (x))\right]$$
(16)
$$\text{ }+A_2^0(\alpha (M^2))\mathrm{exp}\left[_{M^2}^{Q^2}\frac{\text{d}x}{x}\chi _2(\alpha (x))\right],$$
where $`\chi _i`$ are eigenvalues of the $`\chi `$ matrix and $`A_i^0`$ are $`Q`$-independent vectors.
¿From the asymptotic expansion of the box diagrams one finds that only the hard parts contribute to eq. (14). This fixes the scale of $`\alpha `$ in this equation to be $`Q`$.
In the Abelian case, there are no different color amplitudes and there is only one anomalous dimension
$$\chi =4\mathrm{ln}\left(\frac{u}{t}\right).$$
(17)
## 4 Sudakov logarithms in electroweak processes
We are interested in the process $`f^{}\overline{f}^{}f\overline{f}`$. In the Born approximation, its amplitude is of the following form
$$A_B=\frac{ig^2}{s}\underset{I,J=L,R}{}\left(T_f^{}^3T_f^3+t_W^2\frac{Y_f^{}Y_f}{4}\right)A_{IJ}^{f^{}f},$$
(18)
where
$$A_{IJ}^{f^{}f}=\overline{f}_I^{}\gamma _\mu f_I^{}\overline{f}_J\gamma _\mu f_J,$$
(19)
To analyze the electroweak correction to the above process we use the approximation with the $`W`$ and $`Z`$ bosons of the same mass $`M`$ and massless quarks and leptons. The photon is massless, and the corresponding infrared divergent contributions should be accompanied by the real soft photon radiation integrated to some resolution energy $`\omega _{res}`$ to get an infrared safe cross section independent on an auxiliary photon mass. At the same time the massive gauge bosons are supposed to be detected as separate particles. In practice, the resolution energy is much less than the $`W`$ ($`Z`$) boson mass so the soft photon emission is of the QED nature. This cancels the infrared singularities of the QED virtual correction. We should therefore separate the QED virtual correction from the complete result computed with the photon of some mass $`\lambda `$ and then evaluate the QED virtual corrections together with the real soft photon radiation effects with vanishing $`\lambda `$. It is convenient to subtract the QED contribution computed with the photon of the mass $`M`$ from the obtained result for the virtual corrections and then take the limit $`\lambda 0`$ for the sum of QED virtual and real photon contributions to the total amplitude. In the language of the approach of ref. , this prescription means that we use the auxiliary photon mass $`\lambda `$ as a variable of the evolution equation below the scale $`M`$ and the subtraction fixes a relevant initial condition for this differential equation. This leads to a modification of the collinear factor and the soft anomalous dimensions.
Thus we keep always a cutoff $`\omega _{res}M`$ and do not display the QED Sudakov factor arising from this suppression of photon radiation. The remaining “electroweak”universal collinear factor for each fermion-antifermion pair becomes
$$\mathrm{exp}[(T_f(T_f+1)+t_W^2\frac{Y_f^2}{4}s_W^2Q_f^2)\text{ }$$
(20)
$$\text{ }\times (L(s)3l(s))].$$
The soft anomalous dimension for $`I`$ and/or $`J=R`$ is Abelian and, in units of $`g^2/(16\pi ^2)`$, reads
$$\chi =\left(t_W^2Y_f^{}Y_f4s_W^2Q_f^{}Q_f\right)\mathrm{ln}\left(\frac{u}{t}\right),$$
(21)
and the matrix of the soft anomalous dimension for $`I=J=L`$ is
$`\chi _{\lambda \lambda }`$ $`=`$ $`4\mathrm{ln}\left({\displaystyle \frac{u}{s}}\right)`$
$`+\left(t_W^2Y_f^{}Y_f4s_W^2Q_f^{}Q_f+2\right)\mathrm{ln}\left({\displaystyle \frac{u}{t}}\right)`$
$`\chi _{\lambda d}`$ $`=`$ $`{\displaystyle \frac{3}{4}}\mathrm{ln}\left({\displaystyle \frac{u}{t}}\right)`$
$`\chi _{d\lambda }`$ $`=`$ $`4\mathrm{ln}\left({\displaystyle \frac{u}{t}}\right)`$ (22)
$`\chi _{dd}`$ $`=`$ $`\left(t_W^2Y_f^{}Y_f4s_W^2Q_f^{}Q_f\right)\mathrm{ln}\left({\displaystyle \frac{u}{t}}\right).`$
The one-loop leading and subleading logarithms can be directly obtained from eq. (13). The two-loop leading (infrared) logarithms are determined by the second order term of the expansion of the double (soft$`\times `$collinear) logarithmic part of the collinear factors (20). The two-loop next-to-leading logarithms are generated by the interference between the first order terms of the expansion of the double (soft$`\times `$collinear) and single (soft$`+`$collinear$`+`$renormalization group) logarithmic exponents and can also be found in . With the expression for the chiral amplitudes at hand, we can compute the leading and subleading logarithmic corrections to the basic observables for $`e^+e^{}f\overline{f}`$. Let us, for example, consider the total cross sections of the quark-antiquark/$`\mu ^+\mu ^{}`$ production in the $`e^+e^{}`$ annihilation. In the two loop approximation, the leading and next-to-leading Sudakov corrections to the cross sections read
$$\sigma /\sigma _B(e^+e^{}Q\overline{Q})=1+5.30l(s)1.66L(s)$$
$$12.84l(s)L(s)+1.92L^2(s),$$
$$\sigma /\sigma _B(e^+e^{}q\overline{q})=1+20.54l(s)2.17L(s)$$
$$53.72l(s)L(s)+2.79L^2(s),$$
$$\sigma /\sigma _B(e^+e^{}\mu ^+\mu ^{})=1+10.09l(s)1.39L(s)$$
$$21.66l(s)L(s)+1.41L^2(s),$$
(23)
where $`Q=u,c,t`$, $`q=d,s,b`$. Numerically, $`L(s)=0.07`$ $`(0.11)`$ and $`l(s)=0.014`$ $`(0.017)`$ respectively for $`\sqrt{s}=1`$ TeV and $`2`$ TeV.
Clearly, for energies at 1 and 2 TeV the two loop corrections are huge and amount up respectively to $`5\%`$ and $`7\%`$. There is a cancellation between the leading and subleading logarithms and for the above energy interval the subleading contribution even exceeds the leading one. The higher order leading and next-to-leading corrections however do not exceed $`1\%`$ level. They can be in principle resummed using the formulae given above. The leading and subleading corrections to the left right and forward backward asymmetries are typically smaller.
Our result for the one loop double logarithmic contribution is in agreement with . However the result for the one loop single infrared logarithmic contribution differs from . The reason is that, in , only the diagrams with heavy virtual bosons have been taken into account. There is an infrared safe contribution of the diagram with the virtual massless photon where the heavy boson mass serves as an infrared regulator that should be taken into account to get a complete (exponential) result. In one-loop approximation, this contribution comes from the box diagrams with the photon and $`Z`$ boson running inside the loop .
Our result for the two-loop double logarithmic contribution is in agreement with . On the other hand, the coefficients in front of the two-loop leading logarithms in eq. (23) with a few percent accuracy coincide with the result of where the photon contributions were not considered. This is related to the fact that the virtual photon contribution not included to the result of is suppressed by a small factor $`s_W^2`$.
Acknowledgments
The work by V.S. was supported by the Volkswagen Foundation, contract No. I/73611. The work by J.K. and A.P. was supported by the Volkswagen Foundation, by BMBF under grant BMBF-057KA92P and by DFG-Forschergruppe “Quantenfeldtheorie, Computeralgebra und Monte-Carlo-Simulationen” (DFG Contract KU502/6–1).
|
warning/0005/physics0005050.html
|
ar5iv
|
text
|
# Nonlinear relaxation field in charged systems under high electric fields
## I Introduction
High field transport has become a topic of current interest in various fields of physics. In semiconductors the nonlinear transport effects are accessible due to femto - second laser pulses and shrink devices . In plasma physics these field effects can be studied within such short pulse periods . One observable of interest is the current or the electrical conductivity which gives access to properties of dense nonideal plasmas . In high energy physics the transport in strong electric fields is of interest due to pair creation . In order to describe these field effects one can start conveniently from kinetic theory. Within this approach the crucial question is to derive appropriate kinetic equations which include field effects beyond linear response.
At low strength of the external electric field one expects the linear response regime to be valid. Then the contribution of field effects to the conductivity can be condensed into the Debye- Onsager relaxation effect which was first derived within the theory of electrolytes . Debye has given a limiting law of electrical conductivity which stated that the external electric field $`E`$ on a single charge $`Z=1`$ is diminished in an electrolyte solution by the amount
$`\delta E=E(1{\displaystyle \frac{\kappa e^2}{6T}})`$ (1)
where $`e`$ is the elementary charge, $`E`$ the electric field strength, $`T`$ is the temperature of the plasma and $`\kappa `$ is the inverse screening radius of the screening cloud. This law is interpreted as a deceleration force which is caused by the deformed screening cloud surrounding the charge. Later it has been shown by Onsager that this result has to be corrected to
$`\delta E=E(1{\displaystyle \frac{\kappa e^2}{3(2+\sqrt{2})T}})`$ (2)
if the dynamics of ions ($`Z=1`$) is considered. While the linear response theory seems to reproduce this Onsager result , the kinetic theory seems to support more the Debye result . The correct treatment is a matter of ongoing debate. In this paper we will give the result beyond linear response for the statically and dynamically screened approximation. Here different approximations of kinetic theory will be discussed and the one which leads to the closest form to the hydrodynamical approximation (Onsager result) is presented.
The kinetic approach describes the time evolution of the one particle distribution function within an external field $`𝐄`$ as
$`{\displaystyle \frac{}{t}}feZ𝐄{\displaystyle \frac{}{𝐤}}f=I[f,𝐄]`$ (3)
where the field dependent collision integral $`I[f,𝐄]`$ has to be provided by different approximations. Integrating this kinetic equation over the momentum $`𝐤`$ one obtains the balance of the current. For simplicity we assume that the distribution function can be parameterized by a displaced local equilibrium one with a field and time dependent momentum $`f(𝐤,t)=f_0(𝐤𝐩(𝐄,t))`$ which is related to the current $`𝐉`$ as
$`𝐉(E)=nZe{\displaystyle \frac{𝐩(E)}{m}}`$ (4)
if the charge is $`Ze`$, the density $`n`$ and the mass $`m`$. The balance equation for the field and time dependent local momentum $`𝐩(E,t)`$ follows from (3) as
$`{\displaystyle \frac{}{t}}𝐩eZn(1+{\displaystyle \frac{\delta E(E)}{E}})𝐄=R(E)eZn𝐉`$ (5)
where the relaxation field $`\delta E(E)`$ as well as the free conductivity $`R(E)`$ follows from the field dependent collision integral. The total conductivity $`𝐄=\sigma 𝐉`$ is then given by
$`\sigma (E)={\displaystyle \frac{R(E)}{1+\frac{\delta E(E)}{E}}}.`$ (6)
The free conductivity $`R`$ is the subject of intense investigations in the literature . It is known that the Coulomb divergence for small wave vectors is omitted if screening is included and the divergence at large wave vectors is omitted by the De Broglie wavelength i.e. by the quantum effects. We will not consider the discussion of the free conductivity $`R`$ here but concentrate on the relaxation field $`\delta E`$. The free conductivity can be obtained by the same considerations as will be outlined here. We want to point out that the relaxation field will turn out to be free of long wave divergences in the classical limit in contrast to the free conductivity $`R`$.
First we recall the hydrodynamical approach starting from the classical Bogoliubov-Born-Green-Kirkwood-Yvon (BBGKY) hierarchy which results into an analytical formula for the classical relaxation effect already reported . This result is then compared with the quantum kinetic approach. We give a short rederivation of the field dependent kinetic equations in dynamical screened approximation from the Green’s function technique in Sec. III. Two approximations, the static screening as well as dynamical screening are presented. In the fourth section we will derive the field dependent current analytically. We present both the statically as well as dynamically screened treatment as analytical results. The classical expressions for the statically screened result is compared with the classical result from hydrodynamical approximation. The dynamical result is then derived analytically too and it will be shown that only for asymmetric screening the hydrodynamical result can be approached. In Sec. V we shortly discuss the physical limitation of field strengths for the local equilibrium assumption and the gradient approximation. Sec. VI summarizes and the appendix gives the calculation of some involved integrals appearing during the integration of the Lenard-Balescu equation.
## II Approach by classical BBGKY-Hierarchy
The starting point for the classical considerations is the BBGKY hierarchy which reads for the one - particle distribution function $`F_a`$
$`{\displaystyle \frac{F_a}{t}}+𝐯{\displaystyle \frac{F_a}{𝐫}}+{\displaystyle \frac{e_a}{m_a}}\overline{𝐄}{\displaystyle \frac{F_a}{𝐯}}S_aF_a`$ (7)
$`={\displaystyle \underset{b}{}}{\displaystyle \frac{n_be_ae_b}{m_a}}{\displaystyle \frac{}{𝐯}}{\displaystyle 𝑑𝐫^{}𝑑𝐯^{}F_{ab}(𝐫,𝐫^{},𝐯,𝐯^{})\frac{}{𝐫}\frac{1}{\left|𝐫𝐫^{}\right|}}`$ (8)
and the two - particle distribution function $`F_{ab}`$
$`{\displaystyle \frac{F_{ab}}{t}}+𝐯{\displaystyle \frac{F_{ab}}{𝐫}}+𝐯^{}{\displaystyle \frac{F_{ab}}{𝐫^{}}}+{\displaystyle \frac{e_a}{m_a}}\overline{𝐄}{\displaystyle \frac{F_{ab}}{𝐯}}`$ (10)
$`+{\displaystyle \frac{e_b}{m_b}}\overline{𝐄}{\displaystyle \frac{F_{ab}}{𝐯^{}}}S_aF_{ab}S_bF_{ab}`$ (11)
$`=e_ae_b{\displaystyle \frac{}{𝐫}}{\displaystyle \frac{1}{\left|𝐫𝐫^{}\right|}}\left({\displaystyle \frac{1}{m_a}}{\displaystyle \frac{F_{ab}}{𝐯}}{\displaystyle \frac{1}{m_b}}{\displaystyle \frac{F_{ab}}{𝐯^{}}}\right)`$ (12)
$`+{\displaystyle \underset{c}{}}n_ce_c{\displaystyle }d𝐫^{\prime \prime }d𝐯^{\prime \prime }({\displaystyle \frac{e_a}{m_a}}{\displaystyle \frac{}{𝐫}}{\displaystyle \frac{1}{\left|𝐫𝐫^{\prime \prime }\right|}}{\displaystyle \frac{F_{abc}}{𝐯}}`$ (13)
$`+{\displaystyle \frac{e_b}{m_b}}{\displaystyle \frac{}{𝐫^{}}}{\displaystyle \frac{1}{\left|𝐫^{}𝐫^{\prime \prime }\right|}}{\displaystyle \frac{F_{abc}}{𝐯^{}}})`$ (14)
with the external field $`𝐄`$. $`S_a`$ describes a collision integral with some background which we will specify later. This hierarchy is truncated approximating that
$`F_{ab}`$ $`=`$ $`F_aF_b+g_{ab}`$ (15)
$`F_{abc}`$ $`=`$ $`F_aF_bF_c+F_ag_{bc}+F_bg_{ac}+F_cg_{ab}`$ (16)
where $`g_{ab}(𝐫_𝐚,𝐫_𝐛,𝐯_𝐚,𝐯_𝐛)`$ is the two-particle correlation function.
Within the local equilibrium approximation we suppose a stationary (for example a local Maxwellian) distribution for the velocities in the one and two-particle distribution functions
$`f_a(𝐫,𝐯,t)=n_a(𝐫,t)\left({\displaystyle \frac{m_a}{2\pi T}}\right)^{3/2}\mathrm{exp}\left[{\displaystyle \frac{m_a(𝐯𝐮_𝐚)^2}{2T}}\right]`$ (17)
$`g_{ab}(𝐫,𝐫^{},𝐯,𝐯^{},t)=F_{ab}F_aF_b`$ (18)
$`=h_{ab}(𝐫,𝐫^{},t)\left({\displaystyle \frac{m_am_b}{4\pi ^2T^2}}\right)^{3/2}`$ (19)
$`\times \mathrm{exp}\left[{\displaystyle \frac{m_a(𝐯𝐰_{\mathrm{𝐚𝐛}})^2}{2T}}{\displaystyle \frac{m_b(𝐯^{}𝐰_{\mathrm{𝐛𝐚}})^2}{2T}}\right].`$ (20)
Here we have introduced the local one-particle density and the local average velocity
$`n_a(𝐫,t)`$ $`=`$ $`{\displaystyle 𝑑𝐯F_a(𝐫,𝐯,t)}`$ (21)
$`𝐮_𝐚`$ $`=`$ $`{\displaystyle \frac{1}{n_a}}{\displaystyle 𝑑\mathrm{𝐯𝐯}F_a(𝐫,𝐯,t)}`$ (22)
as well as the pair correlation function and the average pair velocity
$`h_{ab}(𝐫,𝐫^{},t)`$ $`=`$ $`{\displaystyle 𝑑𝐯𝑑𝐯^{}g_{ab}(𝐫,𝐫^{},𝐯,𝐯^{},t)},`$ (23)
$`w_{ab}(𝐫,𝐫^{},t)`$ $`=`$ $`{\displaystyle \frac{1}{h_{ab}}}{\displaystyle 𝑑𝐯𝑑𝐯^{}𝐯g_{ab}(𝐫,𝐫^{},𝐯,𝐯^{},t)}.`$ (24)
Further on, we suppose that the particles interact with some background (e.g. neutrals or electrolyte solvent) by the collision integrals $`S_a`$ with the following properties
$`{\displaystyle 𝑑𝐯S_af_a}`$ $`=`$ $`0`$ (26)
$`{\displaystyle 𝑑\mathrm{𝐯𝐯}S_af_a}`$ $`=`$ $`{\displaystyle \frac{1}{b_am_a}}\rho _a𝐮_𝐚,`$ (27)
$`{\displaystyle 𝑑\mathrm{𝐯𝐯}S_ag_{ab}(𝐫,𝐫^{},𝐯,𝐯^{},t)}`$ $`=`$ $`{\displaystyle \frac{1}{b_am_a}}h_{ab}𝐰_{\mathrm{𝐚𝐛}}`$ (28)
where $`b_a`$ is the mobility of particle of type a. This friction with a background serves here to couple the two - particle equations and will be considered infinitesimal small in the end. However, as we will demonstrate this yields to a symmetry breaking in the system which leads basically to different results than neglecting this friction.
Fourier transform of the resulting two equations (15) into momentum space and assuming a homogeneous density $`n(𝐫)=n`$ we arrive at the coupled equation system
$`{\displaystyle \frac{e_a}{T}}\overline{𝐄}.(𝐯_𝐚𝐮_𝐚)f_a={\displaystyle \frac{𝐮_𝐚}{b_am_a}}+{\displaystyle \underset{b}{}}{\displaystyle \frac{4\pi n_be_ae_b}{T}}`$ (29)
$`\times {\displaystyle }{\displaystyle \frac{d𝐤}{(2\pi )^3}}{\displaystyle \frac{i𝐤.(𝐯_𝐚𝐰_{\mathrm{𝐚𝐛}})}{k^2}}f_a(𝐯_𝐚𝐰_{\mathrm{𝐚𝐛}}+𝐮_𝐚)h_{ab}(𝐤)`$ (30)
and
$`i𝐤.(𝐯_𝐚𝐯_𝐛)g_{ab}(e_a(𝐯_𝐚𝐰_{\mathrm{𝐚𝐛}})+e_b(𝐯_𝐛𝐰_{\mathrm{𝐛𝐚}})){\displaystyle \frac{\overline{𝐄}}{T}}g_{ab}`$ (32)
$`=ie_ae_b{\displaystyle \frac{4\pi }{k^2}}𝐤.(𝐯_𝐚𝐮_𝐚𝐯_𝐛+𝐮_𝐛)f_af_b`$ (33)
$`i{\displaystyle \frac{d\overline{𝐤}}{(2\pi )^3}\frac{4\pi e_ae_b}{Tk^2}\overline{𝐤}}.(𝐯_𝐚𝐰_{\mathrm{𝐚𝐛}}𝐯_𝐛+𝐰_{\mathrm{𝐛𝐚}})g_{ab}(𝐤\overline{𝐤})`$ (34)
$`{\displaystyle \underset{c}{}}n_c{\displaystyle }d𝐯_𝐜{\displaystyle \frac{4\pi ie_c}{Tk^2}}[e_a𝐤.(𝐯_𝐚𝐮_𝐚)f_ag_{cb}(𝐤)`$ (35)
$`e_b𝐤.(𝐯_𝐛𝐮_𝐛)f_bg_{ac}(𝐤)]+S_ag_{ab}+S_bg_{ab}`$ (36)
with
$`\overline{𝐄}=𝐄{\displaystyle \underset{b}{}}n_be_b{\displaystyle 𝑑𝐫_𝐛𝑑𝐯_𝐛\frac{}{𝐫_𝐛}\frac{1}{|𝐫_𝐚𝐫_𝐛|}F_b}.`$ (37)
By multiplying the above equation system by $`1,𝐯_𝐚,𝐯_𝐛`$ and integrating over the velocities we obtain the Onsager equation
$`b_a\left[Th_{ab}(𝐤)(1+i{\displaystyle \frac{e_a}{e}}a)+e_a\mathrm{\Phi }_b(𝐤)\right]`$ (38)
$`=b_b\left[Th_{ab}(𝐤)(1i{\displaystyle \frac{e_b}{e}}a)+e_b\mathrm{\Phi }_a(𝐤)\right]`$ (39)
with
$`k^2\mathrm{\Phi }_a(𝐤)`$ $`=`$ $`4\pi e_a+{\displaystyle \underset{c}{}}n_ce_ch_{ac}(𝐤)`$ (40)
$`k^2\mathrm{\Phi }_a(𝐤)`$ $`=`$ $`4\pi e_a+{\displaystyle \underset{c}{}}n_ce_ch_{ca}(𝐤)`$ (41)
for the two -particle correlation function $`h_{ab}`$. Here we use
$`a={\displaystyle \frac{e𝐤.\overline{𝐄}}{k^2T}}.`$ (42)
Let us already remark here that the friction with a background described by the mobilities $`b`$ couple the two sides of the equation (39). If we had not considered this friction, $`S_i=0`$, we would have obtained that the left and the right hand side of (39) vanish separately. This will lead essentially to a different result even for infinite small friction. There is no continuous transition between these two extreme cases pointing to a symmetry breaking in the two treatments. Let us first discuss the case with background friction.
### A With background friction
The system (39) for electrons, $`e_e=e`$, and ions, $`e_i=Ze`$, with charge $`Z`$ reads expanded
$`Th_{ee}`$ $`=`$ $`e{\displaystyle \frac{\mathrm{\Phi }_e(𝐤)+\mathrm{\Phi }_e(𝐤)}{2}}`$ (43)
$`Th_{ei}`$ $`=`$ $`e{\displaystyle \frac{\mathrm{\Phi }_i(𝐤)Z\frac{b_i}{b_e}\mathrm{\Phi }_e(𝐤)}{1+\frac{b_i}{b_e}+ia(1+\frac{b_i}{b_e}Z)}}`$ (44)
$`Th_{ie}`$ $`=`$ $`e{\displaystyle \frac{\mathrm{\Phi }_i(𝐤)Z\frac{b_i}{b_e}\mathrm{\Phi }_e(𝐤)}{1+\frac{b_i}{b_e}ia(1+\frac{b_i}{b_e}Z)}}`$ (45)
$`Th_{ii}`$ $`=`$ $`Ze{\displaystyle \frac{\mathrm{\Phi }_i(𝐤)+\mathrm{\Phi }_i(𝐤)}{2}}.`$ (46)
This we can solve together with (41). First we calculate the effective field strength at the position of the electron in linear response the Onsager result
$`{\displaystyle \frac{\delta E}{E}}𝐄`$ $`=`$ $`i{\displaystyle \frac{𝐄}{E}}{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}k^3𝑑k{\displaystyle \underset{1}{\overset{1}{}}}d(\mathrm{cos}\theta )\mathrm{cos}\theta \mathrm{\Phi }_e(𝐤)`$ (47)
$`=`$ $`𝐄{\displaystyle \frac{\kappa e^2}{3T}}{\displaystyle \frac{Zq}{\sqrt{q}+1}}`$ (48)
with $`\kappa ^2=\kappa _e^2(1+Z)=\frac{4\pi (e^2n_e+Z^2e^2n_i)}{T}`$ and
$`q={\displaystyle \frac{b_e+Zb_i}{(1+Z)(b_e+b_i)}}.`$ (49)
For single charged ions $`Z=1`$ the influence of the mobilities drop out and we recover the result (2).
Since this result is independent of the mobilities one could conclude that this is an universal limiting law. However we will express two doubts here. As one sees for charges $`Z>1`$ the result (2) is approached only in the limit where the ion mobilities are much smaller than the electron mobilities $`b_i/b_e0`$. This means of course that the electrons have different friction with a thought background than the ions. In other words there is an explicit symmetry breaking mechanism included by assuming such collision integrals with the background. Therefore we will obtain another solution if we consider no friction.
The second remark concerns the limit of one-component system which one can obtain by setting $`Z=1`$. The Onsager result or hydrodynamical result with friction (48) leads to twice the Debye result (1) in this case but with opposite sign. Oppositely we will see in the following that the perfectly symmetric treatment of the species without friction with a background will lead to a vanishing one component limit as it should. This again underlies the symmetry breaking if one assumes an infinitesimal small friction with a background.
For completeness, we want to recall the expression of the nonlinear Onsager result which is obtained from the limit $`b_i/b_e0`$ of the system (43)
$`Th_{ee}+e{\displaystyle \frac{\phi _e(𝐤)+\phi _e(𝐤)}{2}}`$ $`=`$ $`0`$ (50)
$`h_{ei}(T+ie{\displaystyle \frac{\mathrm{𝐤𝐄}}{k^2}})+e\phi _i(𝐤)`$ $`=`$ $`o({\displaystyle \frac{b_i}{b_e}})=0`$ (51)
$`h_{ie}(Tie{\displaystyle \frac{\mathrm{𝐤𝐄}}{k^2}})e\phi _i(𝐤)`$ $`=`$ $`o({\displaystyle \frac{b_i}{b_e}})=0`$ (52)
$`Th_{ii}Ze{\displaystyle \frac{\phi _i(𝐤)+\stackrel{~}{\phi }_i(𝐤)}{2}}`$ $`=`$ $`0.`$ (53)
One obtains the result for $`Z=1`$
$`\delta 𝐄`$ $`=`$ $`{\displaystyle \frac{e^2\kappa _e}{3(1+\sqrt{2})T}}𝐄F_H({\displaystyle \frac{eE}{T\kappa _e}})`$ (55)
$`=`$ $`{\displaystyle \frac{e^2\kappa _e}{6T}}𝐄\{\begin{array}{c}2\sqrt{2}+o(E)\\ \\ \frac{3\kappa T}{2eE}+o(1/E)^2\end{array}`$ (56)
with
$`F_H(\alpha )`$ $`=`$ $`{\displaystyle \frac{3(1+\sqrt{2})}{\alpha ^2}}[{\displaystyle \frac{1}{2}}\sqrt{\alpha ^2+2}1+{\displaystyle \frac{1}{\alpha }}\mathrm{arctan}(\alpha )`$ (58)
$`{\displaystyle \frac{1}{\alpha }}\mathrm{arctan}({\displaystyle \frac{\alpha }{\sqrt{\alpha ^2+2}}})].`$
The numerical values of this result will be discussed in chapter (IV C).
### B Without background
Now we reconsider the steps from (36) to (39) without friction with the background. We obtain that both sides of (39) vanish separately
$`Th_{ab}(𝐤)(1+i{\displaystyle \frac{e_a}{e}}a)+e_a\mathrm{\Phi }_b(𝐤)=0`$ (59)
$`Th_{ab}(𝐤)(1i{\displaystyle \frac{e_b}{e}}a)+e_b\mathrm{\Phi }_a(𝐤)=0.`$ (60)
Both equations have identical solutions $`h_{ab}`$ which can be easily verified using the symmetry $`h_{ab}(𝐤)=h_{ba}(𝐤)`$. Together with (41) we can solve for $`\mathrm{\Phi }_e`$ and the relaxation field is obtained instead of (55)
$`\delta 𝐄`$ $`=`$ $`{\displaystyle \frac{e^2\kappa _e\sqrt{1+Z}}{6T}}(Z+1)𝐄F_N({\displaystyle \frac{eE}{T\kappa _e}})`$ (61)
which takes for $`Z=1`$
$`{\displaystyle \frac{\delta E}{E}}={\displaystyle \frac{e^2\kappa _e}{6T}}\{\begin{array}{c}2+o(E)\\ \\ \frac{3\kappa T}{eE}+o(1/E)^2\end{array}`$ (62)
with
$`F_N(\alpha )={\displaystyle \frac{3}{(1+Z)\alpha ^2}}[\sqrt{4+(1+Z)\alpha ^2}`$ (63)
$`+{\displaystyle \frac{4}{\sqrt{1+Z}\alpha }}\mathrm{log}{\displaystyle \frac{2}{\sqrt{1+Z}\alpha +\sqrt{4+(1+Z)\alpha ^2}}}].`$ (64)
We see that the linear response result for $`Z=1`$ is twice the Debye result (1). For equal charged system $`Z=1`$ which would coincide with a one component plasma no relaxation effect appears as one would expect. In other words in a perfectly symmetric mathematical two - component plasma there is another relaxation effect than in a system which distinguishes the components by a different treatment of friction. The Onsager result (55) does not vanish for the limit of one-component plasma $`Z=1`$. This is due to the different treatment of ions and electrons there which assumes explicitly a two component plasma. Therefore the limit $`Z=1`$ does not work there.
This result is quite astonishing. One would expect that the limiting procedure which transforms the system (41) into (60) would also lead to a smooth transitions of the end results. However this is not the case. While the separate limit of $`b_{e,i}\mathrm{}`$ of (41) leads to (60) there is no possibility to transform the result (48) into the linear response result of (62). This underlines that due to even infinitesimal small friction assumed in obtaining (48) there occurs a symmetry breaking in the sense that the electrons and ions are not anymore symmetrically treated.
This lesson we have to keep in mind when we now advance and investigate the systematic treatment by quantum kinetic theory. There we will find also complete different results when we use asymmetric screening compared to symmetric screening. Of course, we will not assume any phenomenological friction since the kinetic theory provides for a systematic description of all occurring processes. Here we want only to point out that the above symmetry breaking is the main reason for the confusion in literature. Following the linear response formalism an asymmetric treatment of two - particle correlation functions is used in that the electrons are statically screened . This seemingly innocent usage leads there to an occasional agreement for $`Z=1`$ with the Onsager result (2).
Another advantage of the kinetic theory we want to point out here. The classical local equilibrium or hydrodynamical approximation does not lead to a mass dependence of the relaxation effect. This will be provided by the kinetic theory.
## III Quantum kinetic theory
We will formulate the kinetic theory within gauge invariant functions not missing field effects. The most promising theoretical tool is the Green function technique . The resulting equations show some typical deviations from the ordinary Boltzmann equation: (i) A collision broadening which consists in a smearing out of the elementary energy conservation of scattering. This is necessary to ensure global energy conservation . (ii) The intra-collisional field effect, which gives additional retardation effects in the momentum of the distribution functions. This comes mainly from the gauge invariance.
One of the most important questions is the range of applicability of these kinetic equations. Up to which field strengths are such modifications important and appropriate described within one-particle equations? In this question has been investigated for semiconductor transport. It was found that for high external fields the intra-collisional field effect becomes negligible. This range is given by a characteristic time scale of field effects $`\tau _F^2=m\mathrm{}/(e𝐄𝐪)`$ which has to be compared with the inverse collision frequency. This criterion is a pure quantum one. It remains the question whether there are also criteria in the classical limit. For a plasma system we will discuss in Sec. V that there is indeed a critical value of the field strength which can be given by classical considerations.
### A Definitions
In order to describe correlations in highly nonequilibrium situations, we define various correlation functions by different products of creation and annihilation operators
$`G^>(1,2)`$ $`=`$ $`<\mathrm{\Psi }(1)\mathrm{\Psi }^+(2)>`$ (65)
$`G^<(1,2)`$ $`=`$ $`<\mathrm{\Psi }^+(2)\mathrm{\Psi }(1)>.`$ (66)
Here $`<>`$ is the average value with the unknown statistical nonequilibrium operator $`\rho `$ and $`1`$ denotes the cumulative variables $`(𝐫_\mathrm{𝟏},s_1,t_1\mathrm{})`$ of space, spin, time etc. The equation of motion for the correlation functions are given in the form of the Kadanoff-Baym equation
$``$ $`i`$ $`\left(G_0^1G^<G^<G_0^1\right)=i\left(G^R\mathrm{\Sigma }^<\mathrm{\Sigma }^<G^A\right)`$ (68)
$`i\left(\mathrm{\Sigma }^RG^<G^<\mathrm{\Sigma }^A\right)`$
where the retarded and advanced functions are introduced as $`A^R(1,2)=i\mathrm{\Theta }(t_1t_2)[A^>\pm A^<]`$ and $`A^A(1,2)=i\mathrm{\Theta }(t_2t_1)[A^>\pm A^<]`$. Here operator notation is employed where products are understood as integrations over intermediate variables (time and space) and the upper/lower sign stands for Fermions/Bosons respectively. The Hartree- Fock drift term reads
$$G_0^1(11^{})=\left(i\mathrm{}\frac{}{t_1}+\frac{\mathrm{}^2}{2m}_{𝐱_\mathrm{𝟏}}^2\mathrm{\Sigma }_{HF}(11^{})\right)\delta (11^{})$$
(69)
with the Hartree Fock self energy
$`\mathrm{\Sigma }_{HF}(1,1^{})`$ (70)
$`=(\delta (𝐫_\mathrm{𝟏}𝐫_\mathrm{𝟏}^{}){\displaystyle }d𝐫_\mathrm{𝟐}V(𝐫_\mathrm{𝟏}𝐫_\mathrm{𝟐})G^<(𝐫_\mathrm{𝟐}t_1^{}𝐫_\mathrm{𝟐}t_1)`$ (71)
$`+V(𝐫_\mathrm{𝟏}𝐫_\mathrm{𝟏}^{})G^<(𝐫_\mathrm{𝟏}t_1𝐫_\mathrm{𝟏}^{}t_1^{}))\delta (t_1t_1^{})`$ (72)
where $`G(𝐫_\mathrm{𝟐},t_1,𝐫_\mathrm{𝟐},t_1)=n(𝐫_\mathrm{𝟐},t_1)`$ is the density.
### B Gauge invariance
In order to get an unambiguous way of constructing approximations we have to formulate our theory in gauge invariant way. This can be done following a procedure known from field theory . This method has been applied to high field problems in . With the help of the Fourier transform of an arbitrary function G(x,X) over the relative coordinates $`x=(𝐫_\mathrm{𝟐}𝐫_\mathrm{𝟏},t_2t_1)=(𝐫,\tau )`$ with the center of mass coordinates $`X=((𝐫_\mathrm{𝟐}+𝐫_\mathrm{𝟏})/2,(t_2+t_1)/2)=(𝐑,t)`$ one can introduce a gauge-invariant Fourier-transform of the difference coordinates $`x`$
$`\overline{G}(k,X)={\displaystyle 𝑑xG(xX)}`$ (73)
$`\times \mathrm{exp}\{{\displaystyle \frac{i}{\mathrm{}}}{\displaystyle \underset{\frac{1}{2}}{\overset{\frac{1}{2}}{}}}𝑑\lambda x_\mu [k^\mu +{\displaystyle \frac{e}{c}}A^\mu (X+\lambda x)]\}.`$ (74)
For constant electric fields, which will be of interest in the following, one obtains a generalized Fourier-transform
$`\overline{G}(k,X)={\displaystyle 𝑑x\mathrm{e}^{\frac{i}{\mathrm{}}[x_\mu k^\mu +e\mathrm{𝐫𝐄}t]}G(x,X)},`$
where the $`\chi `$ function was chosen in such a way that the scalar potential is zero $`A^\mu =(0,c𝐄t)`$. Therefore, we have the following rule in formulating the kinetic theory gauge-invariantly
1. Fourier transformation of the 4-dimensional difference-variable x to canonical momentum p.
2. Shifting the momentum to kinematic momentum according to $`𝐩=𝐤e𝐄t`$.
3. The gauge invariant functions $`\overline{G}`$ are given by
$`G(𝐩,t)=G(𝐤e𝐄t,t)=\overline{G}(𝐤,t)`$ (75)
$`=\overline{G}(𝐩+e𝐄t,t).`$ (76)
We shall make use of these rules in the following sections. In this procedure has been generalized for two - particle Greens functions and leads to the field - dependent Bethe - Salpeter - equation.
### C Equation for Wigner distribution
In the relative and center of mass coordinates the time diagonal part of (68) reads
$`{\displaystyle \frac{}{t}}`$ $`f(𝐩,t)=`$ (79)
$`{\displaystyle \underset{0}{\overset{tt_0}{}}}d\tau [\{G^>(𝐩,t{\displaystyle \frac{\tau }{2}},\tau ),\mathrm{\Sigma }^<(𝐩,t{\displaystyle \frac{\tau }{2}},\tau )\}`$
$`\{G^<(𝐩,t{\displaystyle \frac{\tau }{2}},\tau ),\mathrm{\Sigma }^>(𝐩,t{\displaystyle \frac{\tau }{2}},\tau )\}]`$
Here $`f(𝐩,t)=G^<(𝐩,𝐑,t,\tau =0)`$ denotes the Wigner distribution function and we suppress the center of mass coordinates. $`\{,\}`$ is the anti-commutator understood that the $`\tau `$ variable at the first place comes with a minus sign respectively. This equation is exact in time, but according to the assumed slowly varying space dependence we have used gradient expansion for space variables and dropped all R-dependence for simplicity. This criterion is discussed in the last section (198). With the help of the gauge invariant formulation of Green’s function (III B), we can write the kinetic equation (79) finally in the following gauge-invariant form
$`{\displaystyle \frac{}{t}}f(𝐤,t)+e𝐄_𝐤f(𝐤,t)={\displaystyle \underset{0}{\overset{tt_0}{}}}𝑑\tau `$ (81)
$`[\{G^>(𝐤{\displaystyle \frac{e𝐄}{2}}\tau ,\tau ,t{\displaystyle \frac{\tau }{2}}),\mathrm{\Sigma }^<(𝐤{\displaystyle \frac{e𝐄}{2}}\tau ,\tau ,t{\displaystyle \frac{\tau }{2}})\}_+`$ (82)
$`\{G^<(𝐤{\displaystyle \frac{e𝐄}{2}}\tau ,\tau ,t{\displaystyle \frac{\tau }{2}}),\mathrm{\Sigma }^>(𝐤{\displaystyle \frac{e𝐄}{2}}\tau ,\tau ,t{\displaystyle \frac{\tau }{2}})\}_+].`$ (83)
This kinetic equation is exact in time convolutions. This is necessary because gradient expansions in time are connected with linearization in electric fields and consequently fail . The gradient approximation in space has been applied assuming slow varying processes in space. This corresponds to the limit of a weakly coupled plasma, which we employed already in Section II. Please remind that due to Coulomb gauge we do not have space inhomogeneity by the electric field.
### D Spectral function
The spectral properties of the system are described by the Dyson equation for the retarded Green function. A free particle in a uniform electric field, where the field is represented by a vector potential $`𝐄(t)=\frac{1}{c}\stackrel{.}{𝐀}(t)`$ leads to the following equation
$$\left[i\mathrm{}\frac{}{t}ϵ(𝐩\frac{e}{c}𝐀(t))\right]G_0^R(𝐩,tt^{})=\delta (tt^{}).$$
(85)
This equation is easily integrated
$$G_0^R(𝐩,tt^{})=i\mathrm{\Theta }(tt^{})\mathrm{exp}\left[\frac{i}{\mathrm{}}\underset{t}{\overset{t^{}}{}}𝑑uϵ(𝐩\frac{e}{c}𝐀(u))\right].$$
(86)
For free particles and parabolic dispersions, the gauge invariant spectral function follows
$`A_0(𝐤,\omega )`$ $`=`$ $`2{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑\tau \mathrm{cos}\left(\omega \tau {\displaystyle \frac{k^2}{2m\mathrm{}}}\tau {\displaystyle \frac{e^2E^2}{24m\mathrm{}}}\tau ^3\right)`$ (87)
$`=`$ $`{\displaystyle \frac{2\pi }{ϵ_E}}Ai\left({\displaystyle \frac{k^2/2m\mathrm{}\omega }{ϵ_E}}\right)`$ (88)
where Ai(x) is the Airy function and $`ϵ_E=(\mathrm{}^2e^2E^2/8m)^{1/3}`$. It is instructive to verify that (87) satisfies the frequency sum rule $`𝑑\omega A_0(\omega )=2\pi `$. The interaction-free but field-dependent retarded Green’s function $`G_o^R`$ can be obtained from the interaction-free and field-free Green’s function by a simple Airy transformation . This is an expression of the fact that the solutions of the Schrödinger equation with constant electric field are Airy-functions. The retarded functions can therefore be diagonalized within those eigen-solutions. It can be shown that (87) remains valid even within a quasiparticle picture , where we have to replace simply the free dispersion $`k^2/2m`$ by the quasiparticle energy $`ϵ_k`$.
### E The Problem of the ansatz
In order to close the kinetic equation (79), it is necessary to know the relation between $`G^>`$ and $`G^<`$. This problem is known as an ansatz and must be constructed consistently with the required approximation of self-energy. Assuming the conventional KB ansatz we have a relation between the two time Green functions and the distribution function
$`G^<(𝐤,\omega ,𝐫,t)`$ $`=`$ $`A(𝐤,\omega ,𝐫,t)f(𝐤,𝐫,t)`$ (89)
$`G^>(𝐤,\omega ,𝐫,t)`$ $`=`$ $`A(𝐤,\omega ,𝐫,t)(1f(𝐤,𝐫,t)).`$ (90)
This is quite good as long as the quasi-particle picture holds and no memory effects play any role. As we shall see, the formulation of kinetic equations with high fields is basically connected with a careful formulation of retardation times. Therefore, the simple ansatz, called KB ansatz fails.
Another obscure discrepancy is the fact that with the old ansatz, one has some minor differences in the resulting collision integrals compared with the results from the density operator technique. With the old ansatz, one gets just one half of all retardation times in the various time arguments . This annoying discrepancy remained obscure until the work of Lipavsky, et al. where an expression is given for the $`G^<`$ function in terms of expansion after various times. We can write in Wigner coordinates
$`G^<(𝐩,t,\tau )`$ $`=`$ $`f(𝐩,t{\displaystyle \frac{|\tau |}{2}})A(𝐩,\tau ,t).`$ (91)
This generalized- Kadanoff- Baym (GKB) - ansatz is an exact relation as long as the selfenergy is taken in Hartree- Fock approximation. Together with the requirement of gauge invariance of Sec. III B and using the quasiparticle spectral function (87) with quasiparticle energies $`ϵ_k`$ instead of $`k^2/2m`$, the GKB ansatz finally reads
$`G^<(𝐤,\tau ,𝐫,t)=\mathrm{exp}\left\{{\displaystyle \frac{i}{\mathrm{}}}\left(ϵ_k\tau +{\displaystyle \frac{e^2E^2}{24m}}\tau ^3\right)\right\}`$ (92)
$`\times f(𝐤{\displaystyle \frac{e𝐄|\tau |}{2}},𝐫,t{\displaystyle \frac{|\tau |}{2}}).`$ (93)
In order to get more physical insight into this ansatz one transforms into the frequency representation
$`G^<(𝐤,\omega ,𝐫,t)=2{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑\tau f(𝐤{\displaystyle \frac{e𝐄\tau }{2}},t{\displaystyle \frac{\tau }{2}})`$ (94)
$`\times \mathrm{cos}\left(\omega \tau ϵ(𝐤,𝐫,t){\displaystyle \frac{\tau }{\mathrm{}}}{\displaystyle \frac{e^2E^2}{24m\mathrm{}}}\tau ^3\right).`$ (95)
Neglecting the retardation in $`f`$ one recovers the ordinary ansatz (89) with the spectral function (87). The generalized ansatz takes into account history by an additional memory. This ansatz is superior to the Kadanoff-Baym ansatz in the case of high external fields in several respects : (i) it has the correct spectral properties, (ii) it is gauge invariant, (iii) it preserves causality, (iv) the quantum kinetic equations derived with Eq.(112) coincide with those obtained with the density matrix technique , and (v) it reproduces the Debye-Onsager relaxation effect .
Other choices of ansatz can be appropriate for other physical situations. For a more detailed discussion see .
### F Kinetic equation in dynamically screened approximation
For Coulomb interaction it is unavoidable to consider screening if one does not want to obtain long range or short wave vector divergences. To obtain an explicit form for the kinetic equation we have to determine the selfenergy $`\mathrm{\Sigma }^{>,<}`$. The dynamically screened approximation is given by expressing the self energy by a sum of all ring diagrams. The resulting kinetic equation is the quantum Lenard - Balescu equation, which has been derived for high fields in . We give this approximation in exact time convolutions. The selfenergy is given in terms of the dynamical potential $`𝒱`$
$`\mathrm{\Sigma }_a^<(𝐤,t,t^{})={\displaystyle \frac{d𝐪}{(2\pi \mathrm{})^3}𝒱_{aa}^<(𝐪,t,t^{})G_a^<(𝐤𝐪,t,t^{})}`$ (96)
where the dynamical potential is expressed within Coulomb potentials $`V_{ab}(𝐪)`$
$`𝒱_{aa}^<(𝐪,t,t^{})={\displaystyle \underset{dc}{}}V_{ad}(𝐪)_{dc}^<(𝐪,t,t^{})V_{ca}(𝐪)`$ (98)
via the density-density fluctuation
$`_{ab}^<(𝐪,t,t^{})=\delta _{ab}{\displaystyle 𝑑\overline{t}𝑑\overline{\overline{t}}}`$ (99)
$`\times \left(^r\right)^1(𝐪,t,\overline{t})L_{aa}^<(𝐪,\overline{t},\overline{\overline{t}})\left(^a\right)^1(𝐪,\overline{\overline{t}},t^{}).`$ (100)
Here $`L`$ is the free density fluctuation
$`L_{aa}^<(𝐪,t,t^{})={\displaystyle \frac{d𝐩}{(2\pi \mathrm{})^3}G_a^<(𝐩,t,t^{})G_a^>(𝐩𝐪,t^{},t)}`$ (101)
and $`^{r/a}`$ the retarded/ advanced dielectric function
$`^{r/a}(𝐪,t,t^{})=\delta (tt^{})\pm i\mathrm{\Theta }[\pm (tt^{})]{\displaystyle \underset{b}{}}V_{bb}(𝐪)`$ (103)
$`\times (L^>(𝐪,t,t^{})L^<(𝐪,t,t^{})).`$ (104)
One easily convince oneself that this set of equations (96-104) is gauge invariant.
We can directly introduce this set of equations into the equation for the Wigner function (81) and obtain after some algebra for the in-scattering part of the collision integral
$`I_a^{\mathrm{in}}(𝐤,t)=2{\displaystyle \underset{b}{}}{\displaystyle \frac{d𝐪}{(2\pi \mathrm{})^3}V_{ab}^2(𝐪)\underset{0}{\overset{\mathrm{}}{}}𝑑\tau \frac{d\omega }{2\pi }}`$ (105)
$`\times \mathrm{cos}\left[(ϵ_{kq}^aϵ_k^a\omega )\tau +{\displaystyle \frac{e_a\mathrm{𝐄𝐪}\tau ^2}{2m_a}}\right]`$ (106)
$`\times f_a(𝐤𝐪e_a𝐄\tau ,t\tau )(1f_a(𝐤e_a𝐄\tau ,t\tau ))`$ (107)
$`\times {\displaystyle \frac{L_{bb}^<(𝐪,\omega ,t\frac{1}{2}\tau )}{\left|(𝐪,\omega ,t\frac{1}{2}\tau )\right|^2}}`$ (108)
with the free density fluctuation (101)
$`L_{bb}^<(𝐪,\omega ,t)=2{\displaystyle \frac{d𝐩}{(2\pi \mathrm{})^3}\underset{0}{\overset{\mathrm{}}{}}𝑑\tau }`$ (109)
$`\times \mathrm{cos}\left[(\omega ϵ_p^b+ϵ_{p+q}^b)\tau +{\displaystyle \frac{e_b\mathrm{𝐄𝐪}\tau ^2}{2m_b}}\right]`$ (110)
$`\times f_b(𝐩+𝐪,t{\displaystyle \frac{1}{2}}\tau )(1f_b(𝐩,t{\displaystyle \frac{1}{2}}\tau )).`$ (111)
The out-scattering term $`I^{\mathrm{out}}`$ is given by $`f1f`$. Here we used the ansatz (92) and have employed the approximation $`t\pm \frac{1}{2}\tau t`$ in the density fluctuation (100) which corresponds to a gradient approximation in times for the density fluctuations. Since the center of mass time dependence is carried only by the distribution functions in (100), this approximation is exact in the quasistationary case which we investigate in the next section. All internal time integrations remain exact. Of course, for time dependent phenomenae we have to question this approximation.
Eq. (108) represents the field dependent Lenard-Balescu kinetic equation which was here slightly rewritten and which form will turn out to be very convenient for the later analytical integration. Other standard approximations like the T-matrix approximation resulting into a field dependent Bethe-Salpeter equation can be given.
#### 1 Kinetic equation in statically screened approximation
Using the static approximation for the dielectric function $`(𝐪,0,t)`$ in (108), the kinetic equation for statically screened Coulomb potentials in high electric fields appears
$`{\displaystyle \frac{}{T}}f_a+e𝐄{\displaystyle \frac{}{𝐤_𝐚}}f_a={\displaystyle \underset{b}{}}I_{ab}`$ (112)
$`I_{ab}={\displaystyle \frac{2(2s_b+1)}{\mathrm{}^2}}{\displaystyle \frac{d𝐤_𝐚^{}d𝐤_𝐛d𝐤_𝐛^{}}{(2\pi \mathrm{})^6}\delta \left(𝐤_𝐚+𝐤_𝐛𝐤_𝐚^{}𝐤_𝐛^{}\right)}`$ (113)
$`\times \left\{f_a^{}f_b^{}(1f_a)(1f_b)f_af_b(1f_a^{})(1f_b^{})\right\}`$ (114)
$`\times V_s^2(𝐤_𝐚𝐤_𝐚^{}){\displaystyle \underset{0}{\overset{\mathrm{}}{}}}d\tau \mathrm{cos}\{(ϵ_a+ϵ_bϵ_a^{}ϵ_b^{}){\displaystyle \frac{\tau }{\mathrm{}}}`$ (115)
$`{\displaystyle \frac{𝐄\tau ^2}{2\mathrm{}}}({\displaystyle \frac{e_a𝐤_𝐚}{m_a}}+{\displaystyle \frac{e_b𝐤_𝐛}{m_b}}{\displaystyle \frac{e_a𝐤_𝐚^{}}{m_a}}{\displaystyle \frac{e_b𝐤_𝐛^{}}{m_b}})\}`$ (116)
with $`f_b=f_b(k_be_bE\tau ,T\tau )`$. The potential is the static Debye one
$`V_s(p)={\displaystyle \frac{4\pi e_ae_b\mathrm{}^2}{p^2+\mathrm{}^2\kappa ^2}}`$ (118)
and the static screening length $`\kappa `$ is given by
$`\kappa ^2={\displaystyle \underset{c}{}}{\displaystyle \frac{4\pi e_c^2n_c}{T_c}}`$ (119)
in the equilibrium and nondegenerated limit. Here $`T_c`$ is the temperature of specie $`c`$, charge $`e_c`$, spin $`s_c`$ and mass $`m_c`$ respectively.
If we had used the conventional Kadanoff and Baym ansatz (89) we would have obtained a factor $`1/2`$ in different retardations . This would lead to no relaxation effect at all . Furthermore it is assumed, that no charge or mass transfer will occur during the collision. Otherwise one would obtain an additional term in the $`\mathrm{cos}`$ \- function proportional to $`\tau ^3`$.
Two modifications of the usual Boltzmann collision integral can be deduced from (112): (i)A broadening of the $`\delta `$-distribution function of the energy conservation and an additional retardation in the center-of-mass times of the distribution functions. This is known as collisional broadening and is a result of the finite collision duration . This effect can be observed even if no external field is applied. It is interesting to remark that this collisional broadening ensures the conservation of the total energy . If this effect is neglected one obtains the Boltzmann equation for the field free case. (ii) The electric field modifies the broadened $`\delta `$\- distribution function considerably by a term proportional to $`\tau ^2`$. This broadening vanishes for identical charge to mass ratios of colliding particles. At the same time the momentum of the distribution function becomes retarded by the electric field. This effect is sometimes called intra-collisional-field effect.
## IV Field effects on current
We are now interested in corrections to the particle flux, and therefore obtain from (112) the balance equation for the momentum
$$\frac{}{t}<𝐤_𝐚>n_ae_a𝐄=\underset{b}{}<𝐤_𝐚I_B^{ab}>.$$
(120)
Here we search for the relaxation field (5) which will be represented as renormalization of the external field $`𝐄`$ similar to the Debye-Onsager-Relaxation field in the theory of electrolyte transport . This effect is a result of the deformation of the two-particle correlation function by an applied electric field.
To proceed we assume some important restrictions on the distribution functions. First, we assume a nondegenerate situation, such that the Pauli blocking effects can be neglected. Second, to calculate the current for a quasistationary plasma we choose Maxwellian distributions analog to (17)
$`f_i(p)={\displaystyle \frac{n_i}{2s_i+1}}\lambda _i^3\mathrm{exp}\left\{{\displaystyle \frac{p^2}{2m_iT_i}}\right\}`$ (121)
with the thermal wave length $`\lambda _i^2=2\pi \mathrm{}^2/(m_iT_i)`$, the spin $`s_i`$ and the partial temperature $`T_i`$ for species $`i`$ which can be quite different e.g. in a two - component system.
### A Statically screened result
Before we present the result for the dynamically screened approximation we want to give the static result. The momentum conservation in (112) can be carried out and we get for the relaxation field
$`n_ae_a{\displaystyle \frac{\delta E}{E}}𝐄={\displaystyle \underset{b}{}}(2s_a+1)(2s_b+1)`$ (122)
$`\times {\displaystyle \frac{2}{\mathrm{}^2}}{\displaystyle \frac{d𝐪d𝐐d𝐤}{(2\pi \mathrm{})^9}V_s^2(q)\underset{0}{\overset{\mathrm{}}{}}𝑑\tau (𝐤+e_a𝐄\tau )}`$ (123)
$`\times \mathrm{cos}[({\displaystyle \frac{q^2}{2m_a\mathrm{}}}+{\displaystyle \frac{\mathrm{𝐤𝐪}}{m_a\mathrm{}}}{\displaystyle \frac{\mathrm{𝐪𝐐}}{m_b\mathrm{}}})\tau {\displaystyle \frac{\mathrm{𝐄𝐪}}{2\mathrm{}}}[{\displaystyle \frac{e_b}{m_b}}{\displaystyle \frac{e_a}{m_a}}]\tau ^2]`$ (124)
$`\times \left\{f_a(𝐤)f_b\left(𝐐+{\displaystyle \frac{𝐪}{2}}\right)f_a(𝐤+𝐪)f_b\left(𝐐{\displaystyle \frac{𝐪}{2}}\right)\right\}`$ (125)
where we have shifted the retardation into the distribution functions. The second part of the distribution functions can be transformed into the first one by putting $`𝐤+𝐪𝐤`$ and $`𝐪𝐪`$ with the result
$`n_ae_a{\displaystyle \frac{\delta E}{E}}𝐄={\displaystyle \underset{b}{}}{\displaystyle \frac{2s_as_b}{\mathrm{}^2(2\pi \mathrm{})^9}}{\displaystyle 𝑑𝐤𝑑𝐪𝑑𝐐f_b(𝐐)f_a(𝐤)}`$ (126)
$`\times `$ $`V^2(q)𝐪{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}d\tau \mathrm{cos}[({\displaystyle \frac{q^2}{2\mu \mathrm{}}}{\displaystyle \frac{\mathrm{𝐤𝐪}}{m_a\mathrm{}}}+{\displaystyle \frac{\mathrm{𝐪𝐐}}{m_b\mathrm{}}})\tau `$ (128)
$`+{\displaystyle \frac{\mathrm{𝐄𝐪}}{2\mathrm{}}}[{\displaystyle \frac{e_b}{m_b}}{\displaystyle \frac{e_a}{m_a}}]\tau ^2]`$
with the reduced mass $`\mu ^1=1/m_a+1/m_b`$. The angular integrations can be carried out trivially and we get
$`n_ae_a{\displaystyle \frac{\delta E}{E}}𝐄={\displaystyle \frac{𝐄}{E}}{\displaystyle \underset{b}{}}I_1`$ (129)
$`I_1={\displaystyle \frac{1}{\mathrm{}^{11}4\pi ^6}}{\displaystyle 𝑑qq^3V^2(q)\underset{0}{\overset{\mathrm{}}{}}𝑑\tau \mathrm{js}\left(\frac{Eq}{2\mathrm{}}\left[\frac{e_b}{m_b}\frac{e_a}{m_a}\right]\tau ^2\right)}`$ (130)
$`\times \mathrm{sin}({\displaystyle \frac{q^2\tau }{2\mu \mathrm{}}})I_2[a]I_2[b]`$ (131)
with $`\mathrm{js}(x)=(x\mathrm{cos}x\mathrm{sin}x)/x^2`$. The two integrals over the distribution functions $`I_2`$ can be done with the result
$`I_2[a]`$ $`=`$ $`{\displaystyle \frac{\mathrm{}m_a(2s_a+1)}{q\tau }}{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑kkf_a(k)\mathrm{sin}({\displaystyle \frac{kq\tau }{m_a\mathrm{}}})`$ (133)
$`=`$ $`2\mathrm{}^3n_a\pi ^2\mathrm{e}^{\frac{q^2\tau ^2T_a}{2\mathrm{}^2m_a}}`$ (134)
and correspondingly $`I_2[b]`$. We now introduce the new variables
$`q`$ $`=`$ $`2y\sqrt{\mu T_{ab}}`$ (135)
$`t`$ $`=`$ $`{\displaystyle \frac{2T_{ab}\tau }{\mathrm{}}}`$ (136)
$`T_{ab}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{m_b}{m_a+m_b}}T_a+{\displaystyle \frac{m_a}{m_a+m_b}}T_b\right)`$ (137)
$`e`$ $`=`$ $`{\displaystyle \frac{\mathrm{}\sqrt{\mu }E}{4T_{ab}^{3/2}}}\left[{\displaystyle \frac{e_b}{m_b}}{\displaystyle \frac{e_a}{m_a}}\right]`$ (138)
and obtain
$`I_1={\displaystyle \frac{8n_an_b\mu ^2T_{ab}}{\pi ^2\mathrm{}^4}}{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑yy^3V^2(2y\sqrt{\mu T_{ab}})`$ (139)
$`{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑t\mathrm{js}(yt^2e)\mathrm{sin}(y^2t)\mathrm{e}^{y^2t^2}.`$ (140)
Using the screened Debye potential (118) we finally obtain
$`I_1`$ $`=`$ $`{\displaystyle \frac{8n_an_be_a^2e_b^2}{T_{ab}}}I_3`$ (142)
$`I_3`$ $`=`$ $`{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑z{\displaystyle \frac{z^3}{(z^2+1)^2}}{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑l\mathrm{js}(xzl^2){\displaystyle \frac{\mathrm{sin}(z^2l\zeta )}{\zeta }}\mathrm{e}^{z^2l^2}.`$ (143)
Therein we used $`y=z\zeta `$ and $`l=t\zeta `$ with the quantum parameter
$`\zeta ^2={\displaystyle \frac{\mathrm{}^2\kappa ^2}{4\mu T_{ab}}}`$ (145)
and the classical field parameter
$`x={\displaystyle \frac{e}{\zeta }}={\displaystyle \frac{E}{2T_{ab}\kappa }}({\displaystyle \frac{m_a}{m_a+m_b}}e_b{\displaystyle \frac{m_b}{m_a+m_b}}e_a).`$ (146)
With this form (142) we have given an extremely useful representation because the field effects, contained in $`x`$, are separated from the quantum effects, which are contained in $`\zeta `$. The integral in (142) can be performed analytically in the classical limit $`\zeta 0`$. For the more general quantum case with arbitrary $`\zeta `$ the linear and cubic field effect can be given analytically and are discussed in . We will not discuss them here.
Performing the classical limit $`\zeta 0`$ one obtains from (142) that
$`I_{3c}`$ $`=`$ $`{\displaystyle \frac{\pi x}{24}}F(|x|)`$ (147)
$`F(x)`$ $`=`$ $`{\displaystyle \frac{3}{x^2}}\left[3x+{\displaystyle \frac{1}{1+x}}{\displaystyle \frac{4}{x}}\mathrm{ln}(1+x)\right].`$ (148)
Introducing the classical result (148) into (142) we find from (129) and (120) the following relaxation field
$$\frac{}{t}<𝐤_𝐚>n_ae_a𝐄\left(1+\frac{\delta E_a}{E}\right)=n_ae_a𝐉R(E)$$
(149)
with
$`{\displaystyle \frac{\delta E_a}{E}}`$ $`=`$ $`{\displaystyle \frac{e_a\pi }{6\kappa }}{\displaystyle \underset{b}{}}{\displaystyle \frac{4n_be_b^2}{\mu _{ab}}}{\displaystyle \frac{\frac{e_b}{m_b}\frac{e_a}{m_a}}{\left(\frac{T_b}{m_b}+\frac{T_a}{m_a}\right)^2}}F(|x|)`$ (150)
and $`x`$ from (146). We see that for a plasma consisting of particles with equal charge to mass ratios, no relaxation field appears. The link to the known Debye- Onsager relaxation effect can be found if we assume that we have a plasma consisting of electrons ($`m_e,e_e=e`$) and ions with charge $`e_i=eZ`$ and temperatures $`T_e=T_i=T`$. Then (150) reduces to
$`{\displaystyle \frac{\delta E_a}{E}}`$ $`=`$ $`{\displaystyle \frac{\kappa e_a^2}{6T}}{\displaystyle \frac{Z(1+\frac{m_e}{m_i}Z)}{(1+Z)(1+\frac{m_e}{m_i})}}F\left({\displaystyle \frac{eE}{T\kappa }}{\displaystyle \frac{Z(1+\frac{m_e}{m_i}Z)}{1+\frac{m_e}{m_i}}}\right)`$ (151)
$`=`$ $`{\displaystyle \frac{e^2\kappa _e}{6T}}\{\begin{array}{c}\frac{1}{2}+o(E)\\ \\ \frac{3\kappa T}{2eE}+o(1/E)^2\end{array}\mathrm{for}Z=1.`$ (152)
This formula together with the general form (150) is the main result of this chapter. It gives the classical relaxation effect for statically screened approximation up to any field strength and represents a result beyond linear response. We see that in the case of single charged heavy ions the Debye result (1) is underestimated by a factor of two.
### B Dynamically screened result
The calculation of the current with the collision integral for dynamically screened potentials (108) can be performed analytically as well. For the quasistationary condition we can calculate the frequency integral in (108) analytically using the identity for the classical limit $`o(\mathrm{})`$
$`{\displaystyle \frac{d\omega }{2\pi }\frac{H(\omega )}{\omega }\mathrm{Im}^1(q,\omega )}={\displaystyle \frac{H(0)}{2}}\mathrm{Re}(1{\displaystyle \frac{1}{(q,0)}})`$ (153)
where we set $`H(\omega )=\omega /\mathrm{Im}`$ and which relation is proven in appendix A. We will employ only classical screening. The quantum result for screening is more involved and not yet analytically integrable.
Observing that for the dielectric function (104) together with (121) holds
$`\underset{\omega 0}{lim}{\displaystyle \frac{\omega }{\mathrm{Im}(q,\omega )}}={\displaystyle \frac{q^3}{\sqrt{\pi }\mathrm{}^3}}\left({\displaystyle \underset{b}{}}{\displaystyle \frac{\kappa _b^2}{v_b}}\right)^1`$ (155)
with the partial screening length $`\kappa _b^2=4\pi e_b^2n_b/T_b`$ and the partial thermal velocity $`v_b^2=2T_b/m_b`$, we obtain for the current (129) after similar integrations as in chapter IV A instead of (142)
$`I_1^{\mathrm{dyn}}={\displaystyle \frac{8\kappa ^2e_a^2e_b^2n_an_b\sqrt{m_am_b}}{\sqrt{\pi \mu _{ab}TT_aT_b}\underset{c}{}\frac{\kappa _c^2}{v_c}}}I_3^{\mathrm{dyn}}`$ (156)
$`I_3^{\mathrm{dyn}}={\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑z{\displaystyle \frac{z^2}{1+z^2}}{\displaystyle \underset{1}{\overset{1}{}}}𝑑xx{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑l𝑑l_1\mathrm{e}^{z^2(l^2+l_1^2)}`$ (157)
$`\times {\displaystyle \frac{1}{\zeta }}\mathrm{cos}[M_b\zeta lz^2+Bzl^2x]\mathrm{cos}[M_a\zeta l_1z^2Azl_1^2x].`$ (158)
Here we used the same dimensionless variables as in chapter IV A and the quantum parameter (145). Further we abbreviated $`A=\frac{e_aE}{\kappa T_a}`$, $`B=\frac{e_bE}{\kappa T_b}`$, $`M_a=\sqrt{\frac{2\mu T}{m_aT_a}}`$, $`M_b=\sqrt{\frac{2\mu T}{m_bT_b}}`$.
We like to remark that we neglect any field dependence on the screening $``$ itself here. As presented in a field dependent screening function can be derived. However, this field dependence gives rise to a field dependence starting quadratically and will be not considered in this work.
The classical limit of (156) can be performed again by $`\zeta 0`$. We obtain
$`I_3^{\mathrm{dyn}}={\displaystyle \frac{1}{2}}AM_aI[|A|,|B|](ab)`$ (159)
with the remaining 3-dimensional integral
$`I[A,B]={\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑z{\displaystyle \frac{z^3}{z^2+1}}{\displaystyle \underset{1}{\overset{1}{}}}𝑑x{\displaystyle \frac{x^2}{A^2x^2+z^2}}`$ (160)
$`{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑l\mathrm{e}^{z^2l^2}\mathrm{cos}(Bzl^2x).`$ (161)
#### 1 Linear response
The linear response can be read off directly from (159) and is given by $`I[0,0]`$ of (160). We obtain
$`I_3^{\mathrm{dyn}}={\displaystyle \frac{\pi ^{3/2}}{12}}(M_aAM_bB)`$ (162)
and the linear relaxation field (150) takes the form
$`{\displaystyle \frac{\delta E^{\mathrm{dyn}}}{E}}={\displaystyle \frac{4e\pi \kappa }{3\underset{c}{}\kappa _c^2\sqrt{\frac{m_c}{T_c}}}}{\displaystyle \underset{b}{}}n_be_b^2\sqrt{{\displaystyle \frac{m_am_b}{T_aT_b}}}`$ (163)
$`\times \left({\displaystyle \frac{e_a}{T_a^{3/2}\sqrt{m_a}}}{\displaystyle \frac{e_b}{T_b^{3/2}\sqrt{m_b}}}\right)+o(E).`$ (164)
The difference to (150) becomes more evident if we consider again only electrons and ions with equal temperature
$`{\displaystyle \frac{\delta E^{\mathrm{dyn}}}{E}}`$ $`=`$ $`{\displaystyle \frac{\kappa e^2}{6T}}{\displaystyle \frac{2Z(1+\sqrt{\frac{m_e}{m_i}}Z)}{(Z+\sqrt{\frac{m_e}{m_i}})}}+o(E).`$ (165)
The differences to (151) are obvious in the different mass dependence. This result overestimates the Debye result by a factor of two.
#### 2 Complete classical result
Now we are able to present a complete field dependence beyond linear response. The integral (160) can be done analytically, which is sketched in appendix B. The result reads
$`I[A,B]={\displaystyle \frac{\pi ^{3/2}}{6}}[A,B]`$ (166)
$`[A,B]={\displaystyle \frac{3}{2A^3}}[{\displaystyle \frac{4A(1\sqrt{1+B}}{B}}+{\displaystyle \frac{A^2+\mathrm{log}(1A^2)}{\sqrt{1+\frac{B}{A}}}}`$ (167)
$`+2({\displaystyle \frac{\mathrm{ArcTanh}(\frac{1}{\sqrt{1\frac{B}{A}}})\mathrm{ArcTanh}(\frac{\sqrt{1+B}}{\sqrt{1\frac{B}{A}}})}{\sqrt{1\frac{B}{A}}}}`$ (168)
$`+{\displaystyle \frac{\mathrm{ArcTanh}(\frac{1}{\sqrt{1+\frac{B}{A}}})+\mathrm{ArcTanh}(\frac{\sqrt{1+B}}{\sqrt{1+\frac{B}{A}}})}{\sqrt{1+\frac{B}{A}}}})].`$ (169)
We obtain for (163)
$`{\displaystyle \frac{\delta E_a^{\mathrm{dyn}}}{E}}={\displaystyle \frac{4e_a\pi \kappa }{3\underset{c}{}\kappa _c^2\sqrt{\frac{m_c}{T_c}}}}{\displaystyle \underset{b}{}}n_be_b^2\sqrt{{\displaystyle \frac{m_am_b}{T_aT_b}}}`$ (170)
$`\times \left({\displaystyle \frac{e_a}{T_a^{3/2}\sqrt{m_a}}}[A,B]{\displaystyle \frac{e_b}{T_b^{3/2}\sqrt{m_b}}}[B,A]\right).`$ (171)
Expanding (169) in powers of $`E`$ we recover (163). Once more we choose the case of electrons and ions with equal temperature and obtain
$`{\displaystyle \frac{\delta E^{\mathrm{dyn}}}{E}}`$ $`=`$ $`{\displaystyle \frac{\kappa e^2}{6ϵ_0T}}{\displaystyle \frac{2Z([A,B]+\sqrt{\frac{m_e}{m_i}}Z[B,A])}{(Z+\sqrt{\frac{m_e}{m_i}})}}.`$ (172)
For single charged ions and big mass differences we can further simplify to
$`{\displaystyle \frac{\delta E^{\mathrm{dyn}}}{E}}={\displaystyle \frac{\kappa e^2}{6T}}[{\displaystyle \frac{eE}{\kappa T}}]`$ (173)
$`={\displaystyle \frac{e^2\kappa _e}{6T}}\{\begin{array}{c}2+o(E)\\ \\ \frac{3\kappa T}{\sqrt{2}eE}+o(1/E)^2\end{array}`$ (174)
$`[x]={\displaystyle \frac{3}{x^3}}({\displaystyle \frac{2\left(2x+3\left(1+\sqrt{1+x}\right)\right)}{\sqrt{1+x}}}`$ (175)
$`+\sqrt{2}\left(\mathrm{ArcTanh}({\displaystyle \frac{1}{\sqrt{2}}})+\mathrm{ArcTanh}({\displaystyle \frac{\sqrt{1+x}}{\sqrt{2}}})\right)`$ (176)
$`+{\displaystyle \frac{x^2+\mathrm{log}(1x^2)}{\sqrt{2}}})=2+o(E).`$ (177)
This result will be compare with the statically screened result (151) and the hydrodynamical result (55) in section IV C. Here we remark already that the Debye result is twice overestimated here.
### C Thermally averaged dynamically screened result
We will now give an approximative treatment of the dynamical screening used in . This approximation consists into the replacement of the dynamical screening in the collision integral (108) which is $`\overline{(\omega ,q)^2}`$ by $`(1+\kappa ^2V_{aa}(q)/4\pi )^1`$. This represents a thermal averaging of $`^2`$ which can be proven easily with the help of appendix A. We obtain the relaxation effect of (150) and (151) but with a different field function $`F`$
$`F^{\mathrm{dyn}}(x)`$ $`=`$ $`{\displaystyle \frac{3}{x^2}}\left[2x{\displaystyle \frac{2}{x}}\mathrm{ln}(1+x)\right]`$ (178)
$`=`$ $`\{\begin{array}{c}2+o(x)\\ \\ \frac{3}{x}+o(1/x)^2\end{array}.`$ (179)
Therefore the relaxation effect (151) in linear response for single charged ions takes the form of (1) and twice the static screened result (151) and half of the dynamical screened result (173).
As we see from figure 1 the different approximations lead to very different results. The statically screened result (151) underestimates the Debye result by factor 2 which is corrected by the thermal averaged treatment of the screening. If we calculate instead the complete dynamical screened result (165) or (172) we obtain twice the Debye result (1) and the thermal averaged screened result. However there is a complete different charge dependence. We have to observe that the perfectly symmetric treatment of screening does not reproduce the hydrodynamical result which is the Onsager result (2) for linear response.
### D Asymmetric dynamical screened result
We want now to proceed and ask under which assumptions the Onsager result (2) might be reproduced. Following the results we saw from the hierarchy we have consequently to treat the electrons (specie a) and ions (all other species) asymmetrically. This we will perform in the same spirit as Onsager in that the ions have to be treated dynamical (as before) but the electrons are screened statically.
This means we consider as the potential not the bare Coulomb one but a statically screened Debye potential for specie a. The ions (all other species) will then form the dynamical screening. In comparison with the chapter before we can perform all steps analogously except two modifications, eq. (182) and (187). First we observe that instead of (155) we have now
$`\underset{\omega 0}{lim}{\displaystyle \frac{\omega }{\mathrm{Im}(q,\omega )}}={\displaystyle \frac{qv_b(q^2+\mathrm{}^2\kappa _e^2)}{\sqrt{\pi }\mathrm{}^3\kappa _b^2}}`$ (180)
which leads to a replacement of the sum
$`{\displaystyle \underset{c}{}}{\displaystyle \frac{\kappa _c^2}{v_c}}{\displaystyle \frac{\kappa ^2}{v_i}}`$ (181)
in the for-factor of (156) and (170). This leads in the limit of big mass differences to a for-factor in (165) and (172) respectively
$`\mathrm{modification}\mathrm{I}:{\displaystyle \frac{Z}{1+Z}}.`$ (182)
The second modification is that in (156) one has to replace
$`{\displaystyle \frac{z^2}{1+z^2}}`$ $``$ $`{\displaystyle \frac{z^2}{1+z^2}}{\displaystyle \frac{z^2}{q+z^2}}`$ (183)
$`=`$ $`{\displaystyle \frac{q}{q1}}{\displaystyle \frac{z^2}{q+z^2}}{\displaystyle \frac{1}{q1}}{\displaystyle \frac{z^2}{1+z^2}}`$ (184)
with
$`q={\displaystyle \frac{\kappa _a^2}{\kappa ^2}}.`$ (185)
This shows that in the end results (170), Eq. (172) has to be changed
$`\mathrm{modification}\mathrm{II}:`$ (186)
$`[A,B]{\displaystyle \frac{\sqrt{q}}{q1}}[{\displaystyle \frac{A}{\sqrt{q}}},{\displaystyle \frac{B}{\sqrt{q}}}]{\displaystyle \frac{1}{q1}}[A,B].`$ (187)
Particularly we obtain for the linear response result (165) where for electron-ion plasma $`q=1/(Z+1)`$ and
$`{\displaystyle \frac{\delta E^{\mathrm{asy}}}{E}}`$ $`=`$ $`{\displaystyle \frac{\delta E^{\mathrm{dyn}}}{E}}{\displaystyle \frac{Zq}{\sqrt{q}+1}}`$ (188)
$`=`$ $`{\displaystyle \frac{\kappa e^2}{3T}}{\displaystyle \frac{Zq}{\sqrt{q}+1}}+o(E)`$ (189)
which agrees with (48) if we consider that the mobilities are very different $`b_i/b_e0`$ in (49).
The same result we obtain from the thermally averaged result (148) since there appears no such function as (180) and therefore the modification I of (182) does not apply but solely the modification II of (187). We therefore obtain (151) but
$`F_{\mathrm{asy}}^{\mathrm{dyn}}(x)`$ $`=`$ $`2F^{\mathrm{dyn}}(x)\sqrt{2}F^{\mathrm{dyn}}(\sqrt{2}x)`$ (191)
$`=`$ $`\{\begin{array}{c}2\sqrt{2}+o(x)\\ \\ \frac{3}{2x}+o(1/x)^2\end{array}`$ (192)
with $`F^{\mathrm{dyn}}`$ of (179). The linear response leads then exactly to the same result as from the dynamical screening (188), i.e. the Onsager result with the same charge dependence.
The fact that we reproduce the classical Onsager result with the same charge dependence can be considered as very satisfactory. The more since we have seen how many different considerations are possible. Please note that the special case $`Z=1`$ could lead occasionally to a seemingly agreement between different treatments. We think that the charge dependence incriminates different treatments.
In figure 2 we see that the asymmetrical screened result (172) with (187) approaches the hydrodynamical or Onsager result (2) rather well for small fields while it is too low at high fields. On the other hand the thermally averaged symmetrical screened result (192) agrees with the hydrodynamical approximation (55) in the low and high field limit. Why the hydrodynamical result cannot be reproduced completely within the kinetic theory remains still a puzzle. Probably the remaining difference is due to the neglect of the field effect on the screening itself .
## V Range of applicability
During the derivation of the quantum kinetic equations there has been assumed the gradient approximations which restricted the spatial gradients of the system. Here we want to discuss up to which field strength this assumption is justified.
The electric field is limited to values $`x<<1`$ for $`x`$ from (146). This can be deduced from the expression for the dynamical screened result (173). The expression has a remove-able singularity at $`x=1`$. Therefore we see a smooth curve. Nevertheless this is the field strength where something is happening. For equal masses and temperatures of plasma components this condition translates into
$`E<{\displaystyle \frac{\kappa T}{e}}.`$ (193)
We interpret the occurrence of such singular point that no thermal distributions are pertained in the system. Then we have to take into account non-thermal field dependent distributions which have been employed to study nonlinear conductivity .
The condition (193) allows for different physical interpretations. Within the picture of the screening cloud we can rewrite (193) into
$`eE<m{\displaystyle \frac{v_{\mathrm{th}}^2}{r_D}}.`$ (194)
This means that a particle moving on the radius of the screening cloud $`r_D=1/\kappa `$ with thermal velocity $`v_{\mathrm{th}}^2=T/m`$ should not be pulled away by the acting field force. We can discuss this limit also via the energy density which can be reached in a system by the applied field. We can reformulate once more the condition (193) to find equivalently
$`{\displaystyle \frac{E^2}{4\pi }}<nT.`$ (195)
This means that we have essentially non-thermal effects to be expected if the energy density of the field becomes comparable with the thermal energy density.
The validity criterion (193) can now be used to check the weak space inhomogeneity which has been assumed during our calculation. Quasi- equilibrium in charged systems with external fields can only be assumed if the field current is accompanied by an equivalent diffusion current
$`j_{\mathrm{field}}=e\mu En=j_{\mathrm{diff}}=eD{\displaystyle \frac{dn}{dx}}`$ (196)
using the Einstein condition $`\mu =eD/T`$ one gets
$`eE=T{\displaystyle \frac{1}{n}}{\displaystyle \frac{dn}{dx}}.`$ (197)
Combining this elementary consideration with our condition (193) we obtain a limitation for space gradients
$`{\displaystyle \frac{dn}{d(\kappa x)}}<n`$ (198)
where our treatment of field effects and local equilibrium is applicable.
## VI Summary
The nonlinear relaxation field of a charged system under the influence of high electric fields is investigated. The local equilibrium or hydrodynamical approach starting from the classical BBGKY hierarchy is compared with the results from the quantum kinetic equations.
We come to the same conclusion considering the hydrodynamical approximation or the kinetic theory that a perfectly symmetric two component plasma will lead to a different relaxation effect than the case where we consider the moving charge asymmetrically from the screening surrounding. In the hydrodynamic approach this has been achieved by friction with a background, in the kinetic approach we have realized it due to asymmetric screening. Within this asymmetric treatment the limit to a one component plasma which would be to set the ion charge to $`Z=1`$ leads to a non-vanishing finite quantity. Oppositely in the perfectly symmetrical treatment this limit vanishes in that the relaxation field vanishes as it should. The perfectly symmetrical treatment of species in the system leads to twice the Debye result different from the Onsager result in linear response.
Different approximations of the kinetic approaches are compared and discussed. We found that the symmetrical treatment of species as well as the asymmetrical treatment leads to the same corresponding results as the hydrodynamical approach for linear response. But for higher field strengths there appear minor differences which are probably due to the neglect of the field dependent screening itself. The thermally averaged approximation of screening has the advantage to agree for low and high fields with the hydrodynamical of local equilibrium approach.
###### Acknowledgements.
The author wants to thank warmly Jens Ortner who has contributed the hydrodynamical approach with background. Also many comments regarding the manuscript are gratefully acknowledged. Last but not least Gerd Röpke is thanked for stimulating intellectual disagreements which have provoked this work.
## A Integrals over dielectric functions
Here we proof a very useful relation, which has been given in . Therefore we consider the following integral including the dielectric function
$`I`$ $`=`$ $`{\displaystyle \frac{d\omega }{2\pi }\frac{H(\omega )}{\omega }\mathrm{Im}ϵ^1(\omega )}`$ (A1)
$`=`$ $`{\displaystyle \frac{d\omega }{4\pi i}(\frac{1}{\omega +i\eta }+\frac{1}{\omega i\eta })H(\omega )(f^{}f^+)}`$ (A2)
where $`f^+=11/ϵ`$ and $`f^{}=(f^+)^{}`$. In the following we will assume that the function $`H(\omega )`$ is analytical and vanishes with $`\omega ^2`$ for large $`\omega `$. Since $`f^\pm (\omega )`$ has no poles in the lower/upper half plane we have the identity
$`{\displaystyle \frac{d\omega }{2\pi i}H(\omega )\frac{f^\pm (\omega )}{(\omega \pm i\eta )}}=f^\pm (0)H(0)`$ (A3)
and all other combinations of $`f^\pm `$ with the denominator vanish. If we would use the quantum dielectric function $`ϵ`$ we would have to add the residue of the poles at the Matsubara frequencies. Because we calculate only with the classical dielectric function we can use (A3) With the help of the relation (A3) we compute easily for (A1)
$`I={\displaystyle \frac{1}{2}}H(0)\mathrm{Re}\left(1{\displaystyle \frac{1}{ϵ(0)}}\right)`$ (A4)
which proves relation (153).
## B An Integral
Here we calculate the integral (160)
$`I[a,b]`$ $`=`$ $`{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑z{\displaystyle \frac{z^3}{z^2+1}}{\displaystyle \underset{1}{\overset{1}{}}}𝑑x{\displaystyle \frac{x^2}{a^2x^2+z^2}}`$ (B1)
$`\times `$ $`{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑l\mathrm{e}^{z^2l^2}\mathrm{cos}(bzl^2x).`$ (B2)
The variable substitutions $`lp`$ by $`p=\sqrt{z}l`$, $`xz`$ by $`z=yx`$ and $`pe`$ by $`p\sqrt{x}=l`$ leads to
$`I[a,b]`$ (B3)
$`=2{\displaystyle \underset{0}{\overset{1}{}}}𝑑x{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑y{\displaystyle \frac{y^{5/2}}{y^2x^2+1}}{\displaystyle \frac{x^3}{a^2+y^2}}{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑e\mathrm{e}^{ye^2}\mathrm{cos}(be^2)`$ (B4)
$`={\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑y{\displaystyle \frac{y^{1/2}}{a^2+y^2}}(1{\displaystyle \frac{\mathrm{log}(1+y^2)}{y^2}}){\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑e\mathrm{e}^{ye^2}\mathrm{cos}(be^2)`$ (B5)
where the trivial $`x`$\- integration has been carried out. The variable substitution $`e>q`$ by $`\sqrt{y}e=q`$ and $`yz`$ by $`y=1/z`$ leads to
$`I[a,b]={\displaystyle \frac{1}{a^2}}{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑q\mathrm{e}^{q^2}{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑y\mathrm{cos}(bq^2z){\displaystyle \frac{1z^2\mathrm{log}(1+\frac{1}{z^2})}{z^2+\frac{1}{a^2}}}.`$ (B7)
Now we proceed and use an integral calculated in the next subsection B 1
$`{\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}𝑑y\mathrm{e}^{icy}{\displaystyle \frac{1y^2\mathrm{log}(1+\frac{1}{y^2})}{y^2+\frac{1}{a^2}}}=2\pi {\displaystyle \underset{0}{\overset{1}{}}}𝑑xx^2{\displaystyle \frac{\mathrm{e}^{cx}}{\frac{1}{a^2}x^2}}`$ (B9)
$`+a\pi \mathrm{e}^{c/a}(1+{\displaystyle \frac{\mathrm{log}(1a^2)}{a^2}})`$ (B10)
to obtain for (B7)
$`I[a,b]`$ $`=`$ $`{\displaystyle \frac{\pi ^{3/2}}{4a}}{\displaystyle \frac{1+\frac{\mathrm{log}(1a^2)}{a^2}}{\sqrt{b/a}}}`$ (B12)
$`+{\displaystyle \frac{\pi ^{3/2}}{2a^2}}{\displaystyle \underset{0}{\overset{1}{}}}𝑑x{\displaystyle \frac{x^2}{(\frac{1}{a^2}x^2)\sqrt{1+bx}}}.`$
The last integrals is trivial and we end up with (169).
### 1 Another Integral
Our task remains now to solve the integral
$`I={\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}𝑑y\mathrm{e}^{icy}{\displaystyle \frac{1y^2\mathrm{log}(1+\frac{1}{y^2})}{y^2+\frac{1}{a^2}}}.`$ (B13)
Because the complex function $`\mathrm{log}(1+1/y^2)`$ has a cut from $`(0,i)`$ we perform the integration along the path as depicted in figure 3 and write
$`{\displaystyle \underset{R}{\overset{r}{}}}+{\displaystyle \underset{r}{\overset{R}{}}}+C_R+C_r+{\displaystyle \underset{I}{}}+{\displaystyle \underset{II}{}}`$ (B14)
$`=2\pi i\mathrm{Res}[{\displaystyle \frac{1y^2\mathrm{log}(1+\frac{1}{y^2})}{y^2+\frac{1}{a^2}}},i/a]`$ (B15)
$`=\pi a\mathrm{e}^{c/a}(1+{\displaystyle \frac{\mathrm{log}(1a^2)}{a^2}}).`$ (B16)
It is now easy to proof that in the limit $`r0`$ and $`R\mathrm{}`$ the integration parts $`C`$ vanish. Since the first two parts of (B16) represent just the desired integral $`I`$ we have to calculate
$`{\displaystyle \underset{I}{}}+{\displaystyle \underset{II}{}}`$ $`=`$ $`{\displaystyle \underset{i+r}{\overset{r}{}}}𝑑y\mathrm{e}^{icy}{\displaystyle \frac{1y^2\mathrm{log}(1+\frac{1}{y^2})}{y^2+\frac{1}{a^2}}}`$ (B18)
$`+{\displaystyle \underset{r}{\overset{ir}{}}}𝑑y\mathrm{e}^{icy}{\displaystyle \frac{1y^2\mathrm{log}(1+\frac{1}{y^2})+2\pi i}{y^2+\frac{1}{a^2}}}`$
$`=`$ $`2\pi {\displaystyle \underset{0}{\overset{1}{}}}𝑑x{\displaystyle \frac{x^2\mathrm{e}^{cx}}{\frac{1}{a^2}x^2}}.`$ (B19)
Using (B19) and (B16) we obtain just (B10).
|
warning/0005/hep-ph0005076.html
|
ar5iv
|
text
|
# 1 Two representative choices for the mass parameters affecting the sgoldstino effective couplings. All masses are expressed in GeV.
1. Introduction
Sgoldstinos are the supersymmetric partners of the goldstino $`\stackrel{~}{G}`$, and, in supersymmetric models with a very light gravitino , are expected to be part of the effective theory at the weak scale. In the simplest case, the R–odd goldstino is a gauge singlet, and its R–even superpartners are two neutral spin–0 particles, $`S`$ (CP–even) and $`P`$ (CP–odd): in the following, we will use the generic symbol $`\varphi `$ to denote any of the two sgoldstino states. If the sgoldstino masses $`m_\varphi `$ and the supersymmetry-breaking scale $`\sqrt{F}`$ are not too large, <sup>1</sup><sup>1</sup>1We remind the reader that the supersymmetry–breaking scale $`\sqrt{F}`$ can be put in one–to–one correspondence with the gravitino mass $`m_{3/2}`$, via the relation $`F=\sqrt{3}m_{3/2}M_P`$, where $`M_P(8\pi G_N)^{1/2}2.4\times 10^{18}\mathrm{GeV}`$ is the Planck mass. sgoldstino production and decay may be detectable at the present collider energies.
Previous studies of sgoldstino phenomenology at colliders were performed under the very restrictive assumption of negligible sgoldstino masses. In a recent paper , we started a systematic investigation of the possible phenomenological signatures of massive sgoldstinos, concentrating on $`e^+e^{}`$ colliders and in particular on LEP. We showed that the study of processes such as $`e^+e^{}\gamma \varphi `$, $`Z\varphi `$ or $`e^+e^{}\varphi `$, followed by $`\varphi `$ decaying into two gluon jets, can explore virgin land in the parameter space of models with a superlight gravitino. This was confirmed by a subsequent experimental analysis , where (using not only the two-gluon decay mode, but also the two-photon one, as suggested by previous LEP searches ) new stringent combined bounds on the gravitino and sgoldstino masses were derived.
In the present work, we extend the study of the phenomenology of massive sgoldstinos to the case of hadron colliders, with emphasis on the Tevatron ($`p\overline{p}`$; $`\sqrt{S}=1.8\mathrm{TeV}`$ and $`L100\mathrm{pb}^1`$ in run I; $`\sqrt{S}=2\mathrm{TeV}`$ and $`L2\mathrm{fb}^1`$ in run II). Our study is organized as follows. In sect. 2, we complete the discussion of the sgoldstino effective couplings and branching ratios given in , to encompass a range of sgoldstino masses extending above the $`t\overline{t}`$ threshold. In sect. 3, we discuss the mechanisms for sgoldstino production at hadron colliders, giving explicit formulae for the relevant partonic cross-sections and showing some representative numerical results. In sect. 4, we review the resulting signals at the Tevatron and we attempt a first discussion of the discovery potential of the different channels, leaving a more detailed analysis to our experimental colleagues. We show that inclusive production of sgoldstinos, followed by their decay into two photons, can lead to observable signals or to stringent combined bounds on the gravitino and sgoldstino masses. The sgoldstino decay mode into two gluon jets, dominant over most of the parameter space but plagued by large backgrounds, may provide a useful complementary signature when gluinos are much heavier than the electroweak gauginos and higgsinos. Associated production with an electroweak gauge boson $`(\gamma ,W,Z)`$ and/or other decay modes $`(\gamma Z,ZZ,WW,\stackrel{~}{G}\stackrel{~}{G},t\overline{t})`$ do not lead in general to an increased sensitivity.
To conclude this introduction, we would like to alert the reader on some simplifying assumptions underlying our analysis, and on the analogies between sgoldstinos and other hypothetical spin–0 particles, such as the neutral Higgs bosons of the Standard Model (SM) or its Minimal Supersymmetric extension (MSSM). As in , we will perform our study assuming that there is no mixing between sgoldstinos and Higgs bosons, and that R–odd MSSM particles and Higgs bosons <sup>2</sup><sup>2</sup>2We remind the reader that, in models with a very light gravitino, the MSSM upper bound on the mass of the lightest Higgs boson can be grossly violated . are sufficiently heavy not to play a rôle in sgoldstino production and decay. This should be taken as a benchmark case, which may be further generalized by including an exhaustive treatment of the interplay between $`SU(2)\times U(1)`$ and supersymmetry breaking . Whenever possible, however, we will compare the properties of our ‘pure’ sgoldstinos with the properties of the ‘pure’ SM or MSSM Higgs bosons: in the case of non–negligible mixing, we may expect intermediate properties. In this respect, sgoldstinos represent a motivated and well-defined addition to a long list of other hypothetical spin–0 particles (‘bosonic Higgs’, ‘coloron’, …) that have been proposed with various theoretical motivations, and are often used in data analyses to parametrize some of the searches for new physics. Since the indirect evidence for the SM (or the MSSM) as the correct and complete theory at the weak scale is far from conclusive , we believe that more exotic possibilities such as sgoldstinos still deserve to be taken seriously, even on purely phenomenological grounds. On the theoretical side, it is interesting to notice that sgoldstinos bear some similarities with the spin–0 fields arising from the metric (and antisymmetric tensors) of extra space-time dimensions in some ‘brane-world’ scenarios , whose collider phenomenology has been recently discussed . These similarities could be investigated more closely, leading perhaps to a more unified picture, if progress were made in the discussion of the breaking of supersymmetry and of the electroweak symmetry in this context.
2. Sgoldstino effective couplings and branching ratios
The general theoretical framework for the discussion of sgoldstino phenomenology was reviewed in , to which we refer the reader. Sgoldstino effective interactions and branching ratios were also discussed in , with emphasis on the sgoldstino mass range kinematically accessible at LEP, $`m_\varphi \stackrel{<}{_{}}200\mathrm{GeV}`$. It was shown that the sgoldstino couplings to gauge boson pairs can be parametrized by the supersymmetry-breaking scale $`\sqrt{F}`$, by the gaugino masses $`(M_3,M_2,M_1)`$ and by a mass parameter $`\mu _a`$, associated with the charged higgsino and the antisymmetric combination of neutral higgsinos. To extend the discussion to higher sgoldstino masses, which may be phenomenologically relevant at the Tevatron (or eventually at the LHC), we need only a more detailed discussion of the $`t\overline{t}\varphi `$ effective couplings and of $`\varphi t\overline{t}`$ decays.
As discussed in , the Yukawa couplings of sgoldstinos to leptons and light quarks (for our purposes, all but the top quark) are expected to be suppressed by a factor $`m_f/\sqrt{F}`$, where $`m_f`$ is the fermion mass. We then expect these couplings to be important only for the top quark, and parametrize them as follows:
$$_{\varphi t\overline{t}}=\frac{m_tA_S}{\sqrt{2}F}(Stt^c+\mathrm{h}.\mathrm{c}.)\frac{m_tA_P}{\sqrt{2}F}(iPtt^c+\mathrm{h}.\mathrm{c}.),$$
(1)
where we used two-component spinors in the notation of , and $`A_S,A_P`$ are free parameters with the dimension of a mass. Since $`A_S+A_P=2A_t`$, where $`\delta m^2=m_tA_t`$ is the off-diagonal element in the mass matrix for the stop squarks, we expect $`A_S`$ and $`A_P`$ to be of the order of the other supersymmetry-breaking masses in the strongly interacting sector of the theory. Notice that the $`St\overline{t}`$ coupling is identical in form to the coupling of the SM Higgs, and can be obtained from the latter by performing the substitution
$$\frac{g}{2m_W}\frac{A_S}{\sqrt{2}F}.$$
(2)
Similarly, the $`Pt\overline{t}`$ coupling can be obtained from the coupling of the SM Higgs by inserting, in four-component notation, a $`\gamma _5`$ matrix in the fermion bilinear, and by performing the substitution
$$\frac{g}{2m_W}\frac{iA_P}{\sqrt{2}F}.$$
(3)
Alternatively, the $`Pt\overline{t}`$ coupling can be also obtained from the $`At\overline{t}`$ coupling of the MSSM by performing the substitution of eq. (3) and by making the choice $`\mathrm{tan}\beta =1`$. Notice also that, in general, we can have $`A_SA_P`$. In the following, however, we will make the simplifying assumption $`A_S=A_PA_t`$, and, whenever needed for numerical examples, the representative choice $`A_t=M_3`$. From eq. (1) we find
$$\mathrm{\Gamma }(St\overline{t})=\frac{3m_t^2A_S^2m_S}{16\pi F^2}\left(1\frac{4m_t^2}{m_S^2}\right)^{3/2},\mathrm{\Gamma }(Pt\overline{t})=\frac{3m_t^2A_P^2m_P}{16\pi F^2}\left(1\frac{4m_t^2}{m_P^2}\right)^{1/2}.$$
(4)
Notice the different exponent for the CP-even and the CP-odd state, as in the MSSM.
We are now ready to extend the discussion of the sgoldstino branching ratios to the mass range of interest at the Tevatron. We will focus our attention on the dependences on $`m_\varphi `$ and $`\sqrt{F}`$, by making for the remaining parameters the two representative choices given in Table 1.
The corresponding chargino and neutralino masses (in GeV) are approximately $`(220,380)`$ and $`(175,240,385)`$ for case (a), $`(270,430)`$ and $`(260,350,440)`$ for case (b). There is of course a fourth neutralino, the symmetric higgsino combination, whose mass is controlled by an independent parameter $`\mu _s`$, and does not affect the sgoldstino couplings. For most values of $`\sqrt{F}`$ to be considered in the following, the two parameter choices of Table 1 should be comfortably compatible with the present experimental limits on R-odd supersymmetric particles, coming from LEP and Tevatron searches. For sufficiently large values of the sgoldstino masses, we should also consider sgoldstino decays into pairs of neutralinos, charginos or gluinos. However, the relevant couplings are controlled not only by the parameters of Table 1, but also by $`\mu _s`$, by the Higgs boson masses and by other parameters not related with the spectrum . In the following, we will always assume that the corresponding branching ratios can be safely neglected. In the region of sgoldstino masses considered in this paper, the only kinematically allowed decays could be those into the lightest neutralinos and charginos. For a wide range of the remaining parameters, these decays are sufficiently suppressed by the phase space and other factors, so that our discussion of the dominant decay modes remains a good approximation.
Since all the partial widths for two–body decays are proportional to $`F^2`$, the dependence on $`F`$ drops out of the discussion of the $`\varphi `$ branching ratios. The latter are shown in Figs. 1 and 2, as functions of $`m_\varphi `$, for the two cases of $`S`$ and $`P`$ and for the two parameter choices of Table 1.
We can see that the differences in the couplings of $`S`$ and $`P`$ to the massive weak bosons and to the top quark do not have an important impact on the branching ratios. For both $`S`$ and $`P`$, the most important decay mode is the one into gluons, with the one into goldstinos becoming competitive only for very large sgoldstino masses. In the whole mass range up to 1 TeV, decays into electroweak bosons are suppressed, at the level of $`10\%`$ or less: as we will see, however, these modes could still be relevant for hadron collider searches, because of the much smaller backgrounds of the resulting signals.
As already discussed in , another important quantity is the total $`\varphi `$ width, $`\mathrm{\Gamma }_\varphi `$, controlled by the ratios between the relevant mass parameters and the supersymmetry-breaking scale. Large values of these ratios correspond to broad, strongly coupled sgoldstinos: to keep the particle interpretation and the validity of our approximations, we must require, among the other things, $`\mathrm{\Gamma }_\varphi /m_\varphi 1`$. To compare signal and background in the narrow-width approximation, we must impose more stringent constraints: $`\mathrm{\Gamma }_\varphi /m_\varphi <10^1`$ when discussing decays into jets, $`\mathrm{\Gamma }_\varphi /m_\varphi <\mathrm{few}\times 10^2`$ when discussing decays into photons. We then show in Fig. 3 contours corresponding to constant values of $`\mathrm{\Gamma }_\varphi /m_\varphi `$ in the $`(m_\varphi ,\sqrt{F})`$-plane, for the two parameter choices of Table 1. Since the curves for $`\varphi =S`$ and $`\varphi =P`$ are almost indistinguishable, we draw both of them simultaneously.
As we will see in sects. 3 and 4, while the region of parameter space of experimental interest at LEP is such that sgoldstinos can always be treated as very narrow resonances, this is not necessarily the case at the Tevatron.
As a final remark on the branching ratios, we remind the reader that our computations have been performed at leading order. In analogy with the case of the SM and MSSM neutral Higgs bosons, we expect the NLO QCD corrections to increase the partial width into gluons by a factor $`K1.6`$ . Since the two-gluon decay mode is the dominant one, this may lead to a non-negligible suppression of the rarer decay modes. When important, this will be taken into account in the phenomenological discussion of sect. 4.
3. Production cross-sections
The possible parton-level mechanisms for sgoldstino production at hadron colliders are very similar to those for the SM Higgs and can be easily classified:
$$g+g\varphi ,$$
(5)
$$q+\overline{q}V+\varphi ,$$
(6)
$$q+\overline{q}q+\overline{q}+\varphi ,$$
(7)
$$g+g\mathrm{or}q+\overline{q}t+\overline{t}+\varphi ,$$
(8)
where $`q`$ denotes any quark flavour and $`V`$ any of the electroweak vector bosons, including the photon. We briefly discuss here the corresponding cross-sections, giving some numerical examples for the Tevatron. In the case of the SM Higgs, the dominant Higgs production mechanism at the Tevatron is the one of eq. (5), with those of eqs. (6) and (7) suppressed by roughly one order of magnitude, and the one of eq. (8) suppressed by roughly an additional order of magnitude in the mass region of interest. An important rôle in these hierarchies is played by the fact that the SM Higgs couplings to $`WW`$ and $`ZZ`$ arise at the tree-level and are unsuppressed, whereas the couplings to $`gg`$ (and $`\gamma \gamma `$) arise at the one-loop level, and are therefore considerably suppressed. This compensates in part the hierarchy between strong and electroweak interactions, and allows for the subdominant production mechanisms to be of phenomenological interest. In contrast with the SM Higgs, all sgoldstino couplings to vector boson pairs are essentially on the same footing. The result, which could already be guessed from our study of the branching ratios, is the following: the practical relevance of the subdominant production mechanisms at hadron colliders will be much smaller in the case of sgoldstinos than in the case of the SM and MSSM neutral Higgs bosons.
To produce numerical examples, we will always adopt the CTEQ5 parametrization of the parton distribution functions with $`\mathrm{\Lambda }_5=226`$ MeV, corresponding to $`\alpha _S(m_Z)=0.118`$. Our results are summarized in Fig. 4, which displays some of the lowest-order sgoldstino production cross-sections, as functions of the sgoldstino mass, for $`p\overline{p}`$ collisions at $`\sqrt{S}=2\mathrm{TeV}`$. The cross-sections of Fig. 4 have been obtained for $`\sqrt{F}=1\mathrm{TeV}`$. Since they are all proportional to $`1/F^2`$, the cross-sections for any other value of $`\sqrt{F}`$ can be obtained by a simple rescaling of the values of Fig. 4. Cases (a) and (b) correspond as usual to the two representative parameter choices of Table 1. For simplicity, only the case of $`\varphi =S`$ has been considered. On the scale of Fig. 4, the results for the case $`\varphi =P`$ are very similar.
Gluon-gluon fusion
Since the sgoldstino couplings to light quarks are suppressed by the corresponding quark masses (as it is the case for the SM Higgs), whilst the effective gluon-gluon-sgoldstino couplings are proportional to $`M_3/F`$, the dominant production mechanism of massive sgoldstinos at hadron colliders is in general gluon-gluon fusion, eq. (5). To lowest order, the partonic cross-section can be expressed in terms of the gluonic width of the sgoldstino,
$$\sigma (gg\varphi )=\sigma _0m_\varphi ^2\delta (sm_\varphi ^2),\sigma _0=\frac{\pi ^2}{8m_\varphi ^3}\mathrm{\Gamma }(\varphi gg),$$
(9)
where $`\sqrt{s}`$ is the total energy in the centre-of-mass frame of the incoming partons. With the lowest-order expression for $`\mathrm{\Gamma }(\varphi gg)`$ given in , we find
$$\sigma _0=\frac{\pi M_3^2}{32F^2}.$$
(10)
The lowest-order proton-antiproton cross-section is then, in the narrow-width approximation,
$$\sigma (p\overline{p}\varphi )=\sigma _0\tau _\tau ^1\frac{dx}{x}f_g(x,m_\varphi ^2)f_g(\tau /x,m_\varphi ^2),$$
(11)
where $`\tau =m_\varphi ^2/S`$, $`\sqrt{S}`$ is the total centre-of-mass energy of the proton-antiproton system, and $`f_g(x,Q^2)`$ is the gluon distribution function, the same for proton and antiproton. An identical formula holds for the proton-proton cross-section. Notice that the above formulae are very similar in form to those for the production of a light SM Higgs boson. The only difference is that for the Higgs
$$\sigma _0=\frac{G_F\alpha _s^2}{288\sqrt{2}\pi }.$$
(12)
For a heavy Higgs, one must correct for the form factor originating from the top-quark loop and for the finite Higgs width, whilst the present approximation should be also applicable to heavy sgoldstinos, provided that $`m_\varphi M_3/F\stackrel{<}{_{}}1`$. The formal analogy with Higgs production also allows to include the NLO QCD corrections, by adapting the computation of . For our present purposes, it is sufficient to work at the leading non-trivial order. In the phenomenological discussion of sect. 4, however, we will include the most important QCD corrections by making the rough approximation $`\sigma _{NLO}=K\sigma `$, with $`K2`$.
Associated $`\varphi \gamma `$ production
In analogy with the process $`e^+e^{}\varphi \gamma `$, already considered in , we can consider the process $`q\overline{q}\varphi \gamma `$. This process is made possible by the existence of effective $`\gamma \gamma \varphi `$ and $`\gamma Z\varphi `$ couplings, whose explicit form can be found in . They are controlled by the ratios $`M_{\gamma \gamma }/F`$ and $`M_{\gamma Z}/F`$, respectively, where $`M_{\gamma \gamma }=M_1\mathrm{cos}^2\theta _W+M_2\mathrm{sin}^2\theta _W`$ and $`M_{\gamma Z}=(M_2M_1)\mathrm{sin}\theta _W\mathrm{cos}\theta _W`$ . At the partonic level and before including QCD corrections, there are only two Feynman diagrams to compute, corresponding to s-channel $`\gamma `$ and $`Z`$ exchange. Neglecting both the quark masses and the $`Z`$ width, the differential cross-section for a given quark flavour $`q`$ is
$$\frac{d\sigma }{d\mathrm{cos}\theta }\left(q\overline{q}\varphi \gamma \right)=\frac{|\mathrm{\Sigma }_{\varphi \gamma }|^2s}{64\pi N_cF^2}\left(1\frac{m_\varphi ^2}{s}\right)^3\left(1+\mathrm{cos}^2\theta \right),$$
(13)
where $`N_c=3`$ is a color factor,
$$|\mathrm{\Sigma }_{\varphi \gamma }|^2=\frac{e^2Q_q^2M_{\gamma \gamma }^2}{2s}+\frac{g_Z^2(v_q^2+a_q^2)M_{\gamma Z}^2s}{2(sm_Z^2)^2}+\frac{eQ_qg_Zv_qM_{\gamma \gamma }M_{\gamma Z}}{(sm_Z^2)},$$
(14)
$`v_q=T_{3q}/2Q_q\mathrm{sin}^2\theta _W`$, $`a_q=T_{3q}/2`$, $`T_{3u}=T_{3c}=T_{3t}=T_{3d}=T_{3s}=T_{3b}=1/2`$, $`Q_u=Q_c=Q_t=+2/3`$, $`Q_d=Q_s=Q_b=1/3`$, $`g_Z=e/(\mathrm{sin}\theta _W\mathrm{cos}\theta _W)`$, and $`\theta `$ is the scattering angle in the centre-of-mass frame of the colliding partons.
Associated $`\varphi Z`$ production
Another process to be considered is $`q\overline{q}\varphi Z`$, analogous to the process $`e^+e^{}\varphi Z`$ considered in . For $`\varphi =P`$, the differential cross-section is
$$\frac{d\sigma }{d\mathrm{cos}\theta }\left(q\overline{q}PZ\right)=\frac{|\mathrm{\Sigma }_{PZ}|^2}{32\pi N_cs^2F^2}\sqrt{(sm_P^2m_Z^2)^24m_P^2m_Z^2},$$
(15)
where
$$|\mathrm{\Sigma }_{PZ}|^2=\left(\frac{e^2Q_q^2M_{\gamma Z}^2}{2s}+\frac{g_Z^2(v_q^2+a_q^2)M_{ZZ}^2s}{2(sm_Z^2)^2}+\frac{eQ_qg_Zv_qM_{\gamma Z}M_{ZZ}}{(sm_Z^2)}\right)\left(t^2+u^22m_P^2m_Z^2\right),$$
(16)
and $`(t,u)`$ are the usual Mandelstam variables for two-body scattering. The cross-section for $`\varphi =S`$ has some additional complications, because, as discussed in , the $`SZZ`$ coupling has an additional dependence on the higgsino mass parameter $`\mu _a`$:
$$\frac{d\sigma }{d\mathrm{cos}\theta }\left(q\overline{q}SZ\right)=\frac{|\mathrm{\Sigma }_{SZ}|^2}{32\pi N_cs^2F^2}\sqrt{(sm_S^2m_Z^2)^24m_S^2m_Z^2},$$
(17)
where
$`|\mathrm{\Sigma }_{SZ}|^2`$ $`=`$ $`[{\displaystyle \frac{e^2Q_q^2M_{\gamma Z}^2}{2s}}+{\displaystyle \frac{g_Z^2(v_q^2+a_q^2)M_{ZZ}^2s}{2(sm_Z^2)^2}}+{\displaystyle \frac{eQ_qg_Zv_qM_{\gamma Z}M_{ZZ}}{sm_Z^2}}]\left[t^2+u^2+2m_Z^2(2sm_S^2)\right]`$ (18)
$`+`$ $`{\displaystyle \frac{g_Z^2\mu _a^2m_Z^4(v_q^2+a_q^2)}{(sm_Z^2)^2}}\left(2sm_S^2+{\displaystyle \frac{tu}{m_Z^2}}\right)`$
$`+`$ $`{\displaystyle \frac{g_Z\mu _am_Z^2}{sm_Z^2}}\left[{\displaystyle \frac{g_Z(v_q^2+a_q^2)M_{ZZ}}{sm_Z^2}}+{\displaystyle \frac{eQ_qM_{\gamma Z}v_q}{s}}\right]\left[2s(s+m_Z^2m_S^2)\right].`$
Associated $`\varphi W`$ production
At hadron colliders, we can also consider the associated production of a sgoldstino and a $`W`$ boson, whose partonic cross-section can easily be obtained from the previous ones. For $`\varphi =P`$:
$$\frac{d\sigma }{d\mathrm{cos}\theta }\left(q\overline{q}^{}PW\right)=\frac{|\mathrm{\Sigma }_{PW}|^2}{32\pi N_cs^2F^2}\sqrt{(sm_P^2m_W^2)^24m_P^2m_W^2},$$
(19)
where
$$|\mathrm{\Sigma }_{PW}|^2=\frac{g^2|V_{qq^{}}|^2M_2^2s}{8(sm_W^2)^2}\left(t^2+u^22m_P^2m_W^2\right).$$
(20)
As in the case of $`\varphi Z`$ production, the cross-section for $`\varphi =S`$ has some additional complications, because of the additional dependence of the $`SW^+W^{}`$ coupling on the higgsino mass parameter $`\mu _a`$:
$$\frac{d\sigma }{d\mathrm{cos}\theta }\left(q\overline{q}^{}SW\right)=\frac{|\mathrm{\Sigma }_{SW}|^2}{32\pi N_cs^2F^2}\sqrt{(sm_S^2m_W^2)^24m_S^2m_W^2},$$
(21)
where
$`|\mathrm{\Sigma }_{SW}|^2`$ $`=`$ $`{\displaystyle \frac{g^2|V_{qq^{}}|^2}{4(sm_W^2)^2}}[{\displaystyle \frac{M_2^2s}{2}}[t^2+u^2+2m_W^2(2sm_S^2)]`$ (22)
$`+`$ $`\mu _a^2m_W^4(2sm_S^2+{\displaystyle \frac{tu}{m_W^2}})+2\mu _am_W^2M_2s(s+m_W^2m_S^2)].`$
Vector-boson fusion
The cross-sections for sgoldstino production via vector-boson fusion can be easily calculated, starting from the effective sgoldstino couplings to $`\gamma \gamma `$, $`\gamma Z`$, $`ZZ`$ and $`WW`$. However, their complete analytic expressions are quite involved, and, in analogy with the associated production, also the production via vector boson fusion turns out to be significantly suppressed with respect to the production via gluon-gluon fusion. For these reasons, we omit here a detailed discussion of this production mechanism.
Associated $`t\overline{t}\varphi `$ production
Since the $`t\overline{t}\varphi `$ couplings can be obtained from the Higgs couplings of the SM (or of the MSSM) by the simple rescalings of eqs.(2) and (3), the simplest way to obtain the cross-sections for the associated production of a sgoldstino and a top-antitop pair is to take the corresponding Higgs cross-sections in the SM (or in the MSSM) and to rescale them by the appropriate factor.
Other production mechanisms
As discussed in , the effective lagrangian contains some interaction terms that can also lead to the pair-production of a CP–even and a CP–odd sgoldstino, with cross-section
$$\frac{d\sigma }{d\mathrm{cos}\theta }(q\overline{q}SP)=\frac{(\stackrel{~}{m}_q^4+\stackrel{~}{m}_{q^c}^4)}{512\pi N_cs^2F^4}\left[(sm_S^2m_P^2)^24m_S^2m_P^2\right]^{3/2}\mathrm{sin}^2\theta ,$$
(23)
where $`\theta `$ is the scattering angle in the center-of-mass frame. For plausible values of the parameters, we expect this cross-section to be suppressed by the large numerical factor and the higher power of the supersymmetry-breaking scale at the denominator. Otherwise, the corresponding signal could be seen as an anomaly in the four-jet sample.
To conclude this section, we comment again on the relative importance of the different production mechanisms. We can see from Fig. 4 that, for the parameter choices of Table 1, the dominant sgoldstino production mechanism at the Tevatron is by far gluon-gluon fusion. The processes of eqs. (6) and (7) are suppressed by roughly four orders of magnitude, and the one of eq. (8) by roughly two more orders of magnitude. Therefore, we shall consider only inclusive signals when discussing the phenomenology at the Tevatron.
4. Phenomenological discussion
The inclusive di-jet signal
Since the most important mechanism for sgoldstino production at the Tevatron is gluon-gluon fusion, and the two-gluon decay mode is the dominant one, it is natural to consider a peak in the dijet invariant mass distribution as a possible signal to be looked for. As a first element to measure the Tevatron sensitivity, we draw in Fig. 5 contours of constant $`\sigma (p\overline{p}\varphi +X)\times BR(\varphi gg)`$ in the $`(m_\varphi ,\sqrt{F})`$ plane, for $`\sqrt{S}=1.8\mathrm{TeV}`$ and the two parameter choices of Table 1. Since the curves for $`\varphi =S`$ and $`\varphi =P`$ are almost indistinguishable, we consider only the case $`\varphi =S`$. To account approximately for the NLO QCD corrections , we have multiplied the LO result by a K–factor $`K=2`$.
To interpret the curves of Fig. 5, we need some additional information on the SM backgrounds and on the experimental sensitivity. Fortunately, we can rely on two recent analysis by CDF and D0 , devoted to the search of new particles decaying to dijets, and thus applicable to sgoldstinos. Using $`106\mathrm{pb}^1`$ of data collected at $`\sqrt{S}=1.8\mathrm{TeV}`$, and requiring that both jets have pseudorapidity $`|\eta |<2.0`$ and a scattering angle in the dijet center-of-mass frame $`|\mathrm{cos}\theta ^{}||\mathrm{tanh}[(\eta _1\eta _2)/2]|<2/3`$, CDF plots and tabulates the $`95\%`$ c.l. upper limit on the cross-section times branching ratios for narrow resonances ($`\mathrm{\Gamma }/M<0.1`$) decaying into dijets. D0 uses $`104\mathrm{pb}^1`$, requires $`|\eta |<1.0`$ and $`|\eta _1\eta _2|<1.6`$, and plots a $`95\%`$ c.l. upper limit on the cross-section times branching ratio times acceptance for three representative models of new physics. We have used the published CDF and D0 data and our calculation of the sgoldstino cross-sections and branching ratios to draw in Fig. 5 a sequence of fat dots, representing our tentative estimate of the Tevatron sensitivity after run I. This line was obtained by selecting, for each value of the sgoldstino mass (from 200 to 600 GeV, in steps of 50 GeV), the more stringent of the CDF and D0 limits on $`\sqrt{F}`$, obtained under the assumption $`\mathrm{\Gamma }_\varphi /m_\varphi <0.1`$. By comparing with Fig. 3, however, we can see that for increasing sgoldstino masses this assumption is more and more strongly violated. As a result, we expect to have made a stronger and stronger overestimate of the present bounds for increasing mass values in the region above 300 GeV.
To have an idea of the Tevatron sensitivity after run II, for each given value of the sgoldstino mass we can rescale the corresponding value of $`\sqrt{F}`$ by a factor $`20^{1/8}1.5`$. This amounts to making the naïve assumptions that the cross-section does not vary much when changing $`\sqrt{S}`$ from $`1.8`$ to $`2.0`$ TeV, and that the error in the cross–section measurement will scale as $`1/\sqrt{L}`$, where $`L`$ is the integrated luminosity.
The inclusive di-photon signal
In the case of the SM and MSSM neutral Higgs bosons, the two-photon decay mode has a very suppressed branching ratio, at most $`𝒪(10^3)`$ in the mass region between 100 and 150 GeV, and much smaller for larger masses. As a result, the diphoton signal is marginal for Higgs searches at the Tevatron, and it is much more convenient to exploit other decay modes, in conjunction with associated production mechanisms. In contrast, sgoldstinos have diphoton branching ratios well above $`𝒪(10^2)`$ over the whole mass range between a few and 1000 GeV, and possibly larger inclusive production cross-sections, so we can expect the inclusive diphoton signal to play a major rôle in sgoldstino searches at the Tevatron.
As a first piece of information, we display in Fig. 6 contours of constant $`\sigma (p\overline{p}\varphi +X)\times BR(\varphi \gamma \gamma )`$ in the $`(m_\varphi ,\sqrt{F})`$ plane, for $`\sqrt{S}=1.8\mathrm{TeV}`$ and the two parameter choices of Table 1. Since the curves for $`\varphi =S`$ and $`\varphi =P`$ are almost indistinguishable, we consider only the case $`\varphi =S`$. In this case, the NLO QCD corrections are not included, since they have two competing effects that approximately cancel: an increase in the inclusive production cross-section, but also a similar increase in the two-gluon partial width, which correspondingly reduces the two-photon branching ratio.
To interpret the curves of Fig. 6, we need some additional information on the SM backgrounds and on the experimental sensitivity. Again, we can rely on some recent analysis by CDF and D0 , as well as on some useful information contained in a recent study for run II . The CDF study uses $`100\mathrm{pb}^1`$ of data collected at $`\sqrt{S}=1.8\mathrm{TeV}`$, and requires that both photons have pseudorapidity $`|\eta |<1`$ and transverse energy $`E_T>22\mathrm{GeV}`$. Since the study looks specifically at the invariant mass distribution of diphotons above 50 GeV, it is relatively simple to translate its results into a tentative sensitivity curve in the $`(m_\varphi ,\sqrt{F})`$ plane. The D0 studies were performed with slightly different goals, and some work would be required in order to optimize their data analysis for setting limits on sgoldstinos. On the contrary, the study in , performed with the cuts $`|\eta |<2`$ and $`E_T>20\mathrm{GeV}`$, comes with an analytical formula for the background and can be easily used to obtain a first, rough estimate of the run II sensitivity. For example, following , we can assume a diphoton identification efficiency $`ϵ=0.8`$ and combine the finite sgoldstino width and the experimental resolution in the diphoton invariant mass by defining a quantity $`[\mathrm{\Delta }/(1\mathrm{GeV})]=\sqrt{[\mathrm{\Gamma }_\varphi /(1\mathrm{GeV})]^2+(0.35)^2[m_\varphi /(1\mathrm{GeV})]}`$. We can then ask that, if $`S`$ is the number of signal events and $`B`$ the number of background events in a window of width $`1.2\mathrm{\Delta }`$ centered around $`m_\varphi `$, either $`S/\sqrt{B}>5`$ (if $`B>1`$) or $`S>5`$ (if $`B<1`$). Doing so, and leaving a more detailed and reliable analysis to our experimental colleagues, we could draw in Fig. 6 two series of fat dots and diamonds, representing our tentative estimates of the Tevatron sensitivities after run I and after run II, respectively. A more sophisticated study could proceed along the lines of and make full use of the double differential cross-section in the diphoton invariant mass and in the scattering angle, to take into better account the finite-width effects (important for large sgoldstino masses) and to reduce the dependence on the cuts (important for small sgoldstino masses). For our present purposes, however, it is sufficient to observe that, for the two parameter choices of Table 1, searching for the diphoton signal is by far the most powerful way of constraining the sgoldstino parameter space. <sup>3</sup><sup>3</sup>3The range of accessible sgoldstino masses may extend beyond the right border of Fig. 6. To reliably explore that region, however, we should take into account the effects of sgoldstino decays into neutralinos, charginos and eventually gluinos.
Whilst the searches for very massive sgoldstinos should be relatively straightforward, an interesting phenomenological question is how to extend and maximize the Tevatron reach in the region of small masses, $`m_\varphi 100\mathrm{GeV}`$. In this respect, it may be useful to relax as much as possible the trigger and selection requirements on the photon transverse energy. Also, in the region of very small masses, the associated production mechanisms with electroweak gauge bosons may play a useful role.
Acknowledgements. We would like to thank A. Brignole, A. Castro, G. Landsberg, M.L. Mangano, M. Spira, A.S. Turcot and D. Wood for useful discussions and suggestions.
|
warning/0005/nucl-th0005063.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Currently there is much interest in subthreshold production of heavy mesons in relativistic nuclear collisions. It is expected that measurements will reveal the change of the properties of mesons in dense and hot nuclear matter. Such predictions are made from theoretical reasons and, indeed, hints to a dropping mass of anti-kaons have been found by comparing recent measurements with calculations based on transport models . For $`\varphi `$ meson production there are only a few measurements in the reactions of Ni+Ni at 1.93 A$``$GeV and Ru+Ru at 1.69 A$``$GeV. Although only a limited amount of the phase space was accessible in these experiments, extrapolations to the full phase space point to a surprisingly large production cross section.
First estimates using transport models based on two-particle cross sections, derived in a simple meson exchange model , seem to underestimate the $`\varphi `$ multiplicity. One should however take in to account the lack of precise knowledge concerning elementary $`\pi `$$`\mathrm{\Delta }`$, $`\mathrm{\Delta }`$N and $`\mathrm{\Delta }`$$`\mathrm{\Delta }`$ collisions. In addition multiparticle processes could contribute. Experimentally, such multi-step processes can be investigated by accompanying cluster emission in meson production .
In this work we study whether three-body collisions could remarkably contribute to the production of $`\varphi `$ mesons in proton-nucleus and heavy-ion collisions. At threshold such a mechanism could be dominant because less bombarding energy is required if the projectile nucleon interacts with two target nucleons instead with a single one. For ’free’ nucleons the threshold reduces from 2.6 GeV to 1.8 GeV.
The consideration of $`\varphi `$ production is interesting since also the K<sup>-</sup> production proceeds considerably through an intermediate $`\varphi `$ meson. In the near-threshold proton-proton collision roughly half of the K<sup>-</sup> mesons come from the $`\varphi `$ decay . K<sup>-</sup> mesons have been measured by the KaoS collaboration at the heavy-ion synchrotron SIS of GSI Darmstadt in proton-Au collisions for a bombarding energy of 3.5 GeV , and the ANKE collaboration envisages the K<sup>-</sup> measurements for collisions of protons on various light targets at the cooler synchrotron COSY of FZ Jülich . For a firm understanding of these reactions a necessary prerequisite is the theoretical understanding of the role of the $`\varphi `$ channels.
We are going to calculate the cross section for the reaction p+2N $``$ $`\varphi `$ \+ 3N from the invariant amplitude based on tree-level diagrams. In refs. several diagrams were explored thoroughly for the case of the two-nucleon $`\varphi `$ production, i.e. N+N $``$ $`\varphi `$+2N. As shown there, the available data can be described by diagrams with internal meson conversion in a $`\pi `$$`\rho `$$`\varphi `$ vertex. Based on this investigation we restrict ourselves to such types of diagrams as illustrated in Fig. 1. Three nucleons in states $`N_1`$, $`N_2`$, $`N_3`$ act together via boson exchanges. Literally in the first diagram, after the interaction of particles 1 and 2, the intermediate fermion $`X`$ or pion $`\pi _2`$ has gathered sufficient energy to produce the $`\varphi `$ meson interacting with the third nucleon. In the right hand diagram the time ordering is exchanged.
Similar processes are described within the current transport models by sequential two-step processes which also allow to accumulate the energy of several nucleons. In this case, however, the intermediate particles are real, i.e. they are moving on-shell. Recently also effort is made to implement properly the particles’ off-shell propagation (see ref. ). In the diagrams displayed in Fig. 1 the intermediate particles $`X`$ and $`\pi _2`$ can be either on shell or off shell depending on the initial and final momenta of the particles. There are subspaces in the phase space, where the intermediate particles move on shell. It is our aim to investigate these two effects and to relate them to the standard treatment as sequential two-step processes.
Many-body collisions have already been treated several times in the literature. In ref. this mechanism was applied to analyse the cumulative backward emission of nucleons. In this region of the phase-space off-shell contributions become important. Many-body collisions were also incorporated in the framework of transport models . The results of these investigations show moderate effects; here we extend these studies to very subthreshold reactions for the $`\varphi `$ production.
## 2 Elementary cross section
To calculate the cross section for three-particle collisions we will apply the one-boson exchange model which is often used to parametrize Lorentz invariant cross sections. Our approach is strictly based on tree-level diagrams with effective parameters adjusted to experimental data. For our investigations we restrict ourselves to the diagrams displayed in Fig. 1. We also consider two further diagram types, where the $`\rho `$ and the $`\pi _2`$ mesons are interchanged. For the sake of simplicity in the present exploratory investigation we employ only the nucleon propagator for the intermediate particle $`X`$ and the pion exchange for the interaction between nucleons. Under these propositions the whole set of diagrams we are considering consists of 4 $`\times `$ 36 individual diagrams, where the second factor denotes the number of permutations of the respective three fermion lines in the initial and final states.
In ref. the interaction Lagrangians for these diagrams have been presented which read in standard notation
$`_{\pi NN}`$ $`=`$ $`{\displaystyle \frac{f_{\pi NN}}{m_\pi }}\overline{\psi }\gamma ^5\gamma ^\mu \stackrel{}{\tau }\psi _\mu \stackrel{}{\pi },`$ (1)
$`_{\rho NN}`$ $`=`$ $`g_{\rho NN}\overline{\psi }\gamma ^\mu \stackrel{}{\tau }\psi \stackrel{}{\rho }_\mu {\displaystyle \frac{f_{\rho NN}}{2m_N}}\overline{\psi }\sigma ^{\mu \nu }\stackrel{}{\tau }\psi _\mu \stackrel{}{\rho }_\nu ,`$ (2)
$`_{\pi \rho \varphi }`$ $`=`$ $`{\displaystyle \frac{f_{\pi \rho \varphi }}{m_\varphi }}ϵ_{\mu \nu \alpha \beta }^\mu \varphi ^\nu ^\alpha \stackrel{}{\rho }^\beta \stackrel{}{\pi }.`$ (3)
These equations contain the nucleon field $`\psi `$, the iso-vector meson fields $`\stackrel{}{\pi }`$, $`\stackrel{}{\rho }^\mu `$ and the iso-scalar field $`\varphi ^\mu `$. The Dirac matrices, the Levi-Civita symbol and the corresponding particle masses are denoted by $`\gamma ^\mu `$, $`ϵ_{\mu \nu \alpha \beta }`$ and $`m_i`$, respectively. Greek indices stand for Lorentz indices, and we put $`\mathrm{}=c=1`$. Furthermore, a formfactor is needed for each internal line of an exchanged meson $`i`$ coming from or entering a vertex $`(jkl)`$. We adopt here the standard monopole form factor as a function of the square of the transferred four-momentum $`q_i^\mu `$ and a cut-off parameter $`\mathrm{\Lambda }_{jkl}^i`$:
$`F(q_i^2)={\displaystyle \frac{(\mathrm{\Lambda }_{jkl}^i)^2m_i^2}{(\mathrm{\Lambda }_{jkl}^i)^2q_i^2}}.`$ (4)
The following values of the coupling constants and cut-off parameters have been used: $`f_{\pi NN}=0.989`$, $`g_{\rho NN}=3.718`$, $`f_{\rho NN}=6.1g_{\rho NN}`$, $`f_{\pi \rho \varphi }=1.04`$; $`\mathrm{\Lambda }_{\pi NN}^\pi `$ = 1.6 GeV, $`\mathrm{\Lambda }_{\rho NN}^\rho `$ = 1.3 GeV, $`\mathrm{\Lambda }_{\pi \rho \varphi }^\pi `$ = 0.95 GeV, $`\mathrm{\Lambda }_{\pi \rho \varphi }^\rho `$ = 1.2 GeV. Most of these values are taken from refs. and have been successfully applied in the near-threshold region. For $`\varphi `$ meson production at an energy of 82 MeV above the production threshold we obtain a cross section of 0.04 $`\mu `$b using these parameters. This value is smaller than the recently measured value . However, in ref. it has been demonstrated that the agreement between data and calculation is improved by including in addition the $`\varphi `$ meson bremsstrahlung from external nucleon legs and the final state interaction.
The cross section for $`\varphi `$ meson production by an incoming nucleon with momentum $`p_1=p_{lab}`$ hitting a nucleon with momentum $`p_2`$ which is surrounded by another nucleon with momentum $`p_3`$ in a medium of density $`n_0`$ reads
$`\sigma ={\displaystyle \frac{1}{4m_Np_{lab}}}{\displaystyle \frac{n_0}{2m_N}}{\displaystyle \frac{1}{2}}{\displaystyle \frac{1}{3!}}{\displaystyle \frac{1}{(2\pi )^8}}{\displaystyle \frac{1}{8}}{\displaystyle \underset{projections}{}}{\displaystyle 𝑑𝐋_{inv}\underset{diagrams}{}𝒯^2}.`$ (5)
The first quotient contains the incoming flux with the laboratory momentum $`p_{lab}`$. The second quotient is related to the density $`n_0`$ of the second collision partner; the fact that we calculate the cross section at a single nucleon accounts for the following factor 1/2. The factorial $`3!`$ occurs because the outgoing nucleons are indistinguishable. The first sum (including $`1/8`$) averages over the initial spin-projection quantum-numbers and runs over all outgoing spin and isospin states while the second sum contains the $`4\times 36`$ diagrams mentioned above which are numerically evaluated including all interference terms. The integration in Eq. (5) is carried out over the Lorentz invariant phase space spanned by the momenta of the outgoing particles, $`d𝐋_{inv}=_fd^3p_f/(2p_f^0)\delta ^4(p_fp_i)`$.
## 3 On-shell problem
In some regions of the phase space the intermediate particle $`X`$ can become on-shell and the propagator gets singular. In the following we discuss first the situation for the pion propagator $`\pi _2`$ in Fig. 1 which in a large part of the phase space becomes on shell. The standard procedure to avoid the singularity is to introduce into the propagator an imaginary part $`im\mathrm{\Gamma }`$. That means that one writes the matrix element of Fig. 1 as
$`𝒯=\stackrel{~}{T_1}{\displaystyle \frac{1}{p^2m^2+im\mathrm{\Gamma }}}\stackrel{~}{T_2},`$ (6)
where the quantities $`\stackrel{~}{T_1}`$ and $`\stackrel{~}{T_2}`$ describe the two subprocesses 1+2 $``$ 4+5+$`\pi `$ and $`\pi `$+3 $``$ 6+$`\varphi `$, which are connected by the propagator of the pion with mass $`m=m_\pi `$ and four-momentum $`p^\mu `$. The physical reason for the occurrence of the imaginary part is that the intermediate particle does not propagate in vacuum but in a dense medium instead. Then it can move only a finite distance of the mean free path $`\lambda `$ during its life time $`\tau `$ which is related to the imaginary part in the denominator via $`\mathrm{\Gamma }=1/\tau `$ (see refs. ). The same method is applied if unstable particle are treated in the entrance channels .
However, the on-shell propagation is in fact a process which is already treated in standard transport models as two consecutive two-step processes, where the particle $`X`$ is treated as a real on-shell particle. Now we try to separate the genuine three-body processes out of the contributions given by the whole propagator in Eq. (6).
If the particles were in vacuum the amplitude $`𝒯`$ would also be given by Eq. (6), but the quantity $`m\mathrm{\Gamma }`$ would be replaced by $`\epsilon `$ with Feynman’s prescription $`\epsilon +0`$. The square of this matrix element diverges since it is proportional to the square of the function $`\delta (p^2m^2)`$ in the limit $`\epsilon 0`$. Since the origin (see e.g. ) of the $`\delta `$ function is the integration over the time the pion moves, one can replace it with $`\tau \delta (p^2m^2)/(2\pi m)`$, where the time $`\tau `$ is connected with $`\epsilon `$ via $`\tau =m/\epsilon `$. Thus, the transition probability is proportional to the proper time $`\tau `$ in the rest system of the intermediate particle. This treatment is analogous to the standard derivation of the cross section (5) which becomes proportional to the space-time volume.
Due to the terms coming from the exchange of the initial and final momenta the matrix $`𝒯`$ becomes on shell in different regions of the phase space spanned by the four final-state momenta. In general there are 3 $`\times `$ 3 subspaces defined by $`(p_ap_b)^2=m^2`$, where $`p_a`$ and $`p_b`$ are the total momenta of the three possible initial pairs $`a`$ =\[(12),(23),(31)\] and three final pairs $`b`$ = \[(4,5),(56),(64)\], respectively.
In the near neighbourhood of each subspace we divide the $`𝒯`$ matrix in a smooth part and a singular one,
$`𝒯=\overline{𝒯}+T_1{\displaystyle \frac{1}{p^2m^2+i\epsilon }}T_2.`$ (7)
Contrary to Eq. (6) the amplitudes $`T_1`$ and $`T_2`$ describe now subprocesses with free (on-shell) particles. In calculating the phase space integral we treat the square of the second term as
$`\sigma ^{}`$ $`=`$ $`\sigma _{twostep}+\sigma _{three}^{},`$ (8)
$`\sigma _{twostep}`$ $`=`$ $`C{\displaystyle 𝑑𝐋_{inv}\frac{\pi \tau }{m}\delta (p^2m^2)|\widehat{T}|^2},`$ (9)
$`\sigma _{three}^{}`$ $`=`$ $`C\underset{\epsilon 0}{lim}{\displaystyle 𝑑𝐋_{inv}|\widehat{T}|^2\left[\frac{(p^2m^2)^2}{[(p^2m^2)^2+\epsilon ^2]^2}\frac{\pi }{2\epsilon }\delta (p^2m^2)\right]},`$ (10)
where $`\widehat{T}`$ = $`T_1T_2`$. The quantity $`C`$ comprises the factors standing in front of the integral in Eq. (5). These expressions arise by splitting the propagator in Eq. (7) into the $`\delta `$ function and its principal value, while the last term $`\sigma _{three}^{}`$ is constructed such that it is finite as we will see below.
We identify the first part (9) with the two-step process as it is proportional to the time of flight of the intermediate particle which can only be fixed by additional information on the collision geometry. This part can be transformed into an expression describing a sequence of two two-nucleon collisions. Introducing the identity $`d^4p\delta ^4(p_1+p_2p_4p)=1`$ and using the definition of the two-particle cross sections we obtain
$`\sigma _{twostep}={\displaystyle 𝑑𝐩_\pi \frac{d\sigma }{d𝐩_\pi }(𝐩_1,𝐩_2)\sigma _\varphi (𝐩_\pi ,𝐩_3)(n_0\widehat{v}\tau )}.`$ (11)
Now the cross section has been factorized in the cross sections $`d\sigma /d𝐩_\pi `$ for pion production and $`sigma_\varphi `$ for $`\varphi `$ production via pion absorption by the nucleon 3. The sum over the final subspaces $`b`$ has cancelled the factorials in the definitions of the cross sections. In principle one should sum Eq. (11) over the three subspaces $`a`$, however, we have used only the first one in accordance to Eq. (5). The last factor in the round bracket contains the relative velocity $`\widehat{v}=(p_\pi p_3)v_{rel}`$, $`v_{rel}=\sqrt{(1(mm_N/(p_\pi p_3))^2}`$ between the intermediate particle and its partner 3. It originates from the flux factor in the definition of the production cross section and is proportional to the length $`L=\widehat{v}\tau `$ the pion travels.
A similar expression like Eq. (11) arises if the intermediate nucleon $`X`$ moves on shell. Formally, the integral over the pion momentum in Eq. (11) is replaced by an integral over the scattering angle of the elastic nucleon-nucleon scattering. To derive such an equation the on-shell part of the amplitude $`\widehat{T}`$, see Eq. (9), is transformed into a sum over the spin quantum numbers of the intermediate nucleon using the properties of the Dirac operator,
$`\widehat{T}=T_1(\gamma ^\mu p_\mu +m_N)T_2={\displaystyle \underset{n}{}}T_{1n}T_{2n}.`$ (12)
To arrive at the standard formula the following approximation is needed: $`|_nT_{1n}T_{2n}|^21/2|T_1|^2|T_2|^2`$ which is fulfilled if spin-flip processes are not important.
If we apply Eq. (11) to a p+A collision of a nucleon on a nucleus of mass number $`A`$ the cross section would be proportional to $`A^{4/3}`$ in the subthreshold region as the length $`L`$ is proportional to $`A^{1/3}`$. However, if the reaction proceeds in a dense medium the length reduces to the mean free path, and the factor $`n_0\widehat{v}\tau =1/\sigma _{tot}`$ gives the inverse total cross section. This brings us back to the prescription of Eq. (6) and the two-step formula employed, e.g., in . In this case the cross section of a p+A reaction is proportional to $`A^{2/3}`$ taking into account the fact that the incoming particle suffers rescattering processes too.
After separating the two-step process from the total cross section the remaining parts form the genuine three-body cross section $`\sigma _{three}`$. It contains the smooth terms of the $`𝒯`$ matrix in Eq. (7) and the limit of the divergent contributions of Eq. (10). To see that the expression in the square brackets in Eq. (10) is finite we analytically integrate over $`p^2`$ around the singularity within the interval between $`p^2=m^2\pm mD`$. Assuming that $`\widehat{T}`$ is a smooth function of $`p^\mu `$ we obtain $`2/mD`$ for a small but finite value of $`D`$. Dividing the integration into parts near and far the singularity we write the three-body cross section as
$`\sigma _{three}`$ $`=`$ $`C{\displaystyle }d𝐋_{inv}[\mathrm{\Theta }(|p^2m^2|mD)|𝒯|^2`$ (13)
$`+\delta (p^2m^2)(2mD|\overline{𝒯}|^2+2\pi m(\overline{𝒯}\widehat{T}^{}){\displaystyle \frac{2}{mD}}|\widehat{T}|^2)],`$
where $`\mathrm{\Theta }`$ denotes the step function. Terms of higher than first order in $`mD`$ have been neglected.
## 4 In-medium effects
In Eq. (10) we have assumed that the intermediate particles move freely. Inside a medium we can employ a finite imaginary part as already inserted in Eq. (6). Recently great effort has been made to construct the in-medium propagators and corresponding spectral functions calculating the self-energies in one-loop order and higher approximations to couple the propagating particles to the constituents of the surrounding medium. However, for simplicity reasons we treat these effects phenomenologically as collision broadening introducing in the propagator in Eq. (6) the collision width $`\mathrm{\Gamma }_{coll}=n_0\widehat{v}\sigma _{tot}`$
$`\mathrm{\Gamma }={\displaystyle \frac{1}{\tau }}=\mathrm{\Gamma }_0(p^2)+\mathrm{\Gamma }_{coll}`$ (14)
We have already foreseen that the intermediate particle is a resonance with an energy dependent decay width $`\mathrm{\Gamma }_0`$. Although the definitions in Eq. (10) can be used also for a finite value of $`\epsilon `$ after transforming back the $`\delta `$ functions and their squares into the finite-epsilon representations, this procedure is ambiguous here since there is no singularity anymore. Thus, we define the genuine three-body cross section as the difference of the total cross sections and all possible two-step cross sections:
$`\sigma _{three}=C{\displaystyle 𝑑𝐋_{inv}\left(|𝒯|^2\underset{c}{}|\widehat{T}_c|^2\delta (p_c^2m_c^2)\frac{\pi }{\mathrm{\Gamma }m_c}\right)}.`$ (15)
The sum runs over all open channels $`c`$ for two-step collisions, where resonances are included as long as their decay channels are open, e.g. intermediate $`\rho `$ mesons with masses $`m_c`$ being larger than the two-pion decay threshold.
Due to the surrounding medium the $`𝒯`$ matrix looses its Lorentz invariance as the relative velocity refers to the medium’s rest frame which we fix to be the laboratory system. We mention that the value of $`\widehat{v}`$ becomes very large for pions which create a $`\varphi `$ meson. A pion needs a value of $`\widehat{v}11`$ in a collision with a nucleon resting in the laboratory system, while its cross section amounts to about 25 mb. The corresponding relative velocity for a $`\rho `$ meson is considerably smaller due to its larger mass.
In our calculations we use the effective widths defined in Eq. (14) for those particles only which serve as intermediate particles, and treat particles which occur inside the diagrams of the two consecutive processes like free particles. This corresponds to the widely used standard method by which cross sections are calculated with free particle propagators.
## 5 Numerical results
Now we are going to discuss the cross section for a proton which hits another proton at rest embedded in a protonic surrounding with density $`n_0`$=0.16 fm<sup>-3</sup>. We vary the bombarding energy $`E_{lab}`$ from the three-body threshold of 1.8 GeV for $`\varphi `$ meson production up to the two-nucleon threshold of 2.6 GeV. In this special case of two particles without relative motion the nucleonic intermediate states cannot get on shell below the two-nucleon threshold while pions and $`\rho `$ mesons already allow two-step processes at an bombarding energy of about 0.04 GeV above threshold.
First we have calculated the cross section without the in-medium effect. The two-step cross section (9) is proportional to the flight time $`\tau `$ or the system size. On the other hand the genuine three-body cross section which is calculated according to Eq. (13) does not depend on the system size and is presented in Fig. 2 by the thin full line.
The cross sections are considerably diminished if the rescattering effects of the intermediate particles are taken into account. We have used a total cross section of 25 mb for both pions and $`\rho `$ mesons. The two corresponding two-step processes (defined within the right hand site of Eq. (15)) are also displayed by the dashed and the dot-dashed lines in Fig. 2, respectively. These cross sections are considerably smaller than the original three-body cross section. The sum of these two quantities has to be compared to the total cross section calculated with the complete $`𝒯`$ matrix of Eq. (15) which is presented by the thick full line. The difference should be the contribution of the genuine three-body cross section which, however, turns out to be negative (thick dashed line). The reason is that due to the large width of 200 MeV the pion propagator in the three-body matrix element reaches out very far in the phase space and picks up $`𝒯`$ matrix elements which are smaller than in the on-shell region.
Finally, we calculate the cross section for the collision of a proton on a target nucleus consisting of $`N`$ neutrons and $`Z`$ protons. The production is now a superposition of primary reactions on pp and pn pairs. The production at a pn pair is strongly enhanced compared with that at a pp pair because the isospin coupling allows in the latter case only $`\pi ^0`$ exchanges which have smaller vertex strengths than those of charged mesons. In more complete calculations which, e.g., include also the exchange of $`\sigma `$ mesons such a large difference may not occur. Furthermore, the nucleons have Fermi momentum which can be taken into account by using the spectral function $`\delta (p^0m_N𝐩^2/[2(A1)m_N]+E_{sep})\mathrm{exp}(5𝐩^2/2k_{Fermi}^2)`$ with a Fermi momentum of $`k_{Fermi}`$ = 0.27 GeV and an average separation energy of $`E_{sep}`$ = 20 MeV. The effect of the Fermi motion allows the production already at much lower energy. In view of the exploratory nature of our calculations it is not worth using improved spectral functions which can be found in ref. . Due to the effect of Fermi motion the production already begins at much lower energy. In Fig. 3 the cross sections $`\sigma _{ppp}`$ and $`\sigma _{ppn}`$ for $`\varphi `$ production at a proton in the surrounding of protons or respectively neutrons are presented. They are used to calculate the $`\varphi `$ production in the reaction of a proton with <sup>12</sup>C in accordance to $`\sigma =[(Z(Z1)\sigma _{ppp}+2ZN\sigma _{ppn}+N(N1)\sigma _{pnn}]/A`$. We have used an effective particle number $`A`$ = 6 as in ref. which results from the fact that many target nucleons are screened due to scattering processes of the incoming proton. For comparison we have also included into Fig. 3 the $`\varphi `$ production cross-section (dash-dotted line) calculated as superposition of primary two-nucleon collisions, $`\sigma =Z\sigma _{pp}+N\sigma _{pn}`$, which is smaller than the three-body production, in contrast to the results of ref. .
Finally we should keep in mind that we have constrained our Lagrangian of the nucleon-nucleon-meson interaction. For instance neglecting the intermediate $`\mathrm{\Delta }`$ particles causes an underestimation of the pion production. Therefore, the actual $`\varphi `$ production is expected to be larger than presented in Figs. 2 and 3.
## 6 Conclusion
It has been our aim to study the features and consequences of elementary three-body collisions. This is an important issue when calculating particle production near threshold for nucleon-nucleus and nucleus-nucleus collisions within the frame work of transport models. Here, we have studied the $`\varphi `$ production below the free nucleon-nucleon threshold. It turned out that genuine three-body processes are not important as most of the intensity can be evaluated via consecutive two-nucleon reactions. In the case considered here the two-step processes slightly overestimate the cross section. However, it is necessary that all intermediate particles which can become on shell are treated properly within the transport codes as they contribute essentially to particle production.
|
warning/0005/cond-mat0005267.html
|
ar5iv
|
text
|
# Magnetic fluctuations in 2D metals close to the Stoner instability
## I Introduction
The itinerant ferromagnetism in metals has been the subject of extensive theoretical and experimental studies. In clean systems the ferromagnetic instability is described by the Stoner criterion , which defines the critical value of the spin-exchange interaction constant (at which the system becomes unstable with respect to ferromagnetic ordering). The value of the constant depends on the material. In most simple metals it is small (so that the electronic system exhibits paramagnetic response), while in palladium it approaches the critical value . Theoretically, the Stoner criterion is most easily obtained within the framework of the Hubbard model or in Landau Fermi liquid theory .
In two dimensions disorder tends to localize the electronic system . However, if either the sample size $`L`$ or the dephasing length $`L_\phi `$ is smaller than the localization length, the sample can still be considered as metallic. However, the interaction constants are renormalized from their clean values. In particular, the spin-exchange interaction constant was shown to flow towards strong coupling . It is possible therefore to have a sample with the renormalized constant approaching the critical value while at the same time far from the Anderson localization. In this case a mean-field structure of the ferromagnetically ordered ground state was conjectured recently in Ref. .
An important issue in the physics of disordered systems is the role of mesoscopic fluctuations. Indeed, the Stoner criterion in its usual form reflects the tendency of the system to acquire uniform non zero magnetization (if the interaction constant happens to reach the critical value). However, for each realization of disorder the spin exchange interaction is non-local and random. The effective interaction constant entering the Stoner criterion is essentially the interaction kernel averaged over the whole system. At the same time, averaging over some small part of the system might produce a value of the effective interaction constant different from the system-wide average. In particular, some rare impurity configurations can lead to the locally averaged interaction constant in some region to satisfy the Stoner criterion, while the system-wide average does not. This would mean the appearance of non zero spin polarization in such regions. The similar effect in finite-size mesoscopic systems was considered in Ref. .
In this paper we investigate the plausibility of such a scenario and its effect on magnetic and transport properties of the system. We consider a good metal (characterized by a large dimensionless conductance $`g=2\pi \mathrm{}/e^2R_{\mathrm{}}1`$) in 2D close to the instability, but still in the paramagnetic phase (at the mean field level), so that the naive mean-field value of the total magnetization of the system is zero. We describe the spin exchange interaction by the effective averaged constant $`F_0`$ with all the renormalizations already included (as shown in Ref. ) and by the random, sample specific non-local susceptibility, which, when averaged over the area of a small region, gives the local effective interaction constant. Using the “optimal fluctuation” method we find that there is exponentially small but non zero probability to find such region with non zero spin polarization which we call a local spin droplet (LSD). With the exponential accuracy this probability does not depend on the size of a LSD, therefore droplets of all sizes (up to the size of the order of the thermal length $`L_T=\sqrt{\mathrm{}D/T}`$, ($`D`$ is the diffusion constant and $`T`$ is the temperature) can appear. The total spin of a LSD is also independent of its size. The effective interaction between LSDs is determined by the correlations of the same non-local susceptibility. Since LSDs are extended objects, the correlation functions which determine the effective interaction constant have to be averaged over the area of both interacting LSDs. The oscillating parts of the correlation functions (which usually lead to the RKKY interaction) do not survive this averaging. Instead, we find that the average value of the effective interaction is ferromagnetic and decays with the distance between LSDs only as a power law. However, at large distances fluctuations of the effective interaction constant exceed the average and the sign of the interaction becomes random.
The contribution of LSDs to physical observables manifests itself at low temperatures. Indeed, as any system of weakly interacting moments at temperatures, higher than the point of magnetic ordering, the system of LSDs exhibit the Curie-like susceptibility. We show, that at not so low temperatures, this contribution exceeds the (temperature-independent) Pauli susceptibility of the electron system. Likewise, the LSDs contribute to the dephasing time $`\tau _\phi `$ (extracted from magnetoresistance measurements, see Ref. ). Only, while usually the dephasing time in two dimensional electron systems behaves like $`\tau _\phi ^1T`$, the contribution of LSDs is temperature independent. Thus at temperatures lower than certain cross-over temperature the dephasing time saturates. The cross-over temperature is roughly the same for both quantities. When $`T0`$, interactions of LSDs with each other or with itinerant electrons should lead either to the screening of the local spins or to forming of some spin glass state (due to the randomness of the interaction). We do not consider such regime in this paper.
The paper is organized as follows. In Section II we review the basic physical description of the Stoner instability, establish notations and, in subsection II C we outline the effect of mesoscopic fluctuations. Next, in Section III we discuss the formation of LSDs qualitatively. The subsection IV A is devoted to the calculation of the probability to form a LSD. Then, in subsection IV B we find the distribution of the total spin of the LSD by means of the “optimal fluctuation” method , which we adapt to the non-linear problem. Section V describes the interaction of LSDs, and in Section VI we discuss the contribution of LSDs to physical observables. Our results are summarized in Conclusions. Some mathematical details of the non-local susceptibility correlations are relegated to the Appendix.
## II Stoner Instability
The purpose of this Section is to recall the basic ideas leading to the Stoner criterion (subsections II A and II B) and to contrast the situation in clean systems to that in disordered metals. In subsection II C we demonstrate the role of mesoscopic fluctuations and discuss how they can lead to formation of LSDs. For further details on Stoner ferromagnetism the reader is referred to the standard textbooks Ref. .
### A Renormalized paramagnetic susceptibility
The paramagnetic response of a system of non-interacting electrons is described by the Pauli susceptibility which depends only on the electronic density of states at the Fermi level $`\nu `$. This can be seen from the following observation. The electron energy $`ϵ`$ enters all thermodynamic functions in combination $`ϵ\mu `$ with the chemical potential $`\mu `$. The interaction energy of the electron spin $`\stackrel{}{s}`$ with the external magnetic field $`\stackrel{}{h}`$ (which is proportional to $`\stackrel{}{s}\stackrel{}{h}`$) can thus be considered as a shift of the chemical potential. Since it is proportional to $`\stackrel{}{s}`$, the number of electrons which spin is aligned with $`\stackrel{}{h}`$ exceeds the number of electrons with the opposite spin, resulting in the total magnetization proportional to $`\nu \stackrel{}{h}`$.
Although quite general, the above argument relies on the fact that weak magnetic field does not change the energy spectrum of the electron system. Taking into account the electron-electron interactions, however, changes the distribution function of electrons and thus the electronic energy spectrum. As a result, some physical quantities become renormalized from their bare values. In particular, the paramagnetic susceptibility is renormalized by the the spin exchange interaction (we choose the units with Bohr magneton equal to unity)
$$\chi =\frac{\nu }{1+F_0}.$$
(1)
Here the parameter $`F_0`$ is the effective dimensionless coupling constant of magnetic interaction between electron spins. Within the phenomenology of the Landau Fermi-liquid theory it can be obtained by averaging the spin-exchange part of the Landau function over the Fermi surface. In the case when $`F_0<0`$, the interaction tends to align the electrons spins, competing with the Pauli exclusion principle. If the interaction is strong enough, the gain in the magnetic energy exceeds the kinetic energy cost needed to realign the spins and the ground state of the system changes to the one with non zero total spin \- it becomes ferromagnetic. According to Eq. (1) the instability occurs when $`F_0=1`$, which is known as the Stoner criterion.
More formally, the full susceptibility tensor is given by the commutator of the spin density operators $`\widehat{\sigma }_\alpha (x,t)`$ (hereafter $`x_i`$ denotes the two-component coordinate vector)
$`\chi _{\alpha \beta }(x_1,`$ $`x_2;t_1t_2)`$ (4)
$`=i\theta (t_1t_2)[\widehat{\sigma }_\alpha (x_1,t_1),\widehat{\sigma }_\beta (x_2,t_2)],`$
where $`\alpha ,\beta =x,y,z`$. In the paramagnetic state of an isotropic system the tensor $`\chi _{\alpha \beta }`$ is diagonal and isotropic and can be expressed in terms of the transverse susceptibility $`\chi `$,
$`\chi _{\alpha \beta }=2\chi \delta _{\alpha \beta },`$
which is determined in terms of the commutator of the spin raising and lowering operators $`\widehat{\sigma }^+`$ and $`\widehat{\sigma }^{}`$ similarly to Eq. (4).
The transverse susceptibility can be evaluated in the generalized Hartree-Fock approximation , which amounts to summation of ladder diagrams in the particle-hole channel. In the Galilean invariant system the susceptibility depends only on the coordinate difference and in the momentum representation is given by
$$\chi (q,\omega )=\frac{\mathrm{\Pi }(q,\omega )}{1+U\mathrm{\Pi }(q,\omega )}.$$
(5)
Here $`\mathrm{\Pi }(q,\omega )`$ is the electron polarization operator, which represents the susceptibility of the non-interacting electron gas. In the limit $`\omega =0`$ and $`q=0`$ it gives the Pauli susceptibility \[since $`\mathrm{\Pi }(q=0,\omega =0)=\nu `$\]. The parameter $`U`$ is the spin exchange coupling constant. In the context of the Hubbard model it appears as a phenomenological parameter of the Hamiltonian. In the microscopic Fermi-liquid theory it is the spin exchange part of the vertex function $`\mathrm{\Gamma }^\omega `$ averaged over the Fermi surface (compare with the scattering amplitude $`\mathrm{\Gamma }_2`$ in Ref. ). The instability in the ground state of the system corresponds to the singularity in the static limit of the response function $`\chi (q,\omega =0)`$. The instability criterion is thus $`U\mathrm{\Pi }(q,\omega =0)=1`$. At $`q=0`$ this corresponds to a tendency of the system to acquire spontaneously a uniform (or ferromagnetic) spin density. The criterion for this instability
$$U\nu =1.$$
(6)
corresponds to the Stoner criterion if one identifies the Landau parameter $`F_0`$ with the spin exchange coupling constant.
### B Magnetization energy
The transition to the ferromagnetic state can also be described similarly to the Landau description of the phase transitions with the induced spin density as the order parameter. Close to the instability point, the thermodynamic potential $`\mathrm{\Omega }`$ of the system can be expanded in powers of the spin density $`\stackrel{}{\sigma }`$
$$\mathrm{\Omega }=\frac{dV}{2\nu }\left[(1+F_0)\stackrel{}{\sigma }^2+a^2|\stackrel{}{\sigma }|^2+\frac{1}{2}B(\stackrel{}{\sigma }^2)^2+\mathrm{}\right],$$
(7)
where $`B>0`$. The value of $`\stackrel{}{\sigma }`$ of any particular state of the system (as described by the constants $`F_0`$, $`a`$, and $`B`$) can be found by minimizing the thermodynamic potential $`\mathrm{\Omega }`$. In the limit $`q=0`$ (uniform spin density) the minimum condition is
$$(1+F_0+B\stackrel{}{\sigma }^2)\stackrel{}{\sigma }=0.$$
(8)
Here we have neglected the gradient term relative to the linear term, since the coefficient $`ak_F^2`$ and under our assumptions, $`g1`$ and $`g(1+F_0)1`$ (the latter assumption will be elaborated on in Section III). Therefore the linear term dominates, $`1+F_0>k^2/k_F^2`$ even for large momenta $`k1/l`$ ($`l`$ is the mean free path). As usual, in order to have a non-trivial solution $`\sigma 0`$, one should have $`1+F_0<0`$, which is again the Stoner criterion. In the ferromagnetic phase the solution to Eq. (8) gives the total value of the induced spin density $`\sigma ^2=(1+F_0)/B`$.
### C Mesoscopic fluctuations
In the presence of disorder the local spin density $`\sigma (x)`$ depends on the particular impurity distribution and before averaging can be taken as random. In the paramagnetic susceptibility (5) the coupling $`U`$ is determined by small distances and thus does not depend on disorder. On the contrary, the polarization operator $`\mathrm{\Pi }(x_1,x_2,\omega )`$ includes large distances (since we are interested in the limit $`q=0`$) and is thus strongly affected by disorder. Consequently, the paramagnetic susceptibility is random. Moreover, it depends on both coordinates since translational invariance is lost and it is also non-local. However, if disorder is not too strong, then the non-local, random part of the susceptibility can be separated from the uniform term, which is independent of disorder and represents the susceptibility of the clean system. Preserving the form of Eq. (1), the susceptibility of the disordered system can be found as a solution to the integral equation
$`{\displaystyle }d^2x_2[(1+F_0)\delta (x_1x_2)`$ $`+F_1(x_1,x_2)]\chi (x_2,x_3)`$ (11)
$`=\nu \delta (x_2x_3),`$
where $`F_1(x_1,x_2)`$ is a random quantity with zero mean.
Similarly, the thermodynamic potential (7) becomes
$`\mathrm{\Omega }`$ $`={\displaystyle \frac{1}{2\nu }}{\displaystyle d^2x(1+F_0)\stackrel{}{\sigma }^2(x)}`$ (16)
$`+{\displaystyle \frac{1}{2\nu }}{\displaystyle d^2x_1d^2x_2F_1(x_1,x_2)\left(\stackrel{}{\sigma }(x_1)\stackrel{}{\sigma }(x_2)\right)}`$
$`+{\displaystyle \frac{1}{4\nu }}{\displaystyle \underset{i=1}{\overset{4}{}}d^2x_iB\left[\{x_j\}\right]\left(\stackrel{}{\sigma }(x_1)\stackrel{}{\sigma }(x_2)\right)\left(\stackrel{}{\sigma }(x_3)\stackrel{}{\sigma }(x_4)\right)}.`$
Again, we are interested in the limit $`q0`$ (neglecting the gradient term; see the previous subsection and an estimate below). In the second order term we have neglected the possibility of the presence of spin-orbit coupling. While the spin-orbit interaction can be taken into account, its presence does not affect our main results (see Section V for discussion). In addition to the fourth-order term written in Eq. (LABEL:tpnl) there is a term with different spin structure, namely $`\left[\stackrel{}{\sigma }(x_1)\times \stackrel{}{\sigma }(x_2)\right]\left[\stackrel{}{\sigma }(x_3)\times \stackrel{}{\sigma }(x_4)\right]`$ (see Appendix for details). In what follows we assume the simplest spin structure for a LSD $`\stackrel{}{\sigma }=(0,0,\sigma )`$. In this case the additional cross product term vanishes and we can treat the spin density as a scalar.
Similarly to Eq. (8) the minimum of $`\mathrm{\Omega }`$ can be found from the (now non-local) integral equation
$`(1`$ $`+F_0)\sigma (x_1)+{\displaystyle }d^2x_2F_1(x_1,x_2)\sigma (x_2)`$ (19)
$`+{\displaystyle \underset{i=2}{\overset{4}{}}d^2x_iB\left[\{x_j\}\right]\sigma (x_2)\sigma (x_3)\sigma (x_4)}=0.`$
The coefficient $`B\left[\{x_j\}\right]`$ in the thermodynamic potential Eq. (LABEL:tpnl) is also random. In this paper we take both $`F_1`$ and $`B`$ to be Gaussian random matrices, with the distribution, which in compactified notation is given by
$`w[F_1,B]\mathrm{exp}\left[{\displaystyle \left(\begin{array}{cc}F_1;& B\end{array}\right)\widehat{𝒦}^1\left(\begin{array}{c}F_1\\ B\end{array}\right)}\right]`$ (20)
The Gaussian approximation is valid while the expression in the exponent does not exceed the dimensionless conductance $`g`$, where the log-normal tail appears . The integration is over all the variables of $`F_1`$ and $`B`$. The distribution $`w[F_1,B]`$ should be understood in the operator sense and will be used to evaluate the functional integrals below. The weight operator $`\widehat{𝒦}`$ is constructed from the correlators
$`\widehat{𝒦}=\left(\begin{array}{cc}F_1F_1& F_1B\\ BF_1& BB\end{array}\right),`$ (21)
which are discussed in detail in Appendix, where we give their explicit form.
For a given realization of disorder, the equation (19) might allow for some non-trivial solution $`\sigma ^{(0)}(x)`$. To estimate the value of the total spin corresponding to such a solution, we write the spin density as
$$\sigma ^{(0)}(x)=\sigma _0\psi (x),$$
(22)
where $`\psi (x)`$ is normalized to unity, therefore both $`\psi (x)`$ and $`\sigma _0`$ have dimension of inverse length. The total value of the spin is determined by
$$S=\sigma _0d^2x\psi (x),$$
(23)
while $`\sigma _0`$ can be found from Eq. (19) (in the case when it allows a non-trivial solution) in the form
$$\sigma _0^2=\frac{1+F_0+F_1^{(0)}}{B^{(0)}}.$$
(24)
Here the constants $`F_1^{(0)}`$ and $`B^{(0)}`$ are the “matrix elements” of the non-local operators $`F_1(x_1,x_2)`$ and $`B\left[\{x_j\}\right]`$ (if $`\psi (x)`$ is interpreted as a “wave function”)
$$F_1^{(0)}=d^2x_1d^2x_2F_1(x_1,x_2)\psi (x_1)\psi (x_2)$$
(26)
$$B^{(0)}=\underset{i=1}{\overset{4}{}}d^2x_iB\left[\{x_j\}\right]\underset{k=1}{\overset{4}{}}\psi (x_k).$$
(27)
Consider now a metal close to the Stoner instability, so that $`0<1+F_01`$. While, as follows from Eq. (8), the averaged, uniform spin density is zero, solutions (24) to the non-local equation (19) might exhibit non zero spin density in some rare regions, where due to a particular configuration of impurities the non-local part of the susceptibility $`F_1`$ is negative. If there are several regions with non zero total spin, then the spin-spin or magnetic interaction between them will contribute to the ground state energy of the system and correspondingly to the magnetic susceptibility. If such interaction favors some kind of ordering of the spins, then the appearance of these regions can change the magnetic response of the system, in other words change the ground state.
Fluctuation effects in systems close to a phase transition have been studied extensively (see, for instance, Ref. ). In particular, the picture of smearing the transition point by formation of fluctuation regions with non-zero value of an order parameter was considered in Ref. . In order to determine a value of the order parameter in a fluctuation region a solution of the non-linear Ginsburg-Landau equation in the presence of disorder was needed. Two issues make this case different from ours. First, in Ref. the fourth-order term $`B`$ was not random. Second, the fluctuations of the order parameter were assumed to be local. Therefore, the white noise approximation (i.e. approximating the correlatino functions (21) by the delta-functions) for the disorder was appropriate. As a result, the fluctuation regions did not interact and the percolation scenario of the phase transition was needed. In our case the non-locality of fluctuations (expressed in terms of the correlation functions (21) leads to interaction between LSDs which results in a change of behavior of observable quantities as discussed in Section VI.
## III Qualitative discussion
In the previous subsection II C we indicated how LSDs - local regions with non zero spin polarization - could appear in a metal close to the Stoner instability due to fluctuations in impurity distribution. Here we estimate qualitatively the probability to find a LSD and the value of its total spin.
Treating a LSD as an open region of the size $`R`$ we can characterize it by the Thouless energy $`E_T=D/R^2`$ (where $`D`$ is the diffusion constant). The inverse of the Thouless energy is the “escape time” $`\tau _{esc}=E_T^1`$, which is the time it takes for the diffusing particle to leave the LSD. This time scale serves as the infrared cut-off for the correlation function, which describes the mesoscopic fluctuations of the density of states (DoS)
$`\rho (ϵ)\rho (ϵ+\omega )`$ $`\mathrm{R}e{\displaystyle \frac{R^2d^2Q}{(i\omega +DQ^2+\tau _{esc}^1)^2}}`$ (30)
$`={\displaystyle \frac{\pi }{E_T}}\mathrm{R}e{\displaystyle \frac{1}{i\omega +\tau _{esc}^1}}.`$
The magnetic energy of the LSD, written in terms of its total spin $`S`$ is
$$E(S)=\delta _1(1+F_0)S^2+\delta E(S),$$
(31)
where $`\delta _1`$ is the mean level spacing and $`\delta E(S)`$ denotes contribution of all non-linear terms in Eq. (LABEL:tpnl). All the terms contributing to $`\delta E(S)`$ are random and can be expressed in terms of the fluctuating (random) DoS
$`\delta E(S)={\displaystyle \underset{0}{\overset{\delta _1S}{}}}𝑑s_1{\displaystyle \underset{0}{\overset{s_1}{}}}𝑑s_2\left[\rho (s_2)+\rho (s_2)\right].`$ (32)
The averaged $`\delta E(S)`$ equal to zero, but the average of its square $`(\delta E(S))^2`$ is not and it is determined by the correlator (30)
$`(\delta E(S))^2\{\begin{array}{cc}\frac{\delta _1^2S^4}{g^2},& Sg;\\ \frac{\delta _1^2S^3}{g},& Sg.\end{array}`$ (33)
In this paper we restrict ourselves to consideration of metals close to the instability point, where the overall spin of the LSD is small $`Sg`$. In this case we can treat $`\delta E(S)`$ as a Gaussian random quantity. Moreover, we can expand it in powers of $`S/g`$ so that the magnetic energy of the LSD becomes
$$E(S)=\delta _1(1+F_0)S^2+\zeta \left[\frac{S^2}{g}\frac{S^4}{g^3}\right],$$
(34)
where $`\zeta `$ is a random Gaussian variable with the distribution
$`P(\zeta )\mathrm{exp}\left(\zeta ^2\right).`$ (35)
Note that there is only one random quantity $`\zeta `$ in Eq. (35). The equation (34) is valid only for $`Sg`$. That is why energy minima at$`\zeta >0`$ are spurious and should not be considered.
To find the distribution of the spin value, we need to minimize the energy $`E(S)`$. Differentiating Eq. (34) we obtain the equation
$$\left(1+F_0+\frac{\zeta }{g}\right)S\zeta \frac{S^3}{g^3}=0.$$
(36)
This equation allows for non-trivial solutions $`S>0`$ when $`\zeta <g(1+F_0)`$. As was noted above, we are working in the regime where $`g(1+F_0)1`$, so that the fluctuation that creates the LSD is rare indeed. To determine the distribution of spin values, we solve the equation (36) for $`\zeta `$ and substitute in the Gaussian distribution Eq. (35). As a result, we estimate the distribution (up to numerical coefficients)
$$P(S)=\mathrm{exp}\left(\frac{g^2(1+F_0)^2}{(1\frac{S^2}{g^2})^2}\right).$$
(37)
This distribution is only valid when $`g(1+F_0)1`$ and $`Sg`$, therefore we can expand the denominator in the exponent without loss of accuracy
$$P(S)=\mathrm{exp}\left[g^2(1+F_0)^2(1+F_0)^2S^2\right].$$
(38)
Integrating over the spin $`S`$ we estimate the probability to find a LSD
$$𝒫\mathrm{exp}\left(g^2(1+F_0)^2\right).$$
(39)
It is determined by the first term in the exponent Eq. (38). The second term determines the typical value of the spin of the LSD
$$S\frac{1}{1+F_0}.$$
(40)
Remarkably, this value and the probability Eq. (39) do not depend on the size $`R`$ of the LSD, which is the main qualitative result of this Section. Note that it is similar to the result for zero-dimensional grains . Also the spin value is independent of the dimensionless conductance $`g`$. These facts determine the contribution of LSDs to the physical observables considered below in Section VI.
The applicability of the consideration of this Section and quantitative results of Section IV is limited by two requirements. First, the probability Eq. (37) must be exponentially small, so that $`g(1+F_0)1`$. Second, the Gaussian approximation Eq. (35) \[similarly to the distribution Eq. (20)\] is valid while the expression in the exponent is smaller than the dimensionless conductance of the system , so that in Eq. (37) $`g^2(1+F_0)^2<g`$. Combining the two limits, we obtain the region of applicability of the results Eqs. (37) - (40) as
$$1/g1+F_0<1/\sqrt{g}.$$
(41)
## IV Local spin droplets
In this section we calculate the probability to find a local region with non zero spin, which we call a local spin droplet (LSD) and the value of the total spin of the LSD. The spin and the spatial profile of the LSD can be found from the non-linear equation (19). In subsection IV A we show that to the exponential accuracy the probability to find the LSD Eq. (39) is captured by the linear part of Eq. (19), while the non-linear term fixes the spin value, as shown in subsection IV B.
### A Probability to form a LSD
The calculation of the probability to find a rare fluctuation leading to formation of a LSD can be performed along the lines of the argument used in Ref. to calculate the exponentially small tail in the density of states (DoS) of a particle in a random potential . In the quantum mechanical problem, considered in Ref. , one looks for such fluctuation of the random potential that creates a low energy bound state, thus leading to non zero DoS at that energy. The probability to form the bound state is determined by the distribution of the matrix elements of the random potential. While being exponentially small, the probability should be maximized by choosing the “optimal” fluctuation of the potential.
To gain some intuition about how the optimal fluctuation method can be applied to the problem at hand, in this section we consider the linear part of the equation (19), disregarding for a moment the higher order $`B`$ term. Such an approach can be justified by observing that close to the instability the non-linear term in Eq. (19) is small compared to the linear ones, since the induced spin density on average is equal to zero. The non-linear term stabilizes a non-trivial solution and fixes its amplitude, while the existence of such a solution can be uncovered at the level of the linear problem. Thus the linear equation captures the main contribution to the probability and at the same time demonstrates the similarity of our problem to the problem of tails in the DoS as well as the peculiar differences.
We can write the linear equation in the operator form
$$\widehat{F}_1\psi (x)=E\psi (x).$$
(42)
We write $`\psi (x)`$ instead of $`\sigma (x)`$ to stress the point that the linear equation does not allow us to determine the value of the spin but only the spatial profile of the LSD. Therefore, Eq. (42) is simply the eigenproblem for the operator $`F_1`$ and as such does not fix the normalization of eigenfunctions $`\psi (x)`$, which we are free to normalize to unity for convenience. Since the eigenvalue problem Eq. (42) is similar to the quantum mechanical problem of Ref. , we can adopt the language of the Schrödinger equation, with the (now integral) operator $`\widehat{F}_1`$ playing the role of the “Hamiltonian”, $`E`$ the “energy” and $`\psi (x)`$ the wave-function.
For some particular realizations of the random potential in Eq. (42) there is a low energy bound state with the energy $`E_0[\{F_1\}]=F_1^{(0)}`$ \[given by Eq. (26)\], resulting in non zero DoS at this energy. For energies close to $`E_0[\{F_1\}]`$ only the bound state contributes to the DoS, which before averaging over disorder is given by the single delta-function
$$\rho (E)=\delta (EE_0[\{F_1\}]).$$
(43)
Averaging this DoS over disorder takes into account contributions of all possible realizations of the random potential leading to such bound states and thus results in an exponentially small but smooth function of energy $`E`$. This function is proportional to the probability to find the bound state at energy $`E`$. In particular, for the special value $`E=(1+F_0)`$, it would give the probability to find the non-trivial solution to the linear part of Eq. (19) (or to find the LSD).
The random quantity in the linear problem Eq. (42) is $`F_1`$ itself. Its distribution is obtained from Eq. (21). The averaged probability is then
$`\rho (E)=`$ $`{\displaystyle 𝒟[F_1]\delta (EE_0[\{F_1\}])}`$ (47)
$`\times \mathrm{exp}\left[{\displaystyle d^2x_1d^2x_2d^2y_1d^2y_2A}\right],`$
$`A=F_1(x_1,x_2)K_{FF}^1[\{x_j\},\{y_j\}]F_1(y_1,y_2),`$ (48)
where $`K_{FF}^1[\{x_j\},\{y_j\}]`$ is the inverse of the correlator Eq. (145) i.e.
$`{\displaystyle d^2y_1d^2y_2K_{FF}^1[\{x_j\},\{y_j\}]}`$ $`K_{FF}[\{y_j\},\{z_j\}]=`$
$`\delta (x_1z_1)\delta (x_2z_2).`$
In the “optimal fluctuation” approach one has to evaluate the integral Eq. (IV A) in the saddle point approximation. To find the saddle point one has to minimize the exponent $`A`$ of the Gaussian probability with respect to all functions $`F_1(x_1,x_2)`$ subjected to the condition
$$E=E_0[\{F_1\}],$$
(49)
represented by the delta-function in Eq. (IV A). This involves solving the equations
$`{\displaystyle \frac{\delta }{\delta F_1(x_1,x_2)}}\left[A+\lambda E_0[\{F_1\}]\right]=0,`$
where $`\lambda `$ is the Lagrange multiplier to be found from the condition Eq. (49).
The saddle point solution for $`F_1`$ represents the optimal fluctuation of the random potential,
$`\overline{F}_1(`$ $`x_1,x_2)=`$ (52)
$`\lambda {\displaystyle d^2z_1d^2z_2K_{FF}[\{x_j\},\{z_j\}]\psi (z_1)\psi (z_2)},`$
given in terms of the eigenfunction $`\psi (x)`$ corresponding to the eigenstate $`E_0`$. Substituting Eq. (52) into Eq. (42) we obtain
$`E\psi (x_1)=\lambda {\displaystyle }`$ $`d^2x_2d^2z_1d^2z_2\psi (x_2)`$ (55)
$`\times K_{FF}[\{x_j\},\{z_j\}]\psi (z_1)\psi (z_2),`$
which, together with the normalization condition $`d^2x|\psi (x)|^2=1`$, constitute the analogue of the non-linear Schrödinger equation of Ref. .
Both the saddle point value of $`\lambda `$ and $`\psi (x)`$ should be found from the non-linear equation (55) and the normalization condition. The averaged DoS is given by the Gaussian probability in Eq. (IV A), evaluated at the saddle point,
$`\rho (E)\mathrm{exp}\left({\displaystyle \frac{E^2}{2I_{FF}}}\right),`$ (56)
where
$`I_{FF}=`$ $`{\displaystyle d^2z_1d^2z_2d^2x_1d^2x_2}`$ (59)
$`\times \psi (x_1)\psi (x_2)K_{FF}[\{x_j\},\{z_j\}]\psi (z_1)\psi (z_2).`$
The integral $`I_{FF}`$ is dimensionless (independent of any length scale) since we require the eigenfunction $`\psi (x)`$ to be normalized, so that $`\psi (x)1/R`$, where $`R`$ is roughly the size of the LSD. As we are discussing the single LSD, all four of the eigenfunctions in Eq. (59) are centered around approximately the same point, therefore $`R`$ is the only scale in Eq. (59). In this case the dependence of the correlator $`K_{FF}`$ on $`R`$ is given by Eq. (159), $`K_{FF}g^2R^4`$. Thus the integral Eq. (59) and the probability Eq. (56) are independent of the size of the LSD $`R`$.
The probability to find the LSD is given by Eq. (56), evaluated at the point $`E=(1+F_0)`$,
$`\rho e^{\gamma g^2(1+F_0)^2},`$ (60)
where $`\gamma `$ is the numerical factor which is given by the dimensionless counterpart of the integral Eq. (59). This result is valid while the number in the exponent is large, which corresponds to the lower limit of applicability of our consideration $`1+F_01/g`$ (see the last paragraph of Section III).
To determine the numerical coefficient $`\gamma `$ in Eq. (60) we need to know the precise form of the eigenfunction $`\psi (x)`$. We could not solve the non-linear integral equation (55) analytically and used a variational approach. Since the kernel $`K_{FF}`$ in the integral Eq. (59) decays as $`R^4`$ at large distances, the optimal $`\psi (x)`$ is a limited-range function (i.e. its normalization integral is determined on a limited interval of $`x`$). To estimate the upper limit of $`\gamma `$ we take the variational function of the Gaussian form $`\psi (x)=\pi ^{1/2}R^1\mathrm{exp}(x^2/R^2)`$, substitute in Eq. (59) and evaluate the integral numerically. The resulting estimate is
$`I_{FF}{\displaystyle \frac{1}{0.8\pi ^2g^2}},`$ (61)
and the numerical factor in Eq. (60) is thus $`\gamma 3.9`$.
The most important feature of the result Eq. (60) is its independence on the size $`R`$ of the LSD in agreement with the qualitative results of Section III. This means that LSDs of small and large sizes can appear with equal (in the exponential sense) probability, given the suitable fluctuation of the impurity configuration. We shall return to this point below, when we consider the interaction between LSDs.
### B Spin distribution
The argument that led us to the probability to find an LSD Eq. (60) is not complete because it does not help us to determine the value of the spin of the LSD. This follows from our consideration of the linear part of Eq. (19) only. To determine the value of the spin we must solve the full non-linear equation.
The single-mode approximation to the solution of the non-linear problem was outlined in subsection II C. The formal solution for the amplitude of the spin of a LSD is given by Eq. (24). Again, as we did above in the case of the linear problem, we employ the optimal fluctuation method to find the probability to form a LSD, which is characterized by the total spin $`S`$. However, this time our task is simplified since the probability to form the LSD (regardless of its spin) has already been found. We now need to find how that probability depends on the spin value $`S`$. Therefore we take the function $`\psi (x)`$, which describes the spatial profile of the LSD from the linear problem and focus on the distribution of the spin amplitude $`\sigma `$.
Similar to Eq. (43), the probability to find an LSD characterized by the spin density amplitude $`\sigma `$ is given by the delta-function
$$\rho (\sigma ^2)=\delta (\sigma ^2\sigma _0^2[F_1,B]),$$
(62)
which we write in terms of $`\sigma ^2`$ for convenience.
Averaging over disorder is performed as it was done for the linear problem \[see Eq. (IV A)\]. Only now we have two random quantities, $`F_1`$ and $`B`$, therefore we need to average with the distribution Eq. (20)
$$\rho (\sigma ^2)=𝒟[F_1,B]\delta (\sigma ^2\sigma _0^2)w[F_1,B],$$
(63)
To find the saddle point (or the optimal fluctuation) we have to minimize the exponent
$`A_{nl}=\lambda (\sigma ^2+`$ $`{\displaystyle \frac{1+F_0+F_1^{(0)}}{B^{(0)}}})`$ (66)
$`{\displaystyle \left(\begin{array}{cc}F_1& B\end{array}\right)\widehat{𝒦}^1\left(\begin{array}{c}F_1\\ B\end{array}\right)}`$
where $`F_1^{(0)}`$ and $`B^{(0)}`$ are the integrals Eq. (27) and $`\lambda `$ is again the Lagrange multiplier. The saddle point equations are given by
$`\widehat{𝒦}^1\left(\begin{array}{c}F_1\\ B\end{array}\right)={\displaystyle \frac{\lambda }{2B^{(0)}}}\left(\begin{array}{c}\psi (x_1)\psi (x_2)\\ \sigma ^2\psi (y_1)\psi (y_2)\psi (y_3)\psi (y_4)\end{array}\right),`$ (67)
where the functions $`\psi (x)`$ appeared after differentiating the integrals Eq. (27) with respect to $`F_1`$ and $`B`$. Multiplying both the left-hand and the right-hand sides of Eq. (67) by $`\widehat{𝒦}`$ we obtain two equations with $`F_1`$ and $`B`$ in the left-hand side only. We then multiply the first equation by two functions $`\psi (x)`$ and the second by four functions $`\psi (x)`$ and integrate over their variables to obtain the algebraic equations, with integrals Eq. (27) as the unknowns
$$F_1^{(0)}=\frac{\lambda }{2B^{(0)}}\left(I_{FF}+\sigma ^2I_{FB}\right),$$
(69)
$$B^{(0)}=\frac{\lambda }{2B^{(0)}}\left(I_{FB}+\sigma ^2I_{BB}\right),$$
(70)
where $`I_{FF}`$ is given by Eq. (59) and $`I_{FB}`$ and $`I_{BB}`$ are similarly defined as
$`I_{FB}={\displaystyle d^2x_1}`$ $`d^2x_2{\displaystyle \underset{i=1}{\overset{4}{}}}d^2y_i\psi (x_1)\psi (x_2)`$ (74)
$`\times K_{FB}[\{x_j\},\{y_j\}]{\displaystyle \underset{k=1}{\overset{4}{}}}\psi (y_k),`$
$`I_{BB}={\displaystyle \underset{i=1}{\overset{4}{}}}`$ $`d^2x_id^2y_i{\displaystyle \underset{k=1}{\overset{4}{}}}\psi (x_k)`$ (77)
$`\times K_{BB}[\{x_j\},\{y_j\}]{\displaystyle \underset{k=1}{\overset{4}{}}}\psi (y_k).`$
Equations (70) can be easily solved and we have for the optimal fluctuation
$$\overline{F}_1^{(0)}=\sqrt{\lambda }\frac{I_{FF}+\sigma ^2I_{FB}}{\sqrt{2\left(I_{FB}+\sigma ^2I_{BB}\right)}},$$
(79)
$$\overline{B}^{(0)}=\frac{1}{2}\sqrt{\lambda }\sqrt{2\left(I_{FB}+\sigma ^2I_{BB}\right)}.$$
(80)
Now we only need to find $`\lambda `$ from the constraint $`\sigma ^2=(1+F_0+F_1^{(0)})/B^{(0)}`$ or, equivalently, to find the saddle point solution for $`\lambda `$. Substituting the saddle point solutions (80) into the constraint, we find
$$\sqrt{\lambda }=(1+F_0)\frac{\sqrt{2\left(I_{FB}+\sigma ^2I_{BB}\right)}}{I_{FF}+2\sigma ^2I_{FB}+\sigma ^4I_{BB}}.$$
(81)
The exponent $`A_{nl}`$ at the saddle point Eq. (80) is given by
$$\overline{A}_{nl}(\lambda )=\frac{\lambda }{2\overline{B}^{(0)}}\left[\sigma ^2\overline{B}^{(0)}+\overline{F}_1^{(0)}+2(1+F_0)\right].$$
(82)
Finally, evaluating the exponent (82) for the optimal value of $`\lambda `$ Eq. (81), we obtain the probability to find an LSD with the spin density amplitude $`\sigma `$
$$\rho (\sigma ^2)\mathrm{exp}\left(\frac{(1+F_0)^2}{2\left(I_{FF}+2\sigma ^2I_{FB}+\sigma ^4I_{BB}\right)}\right),$$
(83)
which for $`\sigma =0`$ coincides with the result Eq. (60) of the linear problem, just as the result Eq. (38) of the qualitative argument above.
The final step is to rewrite the distribution Eq. (83) in terms of the spin $`S`$ of the LSD. Converting the spin density amplitude $`\sigma `$ in Eq. (83) to the spin $`S`$ by means of Eq. (23), we obtain the final expression for the spin distribution
$$\rho (S^2)\mathrm{exp}\left(\frac{\gamma g^2(1+F_0)^2}{12\alpha \frac{S^2}{g^2}+\beta \frac{S^4}{g^4}}\right),$$
(84)
where the numerical factors are
$`\alpha ={\displaystyle \frac{g^2|I_{FB}|}{I_{FF}\left[d^2x\psi (x)\right]^2}}1.92,`$ (86)
$`\beta ={\displaystyle \frac{g^4I_{BB}}{I_{FF}\left[d^2x\psi (x)\right]^4}}7.36,`$ (87)
$`\gamma ={\displaystyle \frac{1}{2g^2I_{FF}}}3.9.`$ (88)
The factor $`\gamma `$ is the same as in the linear problem and is listed here for completeness.
The factors Eq. (IV B) are now independent of the size $`R`$ of the LSD. To see that, one needs to notice that all the correlators $`\widehat{𝒦}`$ depend on this scale in the same way (when a single LSD is considered, so that $`R`$ is the only scale in the integral) $`\widehat{𝒦}R^4`$ \[see Eq. (167)\]. The wave function $`\psi (x)`$ is inverse in $`R`$. Therefore, the integrals Eq. (IV B), unlike the integral Eq. (59), which appears in the solution to the linear problem, do depend on $`R`$, since they contain different number of functions $`\psi (x)`$
$$I_{FB}\frac{R^2}{g^4},$$
(90)
$$I_{BB}\frac{R^4}{g^6}.$$
(91)
The different size dependence of the integrals Eq. (59) and $`I_{FF}`$ is compensated in Eq. (IV B) by additional factors of $`d^2x\psi (x)`$.
The distribution Eq. (84) is thus the same as the qualitative result Eq. (37) only now with coefficients Eq. (IV B). The coefficients were evaluated numerically using the solution $`\psi (x)`$, which follows from the linear problem. The coefficients are positive \[the negative sign of $`I_{FB}`$ being taken into account explicitly in Eq. (84)\]. The coefficient $`\beta `$ is positive, ensuring convergence of the expansion Eq. (37), which again is the way to interpret the distribution Eq. (84). In the calculation leading to Eq. (84) the limitation to small spins follows from the separation procedure Eq. (22) since it is valid only close to the instability, where typical spins are small.
The typical value of spin of the LSD is still given by Eq. (40). The spin $`S`$ turns out to be large, $`1Sg`$, so on length scales larger than $`R`$ or at high enough temperature, LSDs behave as classical moments. However, LSDs are extended objects and can have any size with equal probability (with the exponential accuracy). Thus their spins can not be considered as local moments, especially when discussing their interactions.
## V Interactions between LSDs
The importance of LSDs is that their appearance can dramatically change the magnetic response of the system. At high enough temperatures, we can consider them as independent, classical moments, thus we expect the system to be the usual paramagnet with the susceptibility described by the Curie law (but with Curie constant different form that of the free electron gas). As the temperature becomes smaller, the system might change its ground state in a way that depends on the interaction between LSDs, in particular on its sign and typical range.
Let us first recall the basic physics of local moments in a metallic system. In a system of local moments there are two competing types of interaction. First, there is the direct contact interaction , which sign does not change with distance between the moments, but the amplitude decays exponentially beyond the correlation length. This interaction tends to turn the system into a ferromagnet. Then there is the RKKY interaction , which as a function of distance oscillates and decays only as a power law ($`R^2`$ in 2D). The RKKY interaction between local moments tends to form a spin glass at low temperatures .
Our case is different. As we have shown in this paper, in a metal close to Stoner instability there is a non zero probability for LSDs to be spontaneously formed. This probability is independent of the size of LSDs, so that droplets of all sizes can appear. Therefore interaction between the LSDs can not be described in the same way as interaction between local moments. Rather, it is given by the non-local susceptibility $`F_1(x_1,x_2)`$, averaged over the area of the interacting LSDs. Calculation of the average, previously denoted as $`F_1^{(0)}`$, involves integration of $`F_1(x_1,x_2)`$ with two functions $`\psi (xa)`$ which describe the spatial profile of the LSDs. In this section we use the same Gaussian functions we used to evaluate the integrals Eq. (59) and Eq. (IV B). Only now these functions carry explicitly the dependence on the coordinate $`a`$ of the center of the LSD.
Consider now two LSDs separated by distance $`L`$ much larger than the size of both LSDs $`LR`$, see Fig. 1. In this argument we take both LSDs to be of the same size $`R`$, but it can be easily generalized for the case where the interacting LSDs differ is size substantially. Choosing different combinations of the wavefunctions $`\psi (xa)`$ and $`\psi (yb)`$ describing the two LSDs, we can form three different averages
$`F_1^{(1)}={\displaystyle d^2x_1d^2x_2F_1(x_1,x_2)\psi (x_1a)\psi (x_2a)},`$ (93)
$`F_1^{(2)}={\displaystyle d^2y_1d^2y_2F_1(y_1,y_2)\psi (y_1b)\psi (y_2b)},`$ (94)
$`F_1^{(12)}={\displaystyle d^2xd^2yF_1(x,y)\psi (xa)\psi (yb)}.`$ (95)
These averages, as random quantities, have a Gaussian distribution with the weight determined by the correlation function $`F_1F_1`$. We now assume that the LSDs have been already formed. The saddle point solution $`\overline{F}_1`$ \[see Eq. (52)\] describes a single LSD, while the interactions are determined by small deviations from the saddle point. Therefore the weight of the distribution of the averages Eq. (V) can be evaluated at the “optimal fluctuation” point Eq. (52). Then the distribution can be written as
$`w[`$ $`F_1^{(1)},F_1^{(2)},F_1^{(12)}]`$ (98)
$`\mathrm{exp}\left[{\displaystyle \left(\begin{array}{ccc}F_1^{(1)}& F_1^{(2)}& F_1^{(12)}\end{array}\right)\widehat{}^1\left(\begin{array}{c}F_1^{(1)}\\ F_1^{(2)}\\ F_1^{(12)}\end{array}\right)}\right],`$
with the weight matrix
$`\widehat{}=\left(\begin{array}{ccc}I_{FF}& J_1& J_2\\ J_1& I_{FF}& J_2\\ J_2& J_2& J_3\end{array}\right).`$ (99)
The elements of the weight matrix $`\widehat{}`$ are obtained by averaging the correlation function $`K_{FF}`$ over the area of the LSDs, i.e. integrating with four wavefunctions $`\psi (xa)`$, similar to the integral $`I_{FF}`$ \[see Eq. (59)\]. The difference from the case of the single LSD is that now all but two of the elements of $`\widehat{}`$ depend on two different lengthes : the size of the LSD $`R`$ and the distance $`L`$ between them. Thus the estimate Eq. (167) for the correlation function $`K_{FF}`$ does not apply. The elements $`J_i`$ are given by the integrals
$`J_1=`$ $`{\displaystyle d^2x_1d^2x_2d^2y_1d^2y_2K_{FF}[\{x_j\},\{y_j\}]\psi (x_1a)}`$ (103)
$`\times \psi (x_2a)\psi (y_1b)\psi (y_2b){\displaystyle \frac{0.15}{\pi ^2g^2}}{\displaystyle \frac{R^4}{L^4}},`$
$`J_2={\displaystyle }`$ $`d^2x_1d^2x_2d^2y_1d^2y_2K_{FF}[\{x_j\},\{y_j\}]`$ (108)
$`\times \psi (x_1a)\psi (x_2b)\psi (y_1b)\psi (y_2b)`$
$`{\displaystyle \frac{1}{6\pi ^2g^2}}{\displaystyle \frac{R^4}{L^4}}\mathrm{ln}^2{\displaystyle \frac{R}{L}},`$
$`J_3={\displaystyle }`$ $`d^2x_1d^2x_2d^2y_1d^2y_2K_{FF}[\{x_j\},\{y_j\}]`$ (113)
$`\times \psi (x_1a)\psi (x_2b)\psi (y_1a)\psi (y_2b)`$
$`{\displaystyle \frac{1}{6\pi ^2g^2}}{\displaystyle \frac{R^4}{L^4}}\mathrm{ln}^2{\displaystyle \frac{R}{L}}.`$
The integrals Eq. (V) were calculated in the leading order in $`R/L`$. Clearly, $`J_2J_3`$ exactly, but the difference comes in the numerical factor under the logarithm, which we here neglect. This does not have any bearing on our conclusions. The $`1/L^4`$ dependence of all the integrals follows from the frequency integral in Eq. (158), which is determined by the Thouless energy corresponding to the largest length in the problem, which is now $`L`$.
The integral $`I_{FF}`$ is independent of all length scales and therefore is much larger than any of $`J_i`$. Thus the weight matrix $`\widehat{}^1`$ in Eq. (98) to the leading order in $`R/L`$ is
$`\widehat{}^10.8\pi ^2g^2\left(\begin{array}{ccc}1& 0& 1\\ 0& 1& 1\\ 1& 1& f^1\end{array}\right),`$ (114)
where the dimensionless function $`f(R/L)`$ is given by
$$f(R/L)0.72\frac{R^4}{L^4}\mathrm{ln}^2\frac{R}{L}.$$
(115)
To describe the interaction between LSDs we need to find the distribution of $`F_1^{(12)}`$ under the condition that the two LSDs exist, namely that $`F_1^{(1)}<0`$ and $`F_1^{(2)}<0`$. This is given by the conditional probability distribution
$`W[F_1^{(12)}]`$ $`={\displaystyle \frac{1}{N^2}}{\displaystyle 𝒟[F_1^{(1)},F_1^{(2)}]\theta (1F_0F_1^{(1)})}`$ (119)
$`\times \theta (1F_0F_1^{(2)})w[F_1^{(1)},F_1^{(2)},F_1^{(12)}],`$
$`N={\displaystyle 𝒟[F_1^{(1)}]e^{\frac{(F_1^{(1)})^2}{I_{FF}}}\theta (1F_0F_1^{(1)})},`$ (120)
where $`w`$ is the distribution Eq. (98). Since the weight matrix in Eq. (98) is a $`c`$-number, the integrals in Eq. (V) are usual Gaussian integrals and not functional integrals. The $`\theta `$-functions in Eq. (V) make the exact integration in terms of elementary functions impossible, but we can use the small parameter $`f(R/L)1`$ to estimate $`W[F_1^{(12)}]`$ with exponential accuracy, which is all we need to describe interaction between LSDs. Up to the pre-exponential factor
$$W[F_1^{(12)}]\mathrm{exp}\left[\frac{0.8\pi ^2g^2}{f}\left(F_1^{(12)}+2(1+F_0)f\right)^2\right],$$
(121)
so that the distribution is a Gaussian (as it should be since we considered $`F_1`$ to be a Gaussian random quantity from the very beginning).
The average $`F_1^{(12)}`$ given by the distribution Eq. (121) is shifted from zero to the negative value $`2f(R/L)(1+F_0)`$, i.e. the average interaction appears to be ferromagnetic. However, the distribution Eq. (121) also allows for strong fluctuations of $`F_1^{(12)}`$. These fluctuations can be estimated as
$$\frac{(\delta F_1^{(12)})^2}{F_1^{(12)}^2}\frac{1}{g^2(1+F_0)^2f}.$$
(122)
The function $`f`$ \[see Eq. (115)\] decreases with the distance between LSDs. Therefore at large enough distances the fluctuations of $`F_1^{(12)}`$ exceed the average and the sign of $`F_1^{(12)}`$ becomes random.
The cross-over distance $`L^{}`$ can be estimated as (from the condition $`(\delta F_1^{(12)})^2/F_1^{(12)}^21`$)
$$L^{}R\sqrt{g(1+F_0)}R.$$
(123)
This distance has to be compared with the typical distance between LSDs. The latter can be estimated as follows. The concentration of LSDs has to be proportional to the probability Eq. (60)
$$n\frac{1}{R^2}e^{\gamma g^2(1+F_0)^2},$$
(124)
where $`R`$ is the characteristic length of a LSD. Therefore the typical distance between LSDs is exponentially large in the parameter $`g(1+F_0)1`$, while the cross-over length $`L^{}`$ is large only as a power of the same parameter. Thus, the interaction between typical LSDs has random sign.
The energy of interaction between typical LSDs decays as the second power of the distance between them
$$U_t=\frac{1}{\nu gL^2}\frac{1}{\nu gR^2}e^{\gamma g^2(1+F_0)^2},$$
(125)
where $`\nu `$ is the density of states and we have estimated the typical distance between LSDs from Eq. (124).
The results of this Section resemble the results of Ref. , where the electron-mediated interaction between magnetic moments in two ferromagnets separated by a disordered metal was also found to have a random sign. The diference between our model and that of Ref. is that in our case the ferromagnetic regions (LSDs) are created by the very same impurity configurations as those responsible for the interaction between LSDs. Therefore, the interaction between LSDs can be considered only under the condition of their existance and thus is characterized by the conditional probability distribution Eq. (121). As a result, the average $`F_1^{(12)}`$ is negative, i.e. ferromagnetic, and dominates the fluctuations at distances smaller than $`L^{}`$. However, since the typical distance between LSDs is exponentially large, the typical interaction has random sign.
This fact is not surprising, since we are dealing with a disordered system where one could expect to find some spin glass phase (at $`T=0`$, similar to the case of a superconductor in weak magnetic field, considered in Ref. . In the model considered in Ref. mesoscopic fluctuations become uncorrelated beyond the magnetic length. In our case the role similar to that of the magnetic field in Ref. would have been played by the spin-orbit coupling, which we have so far neglected. However, taking the spin-orbit coupling into account does not change the main results of this paper. As we have shown above, the typical interaction between LSDs is random due to large fluctuations in Eq. (121). Should we include the spin-orbit coupling, we would need to compare the spin-orbit length to $`L^{}`$ in order to determine the length beyond which the interaction becomes random. But since the typical distance between LSDs is exponentially larger than $`L^{}`$ the exact value of such cross-over length is not very important.
## VI contribution to physical quantities
In this section we estimate the contribution of the LSDs to observable physical quantities. We consider the examples of paramagnetic susceptibility and the dephasing time. To be observable, the contribution of LSDs should exceed the regular contribution of the electron system. We show that, for both quantities, it happens at the same temperature $`T^{}`$, below which the temperature dependence of both quantities changes. The Pauli susceptibility crosses over to a Curie-type $`1/T`$ dependence, while the dephasing time saturates and becomes temperature independent.
Since the typical interaction energy Eq. (125) is exponentially small, LSDs behave as weakly interacting moments and their contribution to the paramagnetic susceptibility is given by the Curie law $`\chi _{LSD}=C/T`$, where the Curie constant $`C`$ is proportional to the square of the spin $`S`$ of the LSD \[see Eq. (40)\] and the density (or concentration) of LSDs Eq. (124). We have calculated the probability Eq. (60) with the exponential accuracy, therefore the pre-exponential factor in the density of LSDs should be given by the typical length characterizing the distribution of LSDs. This dependence could be elucidated from the following dimensional argument. The probability to find an LSD in unit volume (or, rather, area, which we still denote by $`V`$) of size in between $`R`$ and $`R+dR`$, with spin in between $`S`$ and $`S+dS`$ is given by the distribution
$$d𝒲=\rho (S^2)dVdSd^2R$$
(126)
Clearly, the dimensionality of $`\rho (S^2)`$ is $`L^4`$, where $`L`$ is the characteristic length. Since the exponential in $`\rho (S^2)`$ \[see Eq. (84)\] is independent of length, the size of a LSD can not be pinned to any length in the system. Since individual LSDs contribute to Curie susceptibility independently, we need to sum over all possible sizes and given the $`L^4`$ dependence, the main contribution comes from LSDs of the smallest possible size, namely the mean-free path $`l`$. Thus, contribution to the susceptibility (up to a numerical coefficient) is
$$\chi _{LSD}\frac{1}{\tau T}\frac{\nu }{g(1+F_0)^2}e^{\gamma g^2(1+F_0)^2},$$
(127)
where $`\tau `$ is the scattering time. Here we used the typical value (23) of the spin of a LSD. This contribution has to be compared with the Pauli susceptibility Eq. (1), which is temperature independent. As a result, the contribution Eq. (127) dominates at temperatures smaller than $`T^{}`$, which up to numerical factors is
$$T^{}\frac{1}{\tau }\frac{1}{g(1+F_0)}e^{\gamma g^2(1+F_0)^2}.$$
(128)
This temperature has to be compared with the typical interaction energy Eq. (125). Substituting the mean-free path $`l`$ for the chracteristic size $`R`$ in Eq. (125) we find the ratio $`U_t/T^{}`$ to be
$$\frac{U_t}{T^{}}\frac{(1+F_0)^2}{g(1+F_0)}1.$$
(129)
Therefore, the Curie behavior Eq. (127) persists over a wide temperature range $`U_t<T<T^{}`$.
Similarly, we estimate the contribution of LSDs to the dephasing time $`\tau _\phi `$. LSDs are regions where the impurity configuration makes it energetically favorable for the electrons to align their spins. Some other electron entering such region will “feel” the overall polarization as if it were magnetic field. The corresponding dephasing time can be estimated as the square of the Zeeman energy divided by the Thouless energy (at the size of the LSD). More formally, since the interaction with the polarization stems from the exchange interaction, we can find $`\tau _\phi `$ from the perturbation theory
$`{\displaystyle \frac{1}{\tau _\phi }}={\displaystyle \frac{F_0^2}{\nu ^2}}{\displaystyle \frac{d^2r_1}{V}\frac{d^2r_2}{V}\sigma (r_1)\sigma (r_2)𝒟(r_1r_2)},`$
where $`𝒟(r_1r_2)`$ is the (electron) diffusion propagator and $`\sigma (r_1)\sigma (r_2)`$ is the correlation function of the spins of LSDs. We can now estimate $`\tau _\phi `$ as
$$\frac{1}{\tau _\phi }\frac{F_0^2}{\nu ^2}\frac{nS^2}{D}\frac{F_0^2}{\tau }\frac{1}{g^2(1+F_0)^2}e^{\gamma g^2(1+F_0)^2},$$
(130)
where again the dominant contribution comes from LSDs of smallest size \[$`n`$ is the concentration of LSDs, see Eq. (124)\]. This should be compared with the contribution of the usual Gaussian spin fluctuations
$$\frac{1}{\tau _\phi ^s}\frac{2F_0^2}{(1+F_0)(2+F_0)}\frac{T}{g}\mathrm{ln}g(1+F_0).$$
(131)
Again, up to the numerical factors, the contribution of LSDs dominates at temperatures lower than $`T^{}`$ \[given by Eq. (128)\].
The actual crossover temperatures for different physical quantities might differ by a factor of order $`\mathrm{ln}g(1+F_0)`$, but such difference is beyond the accuracy of our treatment. However, the discussed contribution of LSDs to physical quantities suggests that our scenario of the magnetic fluctuations in a metal close to the Stoner instability can be experimentally observed. The LSDs lead to the saturation of the dephasing time at low temperatures. If such saturation is observed, one should look at the behavior of the paramagnetic susceptibility in the same temperature range. If the LSDs are present in the system, then the onset of the Curie-like temperature dependence should also be detected.
## VII conclusions
In this paper we have considered the effect of disorder on magnetic properties of the ground state of a metal close to the Stoner instability. We have shown that even though on the mean field level the ground state of the metal is paramagnetic ($`1+F_0>0`$), there is non zero (exponentially small) probability to form local spin droplets, i.e. domains of non zero spin polarization. The probability to form a LSD is independent of its size $`R`$, thus LSDs of any size can appear. The total spin of the LSD is also independent of its size and obeys the distribution Eq. (38), with the typical value $`S(1+F_0)^1`$, which is large $`1Sg`$, so that LSDs can be considered as classical spins.
Considered as independent moments, LSDs contribute to the observables, changing the temperature dependence of both the paramagnetic susceptibility and the dephasing time at temperatures lower than certain cross-over temperature Eq. (128). When $`T<T^{}`$, the dephasing time saturates to the temperature-independent value Eq. (130), while the susceptibility acquires the Curie-like $`1/T`$ dependence.
Both the Curie susceptibility Eq. (127) and the dephasing time Eq. (130) were obtained in approximation of non-interacting LSDs. This approach is valid at temperatures larger than the typical value (125) of the interaction between LSDs. Since the cross-over temperature $`T^{}`$ is much larger than the interaction (125), there is a parametrically wide temperature regime \[by our large parameter $`g(1+F_0)`$, see Eq. (129)\], $`U_t<T<T^{}`$, where the Curie behavior of the susceptibility and the saturation of the dephasing time can be observed. At smaller temperatures $`T<U_t`$, however, the LSDs can not be considered as non-interacting moments and the behavior of the system changes. Interaction of LSDs with each other or with itinerant electrons should lead either to screening of the local spins or to forming some spin glass state (due to the random sign of the interaction). Such regime was not considered in this paper..
We acknowledge helpful discussions with L.B. Ioffe and B. Spivak. We also acknowledge the hospitality of Theoretische Physik III, Ruhr-Universität Bochum, where this work was started. I.A. is Packard research fellow. A.L. is supported by the NSF under Grant 9812340.
##
Here we discuss the correlation functions that describe the mesoscopic fluctuations in the system, i.e. give the weight in the distribution Eq. (21) of the random quantities $`F_1`$ and $`B`$. Both quantities are the coefficients in the expansion of the thermodynamic potential Eq. (LABEL:tpnl) in powers of the spin density. Therefore the diagrams for $`F_1`$ and $`B`$ can be obtained by differentiating the diagram for the exact thermodynamic potential. To calculate the correlators one has to multiply the random quantities and then average over the disorder. As a result we get three different correlators, depicted diagrammatically on Figs. 2 - 4.
In terms of exact electronic Green’s functions the thermodynamic potential can be written as
$$\mathrm{\Omega }=\frac{dϵ}{2\pi i}f(ϵ)\mathrm{𝐓𝐫}\mathrm{ln}\frac{G^R(ϵ)}{G^A(ϵ)},$$
(132)
where the Green’s function is defined in the fluctuating field $`\stackrel{}{\sigma }`$, i.e.
$$G^{R(A)}(ϵ)=\left[ϵ\widehat{}\stackrel{}{\tau }\stackrel{}{\sigma }\pm i0\right]^1.$$
(133)
Here $`\widehat{}`$ is the Hamiltonian of the system, $`\tau ^j`$ are the Pauli matrices. The symbol $`\mathrm{𝐓𝐫}`$ denotes the trace over spin indices and the integration over all spatial cooridinates. For brevity we do not explicitly indicate that $`G^{R(A)}(ϵ)`$ depend on spatial coordinates (due to disorder in $`\widehat{}`$).
The expansion (7) can be achieved by taking variational derivatives of the thermodynamic potential (132) with respect to $`\stackrel{}{\sigma }`$. The second variational derivative of the thermodynamic potential corresponds to the second order term in Eq. (7),
$`{\displaystyle \frac{\delta ^2\mathrm{\Omega }}{\delta \sigma ^\alpha (x_1)\delta \sigma ^\beta (x_2)}}`$ $`={\displaystyle \frac{dϵ}{2\pi i}f(ϵ)}`$ (136)
$`\times \mathrm{𝐓𝐫}^{}\left[G^R\tau ^\alpha G^R\tau ^\beta G^A\tau ^\alpha G^A\tau ^\beta \right].`$
Here the prime in $`\mathrm{𝐓𝐫}^{}`$ means that there is no integration over the coordinates in the left-hand side (in this case $`x_1`$ and $`x_2`$). The trace over the Pauli matrices gives (in the absence of spin-orbit coupling) $`\mathrm{Tr}[\tau ^\alpha \tau ^\beta ]=2\delta ^{\alpha \beta }`$. Upon subtracting the average, the derivative (136) is proportional to the non-local quantity $`F_1(x_1,(x_2)`$:
$`F_1(x_1,x_2)=`$ $`{\displaystyle \frac{\delta ^2\mathrm{\Omega }}{\delta \sigma ^\alpha (x_1)\delta \sigma ^\beta (x_2)}}`$ (139)
$`{\displaystyle \frac{\delta ^2\mathrm{\Omega }}{\delta \sigma ^\alpha (x_1)\delta \sigma ^\beta (x_2)}}.`$
Similarly, the coefficient $`B`$ in Eq. (7) is given by the fourth variational derivative of the thermodynamic potential (132) (only the average is zero in this case):
$`B[\{x_i\}]={\displaystyle \frac{dϵ}{2\pi i}}`$ $`f(ϵ)\mathrm{𝐓𝐫}^{}[G^R\tau ^\alpha G^R\tau ^\beta G^R\tau ^\mu G^R\tau ^\nu `$ (142)
$`G^A\tau ^\alpha G^A\tau ^\beta G^A\tau ^\mu G^A\tau ^\nu ].`$
The trace over the Pauli matrices has two parts, $`\mathrm{Tr}[\tau ^\alpha \tau ^\beta \tau ^\mu \tau ^\nu ]=2[\delta ^{\alpha \beta }\delta ^{\mu \nu }(\delta ^{\alpha \mu }\delta ^{\beta \nu }\delta ^{\alpha \nu }\delta ^{\beta \mu })]`$. The first part corresponds to the fourth order term in Eq. (7), while the second part is the cross-product term, which disappears since we consider only the singlet fluctuations $`\stackrel{}{\sigma }=(0,0,\sigma )`$.
We are now ready to calculate correlation functions of the mesoscopic fluctuations $`F_1`$ and $`B`$. Averaging over disorder is performed in a standard manner (see Ref. for details). The correlator of the fluctuation $`F_1(x_1,x_2)`$
$`K_{FF}`$ $`[\{x_j\},\{y_j\}]=`$ (145)
$`K_{FF}(x_1,x_2,y_1,y_2)=F_1(x_1,x_2)F_1(y_1,y_2)`$
is given by a sum of four diagrams, one of which is depicted on Fig. 2.
All four diagrams contribute equally to the integral Eq. (59) (as will be clear from the explicit form of $`K_{FF}`$), therefore we shall proceed evaluating the contribution of the diagrams on Fig. 2. After averaging, each diagram corresponds to a product of four diffusion propagators (or diffusons) and four vertex blocks.
The diffuson in momentum representation is given by
$`𝒟(\omega ;q)={\displaystyle \frac{1}{i\omega +Dq^2}},`$
where $`D`$ is the diffusion constant, and $`q`$ is a 2D momentum vector. Each vertex block (see Fig. 4) contains precisely the same factors $`2\pi \nu \tau ^2`$ as each diffuson, thus in momentum representation the diagrams on Fig. 2 are expressed in terms of the integral
$`K_{FF}`$ $`[\{k_j\}]={\displaystyle \underset{\mathrm{}}{\overset{0}{}}}{\displaystyle \frac{dϵ_1}{\pi \nu }}{\displaystyle \underset{\mathrm{}}{\overset{0}{}}}{\displaystyle \frac{dϵ_2}{\pi \nu }}{\displaystyle d^2Q𝒟(\omega ;Q)}`$ (150)
$`\times 𝒟(\omega ;Qk_1)𝒟(\omega ;Qk_3)𝒟(\omega ;Qk_1k_4)`$
$`\times \delta (k_1+k_4k_2k_3),`$
where $`\omega =ϵ_1ϵ_2`$.
For the purposes of this paper it is more convenient to write the correlator Eq. (145) in the coordinate representation as (we have also evaluated one of the frequency integrals)
$`K_{FF}`$ $`[\{x_j\},\{y_j\}]={\displaystyle \underset{\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \frac{d\omega }{2\pi ^2\nu ^2}}|\omega |𝒟(\omega ;y_1x_1)`$ (153)
$`\times 𝒟(\omega ;x_1y_2)𝒟(\omega ;x_2y_1)𝒟(\omega ;y_2x_2),`$
where $`x_i`$ and $`y_i`$ are 2D coordinate vectors and
$$𝒟(\omega ;x)=\frac{d^2q}{(2\pi )^2}𝒟(\omega ;q)e^{iqx}$$
(154)
is the Fourier transform of the diffusion denominator to the position space.
The remaining frequency integral can be evaluated by using the following integral representation for $`𝒟(\omega ;x)`$. First, we represent the diffusion denominator as an integral over an auxiliary variable
$`{\displaystyle \frac{1}{i\omega +Dq^2}}={\displaystyle \underset{0}{\overset{\mathrm{}}{}}}𝑑t\mathrm{exp}\left[t(i\omega +Dq^2)\right].`$
The momentum integral in Eq. (154) becomes Gaussian and we obtain
$`𝒟(\omega ;x)={\displaystyle \frac{1}{4\pi D}}{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{dt}{t}}\mathrm{exp}\left[i\omega t{\displaystyle \frac{x^2}{4tD}}\right].`$ (155)
Substituting the integral representation Eq. (155) into the correlator Eq. (153), we find the final expression for the contribution of the diagram on Fig. 2 to the correlator Eq. (145)
$`K_{FF}[\{x_j\},\{y_j\}]={\displaystyle \frac{1}{(2\pi \nu D)^2}}{\displaystyle \frac{1}{4\pi ^4}}{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{dt_1dt_2dt_3dt_4}{t_1t_2t_3t_4}}`$ (156)
(157)
$`{\displaystyle \frac{\mathrm{exp}\left[\frac{(x_1y_1)^2}{t_1}\frac{(x_2y_1)^2}{t_2}\frac{(x_1y_2)^2}{t_3}\frac{(x_2y_2)^2}{t_4}\right]}{(t_1+t_2+t_3+t_4)^2}}`$ (158)
To estimate the size of the fluctuation regions with non zero spin (LSD) we need the explicit dependence of the correlator Eq. (158) on the parameters of the problem. To do that we introduce a length scale $`R`$ which characterizes the size if the LSD and write Eq. (158) as
$`K_{FF}`$ $`[\{x_j\},\{y_j\}]={\displaystyle \frac{1}{g^2R^4}}\stackrel{~}{K}_{FF}`$ (159)
where $`\stackrel{~}{K}_{FF}`$ is the dimensionless counterpart of the integral Eq. (158).
The remaining correlators in Eq. (21) are constructed in the same manner as Eq. (158), the only difference being the number of diffusons. The corresponding diagrams are given in Fig. 3 and 4. Again, the total number of diagrams is large, therefore we give expressions for the typical contributions depicted on Figs. 3 and 4. The rest of the diagrams are obtained by interchanging coordinate indices.
The diagram on Fig. 3 contains six diffusons and six vertex blocks. Since each of the vertex blocks carries a factor of $`i`$, the overall sign of the diagram is negative. Using the integral representation Eq. (155) for diffusion denominators, we find
$`K_{FB}[\{x_j\},\{y_j\}]=`$ $`{\displaystyle \frac{1}{2(2\pi )^4}}{\displaystyle \frac{1}{(2\pi D)^4}}`$ (162)
$`\times {\displaystyle \underset{0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{k=1}{\overset{6}{}}}{\displaystyle \frac{dt_k}{t_k}}{\displaystyle \frac{\mathrm{exp}[T_{FB}]}{(\underset{i=1}{\overset{6}{}}t_i)^2}}`$
$`T_{FB}=`$ $`{\displaystyle \frac{(x_1y_3)^2}{t_1}}+{\displaystyle \frac{(x_1y_4)^2}{t_2}}+{\displaystyle \frac{(x_2y_2)^2}{t_3}}`$
$`+{\displaystyle \frac{(x_2y_1)^2}{t_4}}+{\displaystyle \frac{(y_1y_4)^2}{t_5}}+{\displaystyle \frac{(y_3y_2)^2}{t_6}}`$
The correlator on Fig. 4 can be written in the same way. It has positive sign, since it contains eight vertex blocs.
$`K_{BB}[\{x_j\},\{y_j\}]=`$ $`{\displaystyle \frac{1}{8(2\pi )^4}}{\displaystyle \frac{1}{(2\pi D)^6}}`$ (165)
$`\times {\displaystyle \underset{0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{k=1}{\overset{8}{}}}{\displaystyle \frac{dt_k}{t_k}}{\displaystyle \frac{\mathrm{exp}[T_{BB}]}{(\underset{i=1}{\overset{8}{}}t_i)^2}}`$
$`T_{BB}=`$ $`{\displaystyle \frac{(x_1y_3)^2}{t_1}}+{\displaystyle \frac{(x_1y_4)^2}{t_2}}+{\displaystyle \frac{(x_2y_2)^2}{t_3}}`$
$`+{\displaystyle \frac{(x_2y_1)^2}{t_4}}+{\displaystyle \frac{(x_3y_3)^2}{t_5}}+{\displaystyle \frac{(x_3y_2)^2}{t_6}}`$
$`+{\displaystyle \frac{(x_4y_1)^2}{t_7}}+{\displaystyle \frac{(x_4y_4)^2}{t_8}}`$
Altogether, these correlators can be combined in the matrix Eq. (21)
$`\widehat{𝒦}=\left(\begin{array}{cc}K_{FF}[\{x_j\},\{y_j\}]& K_{FB}[\{x_j\},\{y_j\}]\\ K_{FB}[\{x_j\},\{y_j\}]& K_{BB}[\{x_j\},\{y_j\}]\end{array}\right).`$ (166)
Similarly to Eq. (159), the dimensional analysis gives the size dependence in terms of the scale $`R`$
$`\widehat{𝒦}={\displaystyle \frac{1}{g^2R^4}}\left(\begin{array}{cc}\stackrel{~}{K}_{FF}& \frac{1}{g^2}\stackrel{~}{K}_{FB}\\ \frac{1}{g^2}\stackrel{~}{K}_{FB}& \frac{1}{g^4}\stackrel{~}{K}_{BB}\end{array}\right).`$ (167)
The sign of the correlator follows from the diagrams, so that the dimensionless integrals $`\stackrel{~}{K}`$ are positive numbers.
|
warning/0005/gr-qc0005082.html
|
ar5iv
|
text
|
# 1 INTRODUCTION
## 1 INTRODUCTION
The problem of determining the motion of a system of $`N`$ particles mutually interacting through specified forces is one that has attracted attention since the dawn of physics. We continue in this paper our previous explorations of this problem in two spacetime dimensions when the specified interactions are both gravitational and electromagnetic.
Our work here represents the first exact solution to include both interactions in a relativistic framework. Although an exact solution is known in three spatial dimensions for pure Newtonian gravity in the $`N=2`$ case, dissipation of energy in the form of gravitational radiation has obstructed progress toward obtaining exact solutions for the motion of $`N`$ bodies in the general theory of relativity, even when $`N=2`$. However by reducing the number of spatial dimensions this obstruction disappears, at least in the vacuum. Apart from the absence of gravitational radiation, most (if not all) of the remaining conceptual features of relativistic gravity are retained, and so lower dimensional theories of gravity offer the hope of garnering insight into the nature of classical and quantum gravitation in a wide variety of physical situations.
For these reasons we have been investigating the $`N`$-body problem in two dimensional gravity for the past 3 years. We have chosen to work with a 2D theory that models 4D general relativity in that it sets the Ricci scalar equal to the trace of the stress-energy of prescribed matter fields and sources. Hence matter governs the evolution of spacetime curvature which reciprocally governs the evolution of matter . This theory (sometimes referred to as $`R=T`$ theory) has a consistent Newtonian limit , a problematic limit in a generic $`(1+1)`$-dimensional theory of gravity theory . When the stress-energy is that of a cosmological constant, the theory reduces to Jackiw-Teitelboim (JT) theory .
Working in the canonical formalism , we have so far been able to obtain exact solutions to the two body problem in the absence and presence of a cosmological constant. In this paper we extend these considerations to include charged bodies. Specifically, we formulate the charged $`N`$-body problem in relativistic gravity by taking the matter action to be that of $`N`$ charged point-particles minimally coupled to gravity and electromagnetism. We extend the previous canonical formalism we developed for this action in $`R=T`$ lineal gravity to include this case. When $`N=2`$ we obtain exact solutions for the motion of two bodies of unequal (and equal) charge and mass. In the slow motion, weak field limit the Hamiltonian we obtain coincides with that of Newtonian gravity with electromagnetism in $`(1+1)`$ dimensions. We are also able to extend our solutions to include a cosmological constant $`\mathrm{\Lambda }`$, so that in the limit where all bodies are massless and neutral, spacetime has constant curvature (ie the JT theory is obtained). Our solution is the first non-static exact solution to the charged 2-body problem in any relativistic theory of gravity.
Our exact solution to the $`N=2`$ case can be formulated in several ways. We derive an exact solution for the Hamiltonian as a function of the proper separation and the centre-of-inertia momentum of the charged bodies. We are also able to construct a solution in which the proper separation of the two charged point masses is given as a function of their mutual proper time in the equal mass case. If the masses are not equal our exact solution is given in terms of a time-coordinate that is not the proper time.
A scalar (dilaton) field must be included in the action since the Einstein action is a topological invariant in (1+1) dimensions. Canonically reducing the action, we find that the Hamiltonian is given in terms of a spatial integral of the second derivative of the dilaton field, which is a function of the canonical variables of the particles (coordinates and momenta) and is determined from the constraint equations. The matching conditions of the solution to the constraint equations yield an equation which determines the Hamiltonian in terms of the remaining degrees of freedom of the system when $`N=2`$ . We refer to this transcendental equation as the determining equation, since it allows one to determine the Hamiltonian in terms of the centre of inertia momentum and proper separation of the bodies. From this Hamiltonian we can derive the canonical equations of motion. In the equal mass case we find that the separation and momentum are given by differential equations in terms of the proper time, and can be exactly solved in terms of hyperbolic and/or trigonometric functions.
Several different types of motion are expected in the 2 body case, depending upon the signs and magnitudes of the masses, charges, energy and other parameters (e.g. gravitational coupling constant, cosmological constant). If the charges are of opposite sign the particles will remain bounded, whereas if they are of the same sign either bounded or unbounded motion can occur for the same value of the total energy. For a given set of parameters there is in this case a countably infinite series of unbounded motions labelled by an integer $`n`$. A balance condition exists between the bounded and the unbounded cases, and reduces to the expected (Newtonian-like) static balance condition in the absence of particle momenta. We shall analyze these various states of motion, and discuss the transitions which occur between them. A cosmological constant can qualitatively change the motion, rendering bound states unbound and vice-versa. We find several surprising situations, including the diverging separation of the two bodies at finite proper time in the repulsive case, even for $`\mathrm{\Lambda }=0`$. In the $`\mathrm{\Lambda }<0`$ case the motion shows a double maximum behavior for a certain range of the parameters, which is a characteristic effect of the induced momentum dependent $`\mathrm{\Lambda }`$ potential. For classification of the motion we utilize a charge-momentum diagram from which we can easily predict what type of the motion is realized for a given charge.
In the unequal mass case the proper time is no longer the same for the two particles, and a more careful analysis is necessary in order to describe the motion. We obtain phase space trajectories from the determining equation and explicit solutions for the proper separation in terms of a transformed time coordinate which reduces to the mutual proper time in the case of equal mass.
In Sec.II we describe the outline of the canonical reduction of the theory when the charges are included and define the Hamiltonian for the $`N`$\- body system. In Sec.III we solve the constraint equations for the two-body case and get the determining equation of the Hamiltonian, from which the canonical equations of motion are explicitly derived. For the case of equal masses and arbitrary charges the explicit exact solutions to the canonical equations are given in Sec.IV. By using these exact solutions we analyze in Sec.V the motion for $`\mathrm{\Lambda }=0`$ separately for four cases: attractive charges, small repulsive charges, large repulsive charges and $`H<2m`$. We analyze the motion of equal masses for $`\mathrm{\Lambda }0`$ in Sec.VI for four possible combinations of the signs of $`\mathrm{\Lambda }`$ and the charges, where we also develop the plots of phase space trajectories and the analysis of the explicit solutions in terms of the proper time, based on a classification given by a charge-momentum diagram. We treat the unequal mass case in Sec.VII. In Sec.VIII we investigate the static balance problem by using both the canonical equation and the determining equation. Sec.IX contains concluding remarks and directions for further work. The solution of the metric tensor, a linear approximation of the exact solutions and the causal relationships between particles in unbounded motion are given in Appendices.
## 2 CANONICALLY REDUCED HAMILTONIAN OF $`N`$-CHARGED PARTICLES
Our derviation of the canonically reduced Hamiltonian for charged particles is parallel to that given in the uncharged case . Here we briefly review this work, highlighting those aspects that are peculiar to the charged case.
The action integral for gravitational and electromagnetic fields coupled with $`N`$ charged point masses is
$`I`$ $`=`$ $`{\displaystyle }d^2x[{\displaystyle \frac{1}{2\kappa }}\sqrt{g}g^{\mu \nu }\{\mathrm{\Psi }R_{\mu \nu }+{\displaystyle \frac{1}{2}}_\mu \mathrm{\Psi }_\nu \mathrm{\Psi }+{\displaystyle \frac{1}{2}}g_{\mu \nu }\mathrm{\Lambda }\}`$ (1)
$`A_\mu _{,\nu }^{\mu \nu }+{\displaystyle \frac{1}{4\sqrt{g}}}^{\mu \nu }^{\alpha \beta }g_{\mu \alpha }g_{\nu \beta }`$
$`+{\displaystyle \underset{a}{}}{\displaystyle }d\tau _a\{m_a(g_{\mu \nu }(x){\displaystyle \frac{dz_a^\mu }{d\tau _a}}{\displaystyle \frac{dz_a^\nu }{d\tau _a}})^{1/2}+e_a{\displaystyle \frac{dz_a^\mu }{d\tau _a}}A_\mu (x)\}\delta ^2(xz_a(\tau _a))],`$
where $`\mathrm{\Psi }`$ is the dilaton field, $`A_\mu `$ and $`^{\mu \nu }`$ are the vector potential and the field strength, $`g_{\mu \nu }`$ and $`g`$ are the metric and its determinant, $`R`$ is the Ricci scalar, and $`e_a`$ and $`\tau _a`$ are the charge and the proper time of $`a`$-th particle, respectively, with $`\kappa =8\pi G/c^4`$. The symbol $`_\mu `$ denotes the covariant derivative associated with $`g_{\mu \nu }`$.
The field equations derived from the action (1) are
$`Rg^{\mu \nu }_\mu _\nu \mathrm{\Psi }=0,`$ (2)
$`{\displaystyle \frac{1}{2}}_\mu \mathrm{\Psi }_\nu \mathrm{\Psi }{\displaystyle \frac{1}{4}}g_{\mu \nu }^\lambda \mathrm{\Psi }_\lambda \mathrm{\Psi }+g_{\mu \nu }^\lambda _\lambda \mathrm{\Psi }_\mu _\nu \mathrm{\Psi }=\kappa T_{\mu \nu }+{\displaystyle \frac{1}{2}}g_{\mu \nu }\mathrm{\Lambda },`$ (3)
$`_{,\nu }^{\mu \nu }={\displaystyle \underset{a}{}}e_a{\displaystyle 𝑑\tau _a\frac{dz_a^\mu }{d\tau _a}\delta ^2(xz_a(\tau _a))},`$ (4)
$`{\displaystyle \frac{1}{\sqrt{g}}}_{\mu \nu }=_\mu A_\nu _\nu A_\mu ,`$ (5)
$`m_a\left[{\displaystyle \frac{d}{d\tau _a}}\left\{g_{\mu \nu }(z_a){\displaystyle \frac{dz_a^\nu }{d\tau _a}}\right\}{\displaystyle \frac{1}{2}}g_{\nu \lambda ,\mu }(z_a){\displaystyle \frac{dz_a^\nu }{d\tau _a}}{\displaystyle \frac{dz_a^\lambda }{d\tau _a}}\right]=e_a{\displaystyle \frac{dz_a^\nu }{d\tau _a}}\left\{A_{\nu ,\mu }(z_a)A_{\mu ,\nu }(z_a)\right\},`$ (6)
where the stress-energy due to the point masses and the electric field is
$$T_{\mu \nu }=\underset{a}{}m_a𝑑\tau _a\frac{1}{\sqrt{g}}g_{\mu \sigma }g_{\nu \rho }\frac{dz_a^\sigma }{d\tau _a}\frac{dz_a^\rho }{d\tau _a}\delta ^2(xz_a(\tau _a))+\frac{1}{(g)}\left\{_{\mu \alpha }_{\nu \beta }g^{\alpha \beta }\frac{1}{4}g_{\mu \nu }_{\alpha \beta }^{\alpha \beta }\right\},$$
(7)
recalling that in (1+1) dimensions no magnetic component of the field exists. Eq.(3) guarantees the conservation of $`T_{\mu \nu }`$. Inserting the trace of Eq.(3) into Eq.(2) yields
$$R\mathrm{\Lambda }=\kappa T_\mu ^\mu .$$
(8)
Eqs.(4), (5), (6) and (8) form a closed sytem of equations for gravity, electromagnetism, and matter.
On transforming the action (1) to the canonical form we first rewrite the electromagnetic and the particle sectors in a form appropriate to the first-order formalism .
$`I_{E+P}`$ $`=`$ $`{\displaystyle }dx^2[A_\mu _{,\nu }^{\mu \nu }+{\displaystyle \frac{1}{4\sqrt{g}}}^{\mu \nu }^{\alpha \beta }g_{\mu \alpha }g_{\nu \beta }`$ (9)
$`+{\displaystyle \underset{a}{}}{\displaystyle }d\tau _a\{\pi _{a\mu }{\displaystyle \frac{dz_a^\mu }{d\tau _a}}{\displaystyle \frac{1}{2}}\lambda _a^{}(\tau _a)(\pi _{a\mu }\pi _{a\nu }g^{\mu \nu }+m_a^2)+e_a{\displaystyle \frac{dz_a^\mu }{d\tau _a}}A_\mu (x)\}\delta ^2(xz_a(\tau _a))],`$
where $`\pi _a^\mu `$ is essentially the four velocity of the particle and $`\lambda _a^{}`$ is a Lagrange multiplier. The variations with respect to $`A_\mu `$ and $`^{\mu \nu }`$ lead to Eqs.(4) and (5), respectively, and those with respect to $`\pi _{a\mu },z_a^\mu `$ and $`\lambda _a^{}`$ lead to
$`{\displaystyle \frac{dz_a^\mu }{d\tau _a}}\lambda _a^{}\pi _a^\mu =0,`$ (10)
$`{\displaystyle \frac{d\pi _{a\mu }}{d\tau _a}}+{\displaystyle \frac{1}{2}}\lambda _a^{}\pi _{a\lambda }\pi _{a\nu }g_{,\mu }^{\lambda \nu }(z_a)=e_a{\displaystyle \frac{dz_a^\nu }{d\tau _a}}\left\{A_{\nu ,\mu }(z_a)A_{\mu ,\nu }(z_a)\right\},`$ (11)
$`\pi _{a\mu }\pi _{a\nu }g^{\mu \nu }+m_a^2=0.`$ (12)
This set of three equations is equivalent to Eq.(6) when $`\lambda _a^{}=1/m_a`$ is chosen.
By performing the integration over the parameter $`\tau _a`$ and setting
$`z_a^\mu `$ $`=`$ $`(t,z_a),\pi _{a\mu }=(\pi _{a0},\pi _a),`$
$`A_\mu `$ $`=`$ $`(\phi ,A),E=^{01},`$ (13)
$`\lambda _a`$ $`=`$ $`\lambda _a^{}{\displaystyle \frac{d\tau _a}{dz_a^0}}|_{z_a^0(\tau _a)=t}.`$ (14)
the action (9) becomes
$`I_{E+P}`$ $`=`$ $`{\displaystyle }d^2x[A{\displaystyle \frac{E}{t}}{\displaystyle \frac{1}{2}}\sqrt{g}E^2+\phi \{{\displaystyle \frac{E}{x}}{\displaystyle \underset{a}{}}e_a\delta (xz_a(t))\}`$
$`+{\displaystyle \underset{a}{}}{\displaystyle \frac{dz_a}{dt}}(\pi _a+e_aA)\delta (xz_a(t))+{\displaystyle \underset{a}{}}\{\pi _{a0}{\displaystyle \frac{1}{2}}\lambda (\pi _{a\mu }\pi _{a\nu }g^{\mu \nu }+m_a^2)\}\delta (xz_a(t))].`$
Varying the Lagrange multipliers $`\phi `$ and $`\lambda _a`$ yields the constraints
$`{\displaystyle \frac{E}{x}}={\displaystyle \underset{a}{}}e_a\delta (xz_a(t)),`$ (16)
$`\pi _{a\mu }\pi _{a\nu }g^{\mu \nu }+m_a^2=0.`$ (17)
The solution to (16) is
$$E=\frac{1}{2}_x\underset{a}{}e_a|xz_a(t)|.$$
(18)
In (1+1) dimensions the electric field has no independent degrees of freedom. When the charged particles move within a finite region $`(|x|<L)`$, the electric field in an outer region $`(|x|>L)`$ is $`E=\pm \frac{1}{2}_ae_a`$. For a system of zero total charge the electric field vanishes in the outer region. As the solution to (17) we choose
$$\pi _{a0}=\frac{1}{g^{00}}\left\{g^{01}\pi _a+\sqrt{(g^{01}\pi _a)^2g^{00}(g^{11}\pi _a^2+m_a^2)}\right\}.$$
(19)
After the constraints are eliminated, the action (2) is
$`I_{E+P}`$ $`=`$ $`{\displaystyle }d^2x[{\displaystyle \underset{a}{}}\pi _a{\displaystyle \frac{dz_a}{dt}}\delta (xz_a(t))`$
$`+{\displaystyle \underset{a}{}}{\displaystyle \frac{1}{g^{00}}}\{g^{01}\pi _a+\sqrt{(g^{01}\pi _a)^2g^{00}(g^{11}\pi _a^2+m_a^2)}\}\delta (xz_a(t)){\displaystyle \frac{1}{2}}\sqrt{g}E^2(t,x)].`$
From this expression we know that $`\pi _a`$ is the conjugate momentum to $`z_a`$ and hereafter we use the notation
$$p_a\pi _a.$$
Note that the variations of the action (2) with respect to $`A`$ and $`E`$ lead to
$`{\displaystyle \frac{E}{t}}+{\displaystyle \underset{a}{}}e_a{\displaystyle \frac{dz_a}{dt}}\delta (xz_a(t))=0,`$ (21)
$`\sqrt{g}E={\displaystyle \frac{\phi }{x}}{\displaystyle \frac{A}{t}}.`$ (22)
Eq.(21) is automatically satisfied for the solution (18) to the constraint equation, in contrast to the (3+1) dimensional case where the corresponding equation is a true dynamical equation. Upon inserting the solution (18) into the action (2) all terms related to the components of $`A_\mu `$ cancel. Hence we need no longer consider $`A_\mu `$. Its solution, if desired, is straightforwardly obtained by solving (16) after fixing the gauge and obtaining the full solution of the metric.
Consider next the transformation of the gravity sector to canonical form. We decompose the scalar curvature in terms of the extrinsic curvature $`K`$ via
$$\sqrt{g}R=2_0(\sqrt{\gamma }K)+2_1[(N_1K_1N_0)/\sqrt{\gamma }],$$
(23)
where the metric is
$$ds^2=N_0^2dt^2+\gamma \left(dx+\frac{N_1}{\gamma }dt\right)^2,$$
(24)
with $`K=(2N_0\gamma )^1(2_1N_1\gamma ^1N_1_1\gamma _0\gamma )`$, so that $`\gamma =g_{11},N_0=(g^{00})^{1/2}`$ and $`N_1=g_{10}`$. We then rewrite the gravity sector in first-order form . We find that the action (1) becomes
$$I=d^2x\left\{\underset{a}{}p_a\dot{z}_a\delta (xz_a(t))+\pi \dot{\gamma }+\mathrm{\Pi }\dot{\mathrm{\Psi }}+N_0R^0+N_1R^1\right\},$$
(25)
where $`\pi `$ and $`\mathrm{\Pi }`$ are conjugate momenta to $`\gamma `$ and $`\mathrm{\Psi }`$, respectively, and
$`R^0`$ $`=`$ $`\kappa \sqrt{\gamma }\gamma \pi ^2+2\kappa \sqrt{\gamma }\pi \mathrm{\Pi }+{\displaystyle \frac{1}{4\kappa \sqrt{\gamma }}}(\mathrm{\Psi }^{})^2{\displaystyle \frac{1}{\kappa }}\left({\displaystyle \frac{\mathrm{\Psi }^{}}{\sqrt{\gamma }}}\right)^{}{\displaystyle \frac{1}{2}}\sqrt{\gamma }(E^2{\displaystyle \frac{\mathrm{\Lambda }}{\kappa }})`$
$`{\displaystyle \underset{a}{}}\sqrt{{\displaystyle \frac{p_a^2}{\gamma }}+m_a^2}\delta (xz_a(t)),`$
$`R^1`$ $`=`$ $`{\displaystyle \frac{\gamma ^{}}{\gamma }}\pi {\displaystyle \frac{1}{\gamma }}\mathrm{\Pi }\mathrm{\Psi }^{}+2\pi ^{}+{\displaystyle \underset{a}{}}{\displaystyle \frac{p_a}{\gamma }}\delta (xz_a(t)),`$ (26)
with the symbols $`(\dot{})`$ and $`(^{})`$ denoting $`_0`$ and $`_1`$, respectively.
From the action (25) we obtain the set of equations
$`\dot{\pi }`$ $`+`$ $`N_0\{{\displaystyle \frac{3\kappa }{2}}\sqrt{\gamma }\pi ^2{\displaystyle \frac{\kappa }{\sqrt{\gamma }}}\pi \mathrm{\Pi }+{\displaystyle \frac{1}{8\kappa \sqrt{\gamma }\gamma }}(\mathrm{\Psi }^{})^2+{\displaystyle \frac{1}{4\sqrt{\gamma }}}(E^2{\displaystyle \frac{\mathrm{\Lambda }}{\kappa }})`$ (27)
$`{\displaystyle \underset{a}{}}{\displaystyle \frac{p_a^2}{2\gamma ^2\sqrt{\frac{p_a^2}{\gamma }+m_a^2}}}\delta (xz_a(t))\}`$
$`+`$ $`N_1\left\{{\displaystyle \frac{1}{\gamma ^2}}\mathrm{\Pi }\mathrm{\Psi }^{}+{\displaystyle \frac{\pi ^{}}{\gamma }}+{\displaystyle \underset{a}{}}{\displaystyle \frac{p_a}{\gamma ^2}}\delta (xz_a(t))\right\}+N_0^{}{\displaystyle \frac{1}{2\kappa \sqrt{\gamma }\gamma }}\mathrm{\Psi }^{}+N_1^{}{\displaystyle \frac{\pi }{\gamma }}=0,`$
$`\dot{\gamma }N_0(2\kappa \sqrt{\gamma }\gamma \pi 2\kappa \sqrt{\gamma }\mathrm{\Pi })+N_1{\displaystyle \frac{\gamma ^{}}{\gamma }}2N_1^{}=0,`$ (28)
$`R^0=0,`$ (29)
$`R^1=0,`$ (30)
$`\dot{\mathrm{\Pi }}+_1({\displaystyle \frac{1}{\gamma }}N_1\mathrm{\Pi }+{\displaystyle \frac{1}{2\kappa \sqrt{\gamma }}}N_0\mathrm{\Psi }^{}+{\displaystyle \frac{1}{\kappa \sqrt{\gamma }}}N_0^{})=0,`$ (31)
$`\dot{\mathrm{\Psi }}+N_0(2\kappa \sqrt{\gamma }\pi )N_1({\displaystyle \frac{1}{\gamma }}\mathrm{\Psi }^{})=0,`$ (32)
$`\dot{p}_a+{\displaystyle \frac{N_0}{z_a}}\sqrt{{\displaystyle \frac{p_a^2}{\gamma }}+m_a^2}{\displaystyle \frac{N_0}{2\sqrt{\frac{p_a^2}{\gamma }+m_a^2}}}{\displaystyle \frac{p_a^2}{\gamma ^2}}{\displaystyle \frac{\gamma }{z_a}}{\displaystyle \frac{N_1}{z_a}}{\displaystyle \frac{p_a}{\gamma }}`$
$`+N_1{\displaystyle \frac{p_a}{\gamma ^2}}{\displaystyle \frac{\gamma }{z_a}}+{\displaystyle 𝑑xN_0\sqrt{\gamma }E\frac{E}{z_a}}=0,`$ (33)
$`\dot{z_a}N_0{\displaystyle \frac{\frac{p_a}{\gamma }}{\sqrt{\frac{p_a^2}{\gamma }+m_a^2}}}+{\displaystyle \frac{N_1}{\gamma }}=0.`$ (34)
In the equations (33) and (34), all metric components ($`N_0`$, $`N_1`$, $`\gamma `$) are evaluated at the point $`x=z_a`$ and
$$\frac{f}{z_a}\frac{f(x)}{x}|_{x=z_a}.$$
The quantities $`N_0`$ and $`N_1`$ are Lagrange multipliers which yield the constraint equations (29) and (30). The above set of equations can be proved to be equivalent to the set of equations (2), (3) and (6).
To proceed to the canonical reduction of the action (25) we have to eliminate the redundant variables by utilizing the constrant equations to fix the coordinate conditions. Noticing that the only linear terms in the constraint equations (29) and (30) are $`\left(\mathrm{\Psi }^{}/\sqrt{\gamma }\right)^{}`$ and $`\pi ^{}`$, respectively, and the equations may be solved for these quantities, we can transform the total generator obtained from the end point variation into an approriate form to fix the coordinate conditions. Generalizing the procedure described in our previous papers for the case of $`\mathrm{\Lambda }=e_a=0`$, we find that we can consistently choose the coordinate conditions
$$\gamma =1\text{and}\mathrm{\Pi }=0.$$
(35)
Eliminating the constraints, the action (25) reduces to
$$I=𝑑x^2\left\{\underset{a}{}p_a\dot{z}_a\delta (xz_a)\right\},$$
(36)
where the reduced Hamiltonian for the system of particles is defined by
$$H=𝑑x=\frac{1}{\kappa }\mathrm{}\mathrm{\Psi }.$$
(37)
Here $`\mathrm{\Psi }`$ is a function of $`z_a`$ and $`p_a`$ and is determined by solving the constraints which under the coordinate conditions (35) become
$$\mathrm{}\mathrm{\Psi }\frac{1}{4}(\mathrm{\Psi }^{})^2+\kappa ^2\pi ^2+\frac{1}{2}(\kappa E^2\mathrm{\Lambda })+\kappa \underset{a}{}\sqrt{p_a^2+m_a^2}\delta (xz_a)=0,$$
(38)
$$2\pi ^{}+\underset{a}{}p_a\delta (xz_a)=0.$$
(39)
The consistency of this canonical reduction is proved in an analogous way to the case of $`\mathrm{\Lambda }=e_a=0`$ : namely the canonical equations of motion derived from the reduced Hamiltonian (37) are identical with the equations (33) and (34) .
## 3 SOLUTION TO THE CONSTRAINT EQUATIONS AND THE HAMILTONIAN FOR A SYSTEM OF TWO PARTICLES
The standard approach for investigating the dynamics of particles is to get first an explicit expression of the Hamiltonian and to derive the equations of motion, from which the solution of trajectories are obtained. In this section we show how to derive the Hamiltonian from the solution to the constraint equations (38) and (39) and get the explicit Hamiltonian for a system of two charged particles. Since the electric field appears in the combination $`(E^2\mathrm{\Lambda }/\kappa )`$ in all equations we set
$$V(x)E^2C\text{and}\mathrm{\Lambda }_e\mathrm{\Lambda }\kappa C$$
(40)
with $`C\frac{1}{4}(_ae_a)^2`$. Thus $`V(x)`$ vanishes in the outer region and $`\mathrm{\Lambda }_e`$ is an effective cosmological constant, which includes the contribution from the electric field. This latter situation arises from the well-known fact that in (1+1) dimensions the electromagnetic field strength is a 2-form, and so in compact spatial regions it contributes to the stress-energy tensor in the same manner as a cosmological constant, analogous to the way in which a 4-form behaves in 3+1 dimensions . We shall later see that when $`\mathrm{\Lambda }_e`$ vanishes we get the Hamiltonian which leads, in the limit $`\kappa 0`$, to the correct special-relativistic electrodynamics in (1+1) flat space-time.
We express equations (38) and (39) as
$$\mathrm{}\mathrm{\Psi }=\frac{1}{4}\left(\mathrm{\Psi }^{}\right)^2\kappa ^2\left(\chi ^{}\right)^2\frac{1}{2}(\kappa V\mathrm{\Lambda }_e)\kappa \underset{a}{}\sqrt{p_a^2+m_a^2}\delta (xz_a),$$
(41)
$$\mathrm{}\chi =\frac{1}{2}\underset{a}{}p_a\delta (xz_a),$$
(42)
where we set $`\chi ^{}=\pi `$. Rewriting (41) as
$$(1+\frac{\mathrm{\Psi }}{4})\mathrm{}\mathrm{\Psi }=\frac{1}{8}\mathrm{}(\mathrm{\Psi }^24\kappa ^2\chi ^2)\frac{1}{2}(\kappa V\mathrm{\Lambda }_e)+\kappa ^2\chi \mathrm{}\chi \kappa \underset{a}{}\sqrt{p_a^2+m_a^2}\delta (xz_a),$$
(43)
yields, upon insertion into the Right-hand side (RHS) of (37)
$`H`$ $`=`$ $`{\displaystyle \underset{a}{}}{\displaystyle \frac{\sqrt{p_a^2+m_a^2}}{1+\frac{1}{4}\mathrm{\Psi }(z_a)}}+{\displaystyle \frac{\kappa }{2}}{\displaystyle \underset{a}{}}{\displaystyle \frac{p_a\chi (z_a)}{1+\frac{1}{4}\mathrm{\Psi }(z_a)}}+{\displaystyle \frac{1}{2}}{\displaystyle 𝑑x\frac{V(x)}{1+\frac{1}{4}\mathrm{\Psi }(x)}}`$ (44)
$`{\displaystyle \frac{1}{8\kappa }}{\displaystyle 𝑑x\frac{1}{1+\frac{1}{4}\mathrm{\Psi }(x)}\mathrm{}\left(\mathrm{\Psi }^24\kappa ^2\chi ^2+2\mathrm{\Lambda }_ex^2\right)}.`$
an expression which can also be obtained repeated iteration of the insertion of the RHS of (41) into the RHS of (37).
Defining $`\varphi `$ by
$$\mathrm{\Psi }=4\mathrm{ln}|\varphi |,$$
(45)
the constraints (41) and (42) for a two-particle system become
$`\mathrm{}\varphi {\displaystyle \frac{1}{4}}\left\{\kappa ^2\left(\chi ^{}\right)^2+{\displaystyle \frac{\kappa }{2}}V{\displaystyle \frac{1}{2}}\mathrm{\Lambda }_e\right\}\varphi `$ $`=`$ $`{\displaystyle \frac{\kappa }{4}}\{\sqrt{p_1^2+m_1^2}\varphi (z_1)\delta (xz_1)`$ (46)
$`+\sqrt{p_2^2+m_2^2}\varphi (z_2)\delta (xz_2)\},`$
$$\mathrm{}\chi =\frac{1}{2}\left\{p_1\delta (xz_1)+p_2\delta (xz_2)\right\}.$$
(47)
The general solution to (47) is
$$\chi =\frac{1}{4}\left\{p_1xz_1+p_2xz_2\right\}ϵXx+ϵC_\chi .$$
(48)
The factor $`ϵ`$ ($`ϵ^2=1`$) has been introduced in the constants $`X`$ and $`C_\chi `$ so that the T-inversion (time reversal) properties of $`\chi `$ are explicitly manifest. By definition, $`ϵ`$ changes sign under time reversal and so, therefore, does $`\chi `$.
Consider first the case $`z_2<z_1`$, for which we may divide space into three regions: $`z_1<x`$ ((+) region), $`z_2<x<z_1`$ ((0) region) and $`x<z_2`$ ((-) region). In each region, $`V`$ and $`\chi ^{}`$ are constant:
$$V=\{\begin{array}{cc}0\hfill & \text{(+) region},\hfill \\ \frac{1}{2}e_1e_2\hfill & \text{(0) region},\hfill \\ 0\hfill & \text{(-) region},\hfill \end{array}$$
(49)
$$\chi ^{}=\{\begin{array}{cc}ϵX\frac{1}{4}(p_1+p_2)\hfill & \text{(+) region},\hfill \\ ϵX+\frac{1}{4}(p_1p_2)\hfill & \text{(0) region},\hfill \\ ϵX+\frac{1}{4}(p_1+p_2)\hfill & \text{(-) region}.\hfill \end{array}$$
(50)
General solutions to the homogeneous equation $`\mathrm{}\varphi \frac{1}{4}\left\{\kappa ^2\left(\chi ^{}\right)^2+\frac{\kappa }{2}V\frac{1}{2}\mathrm{\Lambda }_e\right\}\varphi =0`$ in each region are
$$\{\begin{array}{c}\varphi _+(x)=A_+e^{\frac{1}{2}K_+x}+B_+e^{\frac{1}{2}K_+x},\hfill \\ \varphi _0(x)=A_0e^{\frac{1}{2}K_0x}+B_0e^{\frac{1}{2}K_0x},\hfill \\ \varphi _{}(x)=A_{}e^{\frac{1}{2}K_{}x}+B_{}e^{\frac{1}{2}K_{}x},\hfill \end{array}$$
(51)
where
$$\{\begin{array}{ccc}K_+\hfill & =\sqrt{\kappa ^2\left(X+\frac{ϵ}{4}(p_1+p_2)\right)^2\frac{1}{2}\mathrm{\Lambda }_e}\hfill & \text{(+) region},\hfill \\ K_0\hfill & =\sqrt{\kappa ^2\left(X\frac{ϵ}{4}(p_1p_2)\right)^2\frac{\kappa }{2}e_1e_2\frac{1}{2}\mathrm{\Lambda }_e}\hfill & \text{(0) region},\hfill \\ K_{}\hfill & =\sqrt{\kappa ^2\left(X\frac{ϵ}{4}(p_1+p_2)\right)^2\frac{1}{2}\mathrm{\Lambda }_e}\hfill & \text{(-) region}.\hfill \end{array}$$
(52)
For these solutions to be the actual solutions to Eq.(46) with delta function source terms, they must satisfy the following matching conditions at $`x=z_1,z_2`$:
$`\varphi _+(z_1)=\varphi _0(z_1)=\varphi (z_1),`$ (53a)
$`\varphi _{}(z_2)=\varphi _0(z_2)=\varphi (z_2),`$ (53b)
$`\varphi _+^{}(z_1)\varphi _0^{}(z_1)={\displaystyle \frac{\kappa }{4}}\sqrt{p_1^2+m_1^2}\varphi (z_1),`$ (53c)
$`\varphi _0^{}(z_2)\varphi _{}^{}(z_2)={\displaystyle \frac{\kappa }{4}}\sqrt{p_2^2+m_2^2}\varphi (z_2).`$ (53d)
Since the magnitudes of both $`\mathrm{\Psi }`$ and $`\chi `$ increase with increasing $`|x|`$, it is necessary to impose a boundary condition which guarantees that the surface terms which arise in transforming the action vanish and simultaneously preserves the finiteness of the Hamiltonian. From the iterative expression (44) we know that we may choose the boundary condition
$$\mathrm{\Psi }^24\kappa ^2\chi ^2+2\mathrm{\Lambda }_ex^2=C_\pm x\text{for (+) and (-) regions}$$
(54)
with $`C_\pm `$ being constants to be determined.
The above matching conditions accompanied by the boundary condition (54) determine the solution $`\varphi `$ ( and also all coefficients) completely. The process of the calculation is quite analogous to the procedure in the previous paper , and we shall omit the details here. The compact expression of the $`\varphi `$ solution is
$`\varphi _+`$ $`=`$ $`\left({\displaystyle \frac{K_1}{_1}}\right)^{\frac{1}{2}}e^{\frac{1}{4}(K_{01}z_1K_{02}z_2)+\frac{1}{2}K_+(xz_1)},`$
$`\varphi _0`$ $`=`$ $`{\displaystyle \frac{1}{4K_0}}e^{\frac{1}{4}(K_{01}z_1K_{02}z_2)}\left\{(K_1_1)^{1/2}e^{\frac{1}{2}K_0(xz_1)}+(K_2_2)^{1/2}e^{\frac{1}{2}K_0(xz_2)}\right\},`$ (55)
$`\varphi _{}`$ $`=`$ $`\left({\displaystyle \frac{K_2}{_2}}\right)^{\frac{1}{2}}e^{\frac{1}{4}(K_{01}z_1K_{02}z_2)\frac{1}{2}K_{}(xz_2)},`$
where
$`K_1`$ $``$ $`2K_0+2K_{}\kappa \sqrt{p_2^2+m_2^2},`$
$`K_2`$ $``$ $`2K_0+2K_+\kappa \sqrt{p_1^2+m_1^2},`$
$`K_{01}`$ $``$ $`K_0K_++{\displaystyle \frac{\kappa ϵ}{2}}p_1,`$ (56)
$`K_{02}`$ $``$ $`K_0K_{}{\displaystyle \frac{\kappa ϵ}{2}}p_2,`$
$`_1`$ $``$ $`\kappa \sqrt{p_1^2+m_1^2}+2K_02K_+,`$
$`_2`$ $``$ $`\kappa \sqrt{p_2^2+m_2^2}+2K_02K_{},`$
and among these quantites there exists one relation
$$K_1K_2=_1_2e^{K_0(z_1z_2)},$$
(57)
which we refer to as the determining equation of $`X`$.
The Hamiltonian (37) becomes
$`H`$ $`=`$ $`{\displaystyle \frac{1}{\kappa }}{\displaystyle 𝑑x\mathrm{}\mathrm{\Psi }}={\displaystyle \frac{4}{\kappa }}\left[{\displaystyle \frac{\varphi ^{}}{\varphi }}\right]_{\mathrm{}}^{\mathrm{}}`$ (58)
$`=`$ $`{\displaystyle \frac{2(K_++K_{})}{\kappa }}.`$
Once the solution $`X`$ to (57) is obtained, the Hamiltonian is explicitly determined. Consequently (57) is the determining equation of the Hamiltonian.
Repeating the analysis for $`z_1<z_2`$ yields a similar solution with $`p_ip_i`$ and so the full solution is obtained by replacing $`p_i`$ and $`z_1z_2`$ by $`\stackrel{~}{p}_i=p_i\text{sgn}(z_1z_2)`$ and $`|z_1z_2|`$, respectively. The determining equation (57) of the Hamiltonian becomes
$$K_1K_2=_1_2e^{K_0|z_1z_2|}.$$
(59)
or more explicitly
$`\left(4K_0^2+[\kappa \sqrt{p_1^2+m_1^2}2K_+][\kappa \sqrt{p_2^2+m_2^2}2K_{}]\right)\text{tanh}\left({\displaystyle \frac{1}{2}}K_0|z_1z_2|\right)`$
$`=2K_0\left([\kappa \sqrt{p_1^2+m_1^2}2K_+]+[\kappa \sqrt{p_2^2+m_2^2}2K_{}]\right)`$ (60)
where the momentum $`p_i`$ is replaced by $`\stackrel{~}{p}_i`$.
For the expression (58) to have a definite meaning as the Hamiltonian, $`K_\pm `$ must both be real, with positive sum. This imposes a restriction on $`X`$ corresponding to a value of the cosmological constant $`\mathrm{\Lambda }_e`$. However $`K_0`$ is not necessarily real. If $`e_1e_2`$ takes a large positive value (strong electromagnetic repulsion), $`K_0`$ may be imaginary. In this case we need to reconsider the above analysis, because in the (0) region the soluton to the $`\varphi `$ equation (46) becomes
$$\varphi _0(x)=A_s\text{sin}\frac{1}{2}\stackrel{~}{K}_0x+A_c\text{cos}\frac{1}{2}\stackrel{~}{K}_0x,$$
(61)
where
$`\stackrel{~}{K}_0`$ $`=`$ $`iK_0`$ (62)
$`=`$ $`\sqrt{{\displaystyle \frac{\kappa }{2}}e_1e_2+{\displaystyle \frac{1}{2}}\mathrm{\Lambda }_e\kappa ^2\left(X{\displaystyle \frac{ϵ}{4}}(\stackrel{~}{p}_1\stackrel{~}{p}_2)\right)^2}.`$
Under the same matching conditions (53a-53d) and the boundary condition (54) we get, instead of (3), a new determining equation of the Hamiltonian
$`\left(4\stackrel{~}{K}_0^2[\kappa \sqrt{p_1^2+m_1^2}2K_+][\kappa \sqrt{p_2^2+m_2^2}2K_{}]\right)\text{tan}\left({\displaystyle \frac{1}{2}}\stackrel{~}{K}_0|z_1z_2|\right)`$
$`=2\stackrel{~}{K}_0\left([\kappa \sqrt{p_1^2+m_1^2}2K_+]+[\kappa \sqrt{p_2^2+m_2^2}2K_{}]\right).`$ (63)
Actually, this is just the equation derived from (3) by formally replacing $`K_0`$ with $`i\stackrel{~}{K}_0`$. The solution of $`\varphi `$ for imaginary $`K_0`$ is also identical with that obtained from (3) by the same replacement. We can therefore use equation (59) for all values of $`K_0`$; it is a transcendental equation which determines $`H`$ in terms of the momenta and positions of the particles. We have previously shown that in the case of zero cosmological constant and no charges the solution for $`H`$ is expressed in terms of the Lambert $`W`$ function. More generally, with $`\mathrm{\Lambda }_e`$ and $`e_a`$ all nonzero, a solution to (59) for $`H`$ cannot be explicitly expressed in terms of known fuctions. However it can be obtained in successive approximations in the parameters $`c^1`$, $`\kappa `$, etc. The examples will be shown later in the sections VI and VIII.
Though in the general case the Hamiltonian cannot be expressed explicitly in terms of known functions, the canonical equations of motion can be exactly derived from the determining equation (59) by differentiating it with respect to the variables $`z_i`$ and $`p_i`$ . For the variables $`p_1`$ and $`z_1`$ we have
$`\dot{p}_1`$ $`=`$ $`{\displaystyle \frac{H}{z_1}}={\displaystyle \frac{2}{\kappa }}\left({\displaystyle \frac{K_+}{z_1}}+{\displaystyle \frac{K_{}}{z_1}}\right)=2\left({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}}\right){\displaystyle \frac{X}{z_1}}`$ (64)
$`=`$ $`{\displaystyle \frac{2}{\kappa }}\left({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}}\right){\displaystyle \frac{K_0K_1K_2}{J}},`$
$`\dot{z}_1`$ $`=`$ $`{\displaystyle \frac{H}{p_1}}={\displaystyle \frac{2}{\kappa }}\left({\displaystyle \frac{K_+}{p_1}}+{\displaystyle \frac{K_{}}{p_1}}\right)`$ (65)
$`=`$ $`{\displaystyle \frac{ϵ}{2}}\left({\displaystyle \frac{Y_+}{K_+}}{\displaystyle \frac{Y_{}}{K_{}}}\right)+2\left({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}}\right){\displaystyle \frac{X}{p_1}}`$
$`=`$ $`ϵ{\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{8}{J}}\left({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}}\right){\displaystyle \frac{K_0K_1}{_1}}\left\{{\displaystyle \frac{p_1}{\sqrt{p_1^2+m_1^2}}}ϵ{\displaystyle \frac{Y_+}{K_+}}\right\},`$
where
$`Y_+`$ $``$ $`\kappa \left[X+{\displaystyle \frac{ϵ}{4}}(p_1+p_2)\right],`$
$`Y_0`$ $``$ $`\kappa \left[X{\displaystyle \frac{ϵ}{4}}(p_1p_2)\right],`$ (66)
$`Y_{}`$ $``$ $`\kappa \left[X{\displaystyle \frac{ϵ}{4}}(p_1+p_2)\right],`$
and
$`J`$ $`=`$ $`2\left\{\left({\displaystyle \frac{Y_0}{K_0}}+{\displaystyle \frac{Y_+}{K_+}}\right)K_1+\left({\displaystyle \frac{Y_0}{K_0}}+{\displaystyle \frac{Y_{}}{K_{}}}\right)K_2\right\}`$ (67)
$`2\left\{\left({\displaystyle \frac{Y_0}{K_0}}{\displaystyle \frac{Y_+}{K_+}}\right){\displaystyle \frac{1}{_1}}+\left({\displaystyle \frac{Y_0}{K_0}}{\displaystyle \frac{Y_{}}{K_{}}}\right){\displaystyle \frac{1}{_2}}\right\}K_1K_2{\displaystyle \frac{Y_0}{K_0}}K_1K_2(z_1z_2).`$
Similarly for particle 2 the equations are
$`\dot{p}_2`$ $`=`$ $`{\displaystyle \frac{2}{\kappa }}\left({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}}\right){\displaystyle \frac{K_0K_1K_2}{J}},`$ (68)
$`\dot{z}_2`$ $`=`$ $`ϵ{\displaystyle \frac{Y_{}}{K_{}}}+{\displaystyle \frac{8}{J}}\left({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}}\right){\displaystyle \frac{K_0K_2}{_2}}\left\{{\displaystyle \frac{p_2}{\sqrt{p_2^2+m_2^2}}}+ϵ{\displaystyle \frac{Y_{}}{K_{}}}\right\}.`$ (69)
These canonical equations guarantee the conservation of the Hamiltonian and the total momentum $`p_1+p_2`$.
On the other hand the equations of motion (33) and (34) derived from the action (25) become under the coordinate conditions (35)
$`\dot{p}_a`$ $`=`$ $`{\displaystyle \frac{N_0}{z_a}}\sqrt{p_a^2+m_a^2}+{\displaystyle \frac{N_1}{z_a}}p_a+{\displaystyle \frac{1}{2}}{\displaystyle \underset{b}{}}e_ae_bN_0{\displaystyle \frac{}{z_a}}|z_az_b|,`$ (70)
$`\dot{z_a}`$ $`=`$ $`N_0{\displaystyle \frac{p_a}{\sqrt{p_a^2+m_a^2}}}N_1.`$ (71)
It is straightforward to verify that insertion of the solutions of the metric components given in Appendix A into (70) and (71) reproduces the canonical equations of motion (64), (65), (68) and (69) where the partial derivatives at $`z1,z2`$ are defined by
$$\frac{N_{0,1}}{z_i}\frac{1}{2}\left\{\frac{N_{0,1}}{x}|_{x=z_i+0}+\frac{N_{0,1}}{x}|_{x=z_i0}\right\}.$$
(72)
Thus consistency between the geodesic equations and the canonical equations of motion is explicitly assured, while formal proof of consistency in the case of $`\mathrm{\Lambda }_e=0`$ and $`e_a=0`$ can be easily generalized .
The components of the metric are determined from the equations (27), (28), (31) and (32) under the coordinate conditions (35). The derivation and the explicit solutions of the metric are given in Appendix A. With these solutions we can trace how the structure of space-time changes due to the motion of the two bodies.
## 4 EXACT SOLUTIONS OF THE TRAJECTORIES FOR EQUAL MASSES AND ARBITRARY CHARGES
In this section we consider a system of two particles with equal mass. Since the total momentum is conserved we can always choose the center of inertia (C.I.) system $`p_1=p_2=p`$. Corresponding to the sign of $`(\sqrt{H^2+8\mathrm{\Lambda }_e/\kappa ^2}2ϵ\stackrel{~}{p})^28e_1e_2/\kappa 8\mathrm{\Lambda }_e/\kappa ^2`$ the determining equations (3) and (3) become
$$(𝒥_\mathrm{\Lambda }^{\mathrm{\hspace{0.33em}2}}+B^2)\text{tanh}\left(\frac{\kappa }{8}𝒥_\mathrm{\Lambda }|r|\right)=2𝒥_\mathrm{\Lambda }B,$$
(73)
and
$$(\stackrel{~}{𝒥}_\mathrm{\Lambda }^{\mathrm{\hspace{0.33em}2}}B^2)\text{tan}\left(\frac{\kappa }{8}\stackrel{~}{𝒥}_\mathrm{\Lambda }|r|\right)=2\stackrel{~}{𝒥}_\mathrm{\Lambda }B,$$
(74)
respectively, where
$`𝒥_\mathrm{\Lambda }=\sqrt{\left(\sqrt{H^2+{\displaystyle \frac{8\mathrm{\Lambda }_e}{\kappa ^2}}}2ϵ\stackrel{~}{p}\right)^2{\displaystyle \frac{8e_1e_2}{\kappa }}{\displaystyle \frac{8\mathrm{\Lambda }_e}{\kappa ^2}}},`$
$`\stackrel{~}{𝒥}_\mathrm{\Lambda }=\sqrt{{\displaystyle \frac{8e_1e_2}{\kappa }}+{\displaystyle \frac{8\mathrm{\Lambda }_e}{\kappa ^2}}\left(\sqrt{H^2+{\displaystyle \frac{8\mathrm{\Lambda }_e}{\kappa ^2}}}2ϵ\stackrel{~}{p}\right)^2},`$
$`B=H2\sqrt{p^2+m^2}.`$ (75)
The equation (73) is further divided into two types:
$$\text{tanh}\left(\frac{\kappa }{16}𝒥_\mathrm{\Lambda }|r|\right)=\frac{B}{𝒥_\mathrm{\Lambda }},\text{(tanh-type A)}$$
(76)
or
$$\text{tanh}\left(\frac{\kappa }{16}𝒥_\mathrm{\Lambda }|r|\right)=\frac{𝒥_\mathrm{\Lambda }}{B},\text{(tanh-type B)}.$$
(77)
In the case of $`\mathrm{\Lambda }_e=0`$ and $`e_a=0`$ the tanh-type B equation is excluded, because $`𝒥_\mathrm{\Lambda }/B`$ exceeds unity. When a cosmological constant and/or charge are introduced, this equation may have a solution in some restricted range of the parameters.
Eq.(74) is also divided into
$$\text{tan}\left(\frac{\kappa }{16}\stackrel{~}{𝒥}_\mathrm{\Lambda }|r|\right)=\frac{B}{\stackrel{~}{𝒥}_\mathrm{\Lambda }},\text{(tan-type A)}$$
(78)
or
$$\text{tan}\left(\frac{\kappa }{16}\stackrel{~}{𝒥}_\mathrm{\Lambda }|r|\right)=\frac{\stackrel{~}{𝒥}_\mathrm{\Lambda }}{B},\text{(tan-type B)}.$$
(79)
For all four types of the determining equations the canonical equations of motion are identical:
$`\dot{p}`$ $`=`$ $`{\displaystyle \frac{\kappa 𝒥_\mathrm{\Lambda }^2(𝒥_\mathrm{\Lambda }^2B^2)}{16C}}\text{sgn(r)},`$ (80)
$`\dot{r}`$ $`=`$ $`2ϵ\sqrt{1+{\displaystyle \frac{8\mathrm{\Lambda }_e}{\kappa ^2H^2}}}\left(1{\displaystyle \frac{𝒥_\mathrm{\Lambda }^2}{C}}\right)\text{sgn(r)}+{\displaystyle \frac{2𝒥_\mathrm{\Lambda }^2}{C}}{\displaystyle \frac{p}{\sqrt{p^2+m^2}}}.`$ (81)
where
$$C=\frac{1}{\sqrt{1+\frac{8\mathrm{\Lambda }_e}{\kappa ^2H^2}}}\left\{\sqrt{1+\frac{8\mathrm{\Lambda }_e}{\kappa ^2H^2}}𝒥_\mathrm{\Lambda }^2\left(\sqrt{H^2+\frac{8\mathrm{\Lambda }_e}{\kappa ^2}}2ϵ\stackrel{~}{p}\right)\left(B+\frac{\kappa }{16}(𝒥_\mathrm{\Lambda }^2B^2)r\right)\right\}.$$
(82)
For given values of $`\mathrm{\Lambda }_e`$ and $`e_a`$, the equations (73) or (74) describe the surface in $`(r,p,H)`$ space of all allowed phase-space trajectories. Since $`H`$ is a constant of motion, we can draw a phase space trajectory in $`(r,p)`$ space by setting $`H=H_0`$ in (73) or (74). This same trajectory can also be obtained directly from the solution $`r(t),p(t)`$ to the canonical equations (80) and (81) by eliminating the time variable $`t`$. Numerical solutions to (80) and (81) confirm this, but in $`r(t)`$ and $`p(t)`$ superficial singularities appear due to the zero points of $`C`$.
It is therefore preferable to describe the particles’ trajectories in terms of some invariant parameter. The common proper time $`\tau _a`$ of each particle is the best candidate as seen in the starting action (1). From the metric components given in the Appendix A and the canonical equations (71), the proper time is
$`d\tau _a^2`$ $`=`$ $`dt^2\left\{N_0(z_a)^2(N_1(z_a)+\dot{z}_a)^2\right\},`$ (83)
$`=`$ $`dt^2N_0(z_a)^2{\displaystyle \frac{m_a^2}{p_a^2+m_a^2}}(a=1,2).`$
For the equal mass case it is identical for particles 1 and 2
$$d\tau =d\tau _1=d\tau _2=\frac{m}{\sqrt{p^2+m^2}}\frac{𝒥_\mathrm{\Lambda }^2}{C}dt,$$
(84)
from which the canonical equations (80) and (81) may be expressed in the form
$`{\displaystyle \frac{dp}{d\tau }}={\displaystyle \frac{\kappa \sqrt{p^2+m^2}(𝒥_\mathrm{\Lambda }^2B^2)}{16m}}\text{sgn(r)},`$ (85)
$`{\displaystyle \frac{dr}{d\tau }}={\displaystyle \frac{2ϵ}{m}}\left\{{\displaystyle \frac{\sqrt{p^2+m^2}C}{𝒥_\mathrm{\Lambda }^2}}(\sqrt{p^2+m^2}ϵ\stackrel{~}{p})\right\}\text{sgn(r)}.`$ (86)
Remarkably these equations can be solved exactly. First we solve Eq.(85) for $`p(\tau )`$ and then obtain $`r(\tau )`$, either by directly solving (86) after substituting the solution for $`p`$ or by solving (76)-(79) for $`r`$. For the $`r>0`$ region Eq.(85) leads to
$`{\displaystyle _{p_0}^p}{\displaystyle \frac{dp}{\left\{\sqrt{p^2+m^2}ϵ\sqrt{1+\frac{8\mathrm{\Lambda }_e}{\kappa ^2H^2}}p\frac{1}{H}\left(m^2+\frac{2e_1e_2}{\kappa }\right)\right\}\sqrt{p^2+m^2}}}`$ $`=`$ $`{\displaystyle \frac{\kappa H}{4m}}{\displaystyle _{\tau _0}^\tau }𝑑\tau `$ (87)
$`=`$ $`{\displaystyle \frac{\kappa H}{4m}}(\tau \tau _0).`$
This expression implies the condition
$$1+\frac{8\mathrm{\Lambda }_e}{\kappa ^2H^2}0$$
(88)
which is satisfied for all $`\mathrm{\Lambda }_e>0`$. For negative $`\mathrm{\Lambda }_e`$ the motion is allowed provided $`H`$ satisfies
$$H\sqrt{\frac{8\mathrm{\Lambda }_e}{\kappa ^2}}.$$
(89)
We perfrom the integration of the Left-hand side (LHS) of (87) in three cases, separately, depending on the value of $`\mathrm{\Lambda }_e`$. The solution $`p(\tau )`$ is
$$p(\tau )=\frac{ϵm}{2}\left(f(\tau )\frac{1}{f(\tau )}\right)$$
(90)
with
$$f(\tau )=\{\begin{array}{cc}\frac{\frac{H}{m}\left(1+\sqrt{\gamma _H}\right)\left\{1\eta e^{\frac{ϵ\kappa m}{4}\sqrt{\gamma _m}(\tau \tau _0)}\right\}}{\gamma _e+\sqrt{\gamma _m}+\left(\sqrt{\gamma _m}\gamma _e\right)\eta e^{\frac{ϵ\kappa m}{4}\sqrt{\gamma _m}(\tau \tau _0)}}\hfill & \gamma _m>0,\hfill \\ & \\ \frac{1+\sqrt{\gamma _H}}{\frac{m}{H}\gamma _e+\frac{\sigma }{m\sigma \frac{ϵ\kappa H}{8}(\tau \tau _0)}}\hfill & \gamma _m=0,\hfill \\ & \\ \frac{\frac{H}{m}(1+\sqrt{\gamma _H})}{\gamma _e+\sqrt{\gamma _m}\frac{\sigma +\frac{m^2}{H}\sqrt{\gamma _m}\mathrm{tan}\left[\frac{ϵ\kappa m}{8}\sqrt{\gamma _m}(\tau \tau _0)\right]}{\frac{m^2}{H}\sqrt{\gamma _m}\sigma \mathrm{tan}\left[\frac{ϵ\kappa m}{8}\sqrt{\gamma _m}(\tau \tau _0)\right]}}\hfill & \gamma _m<0,\hfill \end{array}$$
(91)
where
$$\begin{array}{cc}\gamma _H=1+\frac{8\mathrm{\Lambda }_e}{\kappa ^2H^2},\hfill & \gamma _e=1+\frac{2e_1e_2}{\kappa m^2},\hfill \\ \gamma _m=\gamma _e^2+\frac{8\mathrm{\Lambda }_e}{\kappa ^2m^2},\hfill & \sigma =(1+\sqrt{\gamma _H})(\sqrt{p_0^2+m^2}ϵp_0)\frac{m^2}{H}\gamma _e,\hfill \\ \eta =\frac{\sigma \frac{m^2}{H}\sqrt{\gamma _m}}{\sigma +\frac{m^2}{H}\sqrt{\gamma _m}},\hfill & \end{array}$$
(92)
with $`p_0`$ being the initial momentum at $`\tau =\tau _0`$.
Similarly the solution in $`r<0`$ region is
$$p(\tau )=\frac{ϵm}{2}\left(\overline{f}(\tau )\frac{1}{\overline{f}(\tau )}\right)$$
(93)
with
$$\overline{f}(\tau )=\{\begin{array}{cc}\frac{\frac{H}{m}\left(1+\sqrt{\gamma _H}\right)\left\{1\overline{\eta }e^{\frac{ϵ\kappa m}{4}\sqrt{\gamma _m}(\tau \tau _0)}\right\}}{\gamma _e+\sqrt{\gamma _m}+\left(\sqrt{\gamma _m}\gamma _e\right)\overline{\eta }e^{\frac{ϵ\kappa m}{4}\sqrt{\gamma _m}(\tau \tau _0)}}\hfill & \gamma _m>0,\hfill \\ & \\ \frac{1+\sqrt{\gamma _H}}{\frac{m}{H}\gamma _e+\frac{\overline{\sigma }}{m\overline{\sigma }\frac{ϵ\kappa H}{8}(\tau \tau _0)}}\hfill & \gamma _m=0,\hfill \\ & \\ \frac{\frac{H}{m}(1+\sqrt{\gamma _H})}{\gamma _e+\sqrt{\gamma _m}\frac{\overline{\sigma }+\frac{m^2}{H}\sqrt{\gamma _m}\mathrm{tan}\left[\frac{ϵ\kappa m}{8}\sqrt{\gamma _m}(\tau \tau _0)\right]}{\frac{m^2}{H}\sqrt{\gamma _m}\overline{\sigma }\mathrm{tan}\left[\frac{ϵ\kappa m}{8}\sqrt{\gamma _m}(\tau \tau _0)\right]}}\hfill & \gamma _m<0,\hfill \end{array}$$
(94)
where
$$\overline{\sigma }=(1+\sqrt{\gamma _H})(\sqrt{p_0^2+m^2}+ϵp_0)\frac{m^2}{H}\gamma _e,\overline{\eta }=\frac{\overline{\sigma }\frac{m^2}{H}\sqrt{\gamma _m}}{\overline{\sigma }+\frac{m^2}{H}\sqrt{\gamma _m}}.$$
(95)
Corresponding to each type of the determining equations (76)- (79) the solution for $`r(\tau )`$ is obtained as follows,
tanh-type A:
$$r(\tau )=\{\begin{array}{cc}\frac{16\text{tanh}^1\left[\frac{\kappa \left(Hm\left|f(\tau )+\frac{1}{f(\tau )}\right|\right)}{\sqrt{\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (f(\tau )\frac{1}{f(\tau )})\right)^28\kappa e_1e_28\mathrm{\Lambda }_e}}\right]}{\sqrt{\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (f(\tau )\frac{1}{f(\tau )})\right)^28\kappa e_1e_28\mathrm{\Lambda }_e}}\hfill & r>0,\hfill \\ & \\ \frac{16\text{tanh}^1\left[\frac{\kappa \left(Hm\left|\overline{f}(\tau )+\frac{1}{\overline{f}(\tau )}\right|\right)}{\sqrt{\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (\overline{f}(\tau )\frac{1}{\overline{f}(\tau )})\right)^28\kappa e_1e_28\mathrm{\Lambda }_e}}\right]}{\sqrt{\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (\overline{f}(\tau )\frac{1}{\overline{f}(\tau )})\right)^28\kappa e_1e_28\mathrm{\Lambda }_e}}\hfill & r<0,\hfill \end{array}$$
(96)
tanh-type B:
$$r(\tau )=\{\begin{array}{cc}\frac{16\text{tanh}^1\left[\frac{\sqrt{\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (f(\tau )\frac{1}{f(\tau )})\right)^28\kappa e_1e_28\mathrm{\Lambda }_e}}{\kappa \left(Hm\left|f(\tau )+\frac{1}{f(\tau )}\right|\right)}\right]}{\sqrt{\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (f(\tau )\frac{1}{f(\tau )})\right)^28\kappa e_1e_28\mathrm{\Lambda }_e}}\hfill & r>0,\hfill \\ & \\ \frac{16\text{tanh}^1\left[\frac{\sqrt{\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (\overline{f}(\tau )\frac{1}{\overline{f}(\tau )})\right)^28\kappa e_1e_28\mathrm{\Lambda }_e}}{\kappa \left(Hm\left|\overline{f}(\tau )+\frac{1}{\overline{f}(\tau )}\right|\right)}\right]}{\sqrt{\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (\overline{f}(\tau )\frac{1}{\overline{f}(\tau )})\right)^28\kappa e_1e_28\mathrm{\Lambda }_e}}\hfill & r<0,\hfill \end{array}$$
(97)
tan-type A:
$$r(\tau )=\{\begin{array}{cc}\frac{16\left(\text{tan}^1\left[\frac{\kappa \left(m\left|f(\tau )+\frac{1}{f(\tau )}\right|H\right)}{\sqrt{8\mathrm{\Lambda }_e+8\kappa e_1e_2\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (f(\tau )\frac{1}{f(\tau )})\right)^2}}\right]+n\pi \right)}{\sqrt{8\mathrm{\Lambda }_e+8\kappa e_1e_2\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (f(\tau )\frac{1}{f(\tau )})\right)^2}}\hfill & r>0,\hfill \\ & \\ \frac{16\left(\text{tan}^1\left[\frac{\kappa \left(m\left|\overline{f}(\tau )+\frac{1}{\overline{f}(\tau )}\right|H\right)}{\sqrt{8\mathrm{\Lambda }_e+8\kappa e_1e_2\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }}m\kappa (\overline{f}(\tau )\frac{1}{\overline{f}(\tau )})\right)^2}}\right]+n\pi \right)}{\sqrt{8\mathrm{\Lambda }_e+8\kappa e_1e_2\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (\overline{f}(\tau )\frac{1}{\overline{f}(\tau )})\right)^2}}\hfill & r<0,\hfill \end{array}$$
(98)
tan-type B:
$$r(\tau )=\{\begin{array}{cc}\frac{16\left(\text{tan}^1\left[\frac{\sqrt{8\mathrm{\Lambda }_e+8\kappa e_1e_2\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (f(\tau )\frac{1}{f(\tau )})\right)^2}}{\kappa \left(Hm\left|f(\tau )+\frac{1}{f(\tau )}\right|\right)}\right]+n\pi \right)}{\sqrt{8\mathrm{\Lambda }_e+8\kappa e_1e_2\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (f(\tau )\frac{1}{f(\tau )})\right)^2}}\hfill & r>0,\hfill \\ & \\ \frac{16\left(\text{tan}^1\left[\frac{\sqrt{8\mathrm{\Lambda }_e+8\kappa e_1e_2\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (\overline{f}(\tau )\frac{1}{\overline{f}(\tau )})\right)^2}}{\kappa \left(Hm\left|\overline{f}(\tau )+\frac{1}{\overline{f}(\tau )}\right|\right)}\right]+n\pi \right)}{\sqrt{8\mathrm{\Lambda }_e+8\kappa e_1e_2\left(\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}m\kappa (\overline{f}(\tau )\frac{1}{\overline{f}(\tau )})\right)^2}}\hfill & r<0.\hfill \end{array}$$
(99)
## 5 ANALYSIS OF ELECTRODYNAMIC MOTION WITH $`\mathrm{\Lambda }_e=0`$
Based on the exact solutions in the previous section we can analyze the dynamics of two-body problem. In this section we investigate electrodynamics in a space-time with $`\mathrm{\Lambda }_e=0`$. The solution $`p(\tau )`$ in this case is given for $`r>0`$ by
$$p(\tau )=\frac{ϵm}{2}\left(f_e(\tau )\frac{1}{f_e(\tau )}\right)$$
(100)
with
$`f_e(\tau )`$ $`=`$ $`{\displaystyle \frac{H}{m\gamma _e}}\left\{1\eta _ee^{\frac{ϵ\kappa m}{4}\gamma _e(\tau \tau _0)}\right\},`$
$`\eta _e`$ $`=`$ $`{\displaystyle \frac{\sqrt{p_0^2+m^2}ϵp_0\frac{m^2}{H}\gamma _e}{\sqrt{p_0^2+m^2}ϵp_0}},`$
and for $`r<0`$ by
$$p(\tau )=\frac{ϵm}{2}\left(\overline{f}_e(\tau )\frac{1}{\overline{f}_e(\tau )}\right)$$
(102)
with
$`\overline{f}_e(\tau )`$ $`=`$ $`{\displaystyle \frac{H}{m\gamma _e}}\left\{1\overline{\eta }_ee^{\frac{ϵ\kappa m}{4}\gamma _e(\tau \tau _0)}\right\},`$
$`\overline{\eta }_e`$ $`=`$ $`{\displaystyle \frac{\sqrt{p_0^2+m^2}+ϵp_0\frac{m^2}{H}\gamma _e}{\sqrt{p_0^2+m^2}+ϵp_0}}.`$
The relative distance $`r(\tau )`$ is obtained from (76) - (79) by simply replacing $`f(\tau )`$ and $`\overline{f}(\tau )`$ with $`f_e(\tau )`$ and $`\overline{f}_e(\tau )`$, respectively.
Relative motion of two charged particles is classified by the signs and the magnitudes of the charges. Since the charges always appear as the product $`e_1e_2`$ in all solutions, it is sufficient to analyze the attractive case by setting $`e_1=e_2=q`$ and the repulsive case by setting $`e_1=e_2=q`$ so that the charges have equal magnitude. Throughout this paper, in the numerical analysis, we set $`ϵ=1,\kappa =1`$ and rescale everything in units of $`m`$ (effectively setting $`m=1`$) for simplicity, except when otherwise stated. It should be remarked that for every phase space trajectory there exists a time-reversed trajectory with $`ϵ=1`$, which is obtained by reversing the former with respect to the $`r`$-axis.
### 5.1 The attractive case: $`e_1=e_2=q`$
When the electric force between charges is attractive, the value of $`\kappa ^2\left(Hmf(\tau )+\frac{m}{f(\tau )}\right)^28\kappa e_1e_2`$ is always positive and two particles follow a bounded periodic motion described by tanh-type solution (76). The period is determined from the initial value of $`p_0=\sqrt{(H/2)^2m^2}`$ at $`r=0`$:
$$T=\frac{16}{\kappa m\gamma _e}\text{tanh}^1\left(\frac{\gamma _e\sqrt{H^24m^2}}{(2\gamma _e)H}\right).$$
(104)
Although the above expression diverges when $`\gamma _e=\frac{2H}{2p_0+H}`$, this situation is never realized in the attractive case, since $`\gamma _e<1`$ whereas $`\frac{2H}{2p_0+H}>1`$.
In Figs.1 and 2 we plot $`r(\tau )`$ and the phase space trajectories , respectively, for two particles initially at $`r=0`$ for fixed $`|q|=1`$ and four different values of $`H`$. We see that as $`H`$ increases, the phase space trajectory becomes more $`S`$-shaped, $`r(\tau )`$ becomes steep and the amplitude increases, but the period approaches a constant value $`(16/\kappa m\gamma _e)\text{tanh}^1[\gamma _e/(2\gamma _e)]`$. This $`S`$-shaped deformation of the trajectory at higher energy is caused by the $`p`$-dependence of the gravitational potential and is common to the $`q=0`$ case described with the $`W`$ function .
Figs.3 and 4 show similar plots for fixed $`H=3`$ and four different values of $`|q|`$. For large $`|q|`$ ($`\gamma _e<0`$) due to the attractive force between charges the phase space trajectory contracts toward the origin, and the period and the amplitude of $`r(\tau )`$ become small. As $`|q|`$ becomes small, $`\gamma _e`$ passes zero at $`|q|=1/\sqrt{2}`$ and approaches 1 at $`|q|=0`$. As we can see from the figures, no peculiar changes in either $`r(\tau )`$ or the phase space trajectories occur except for a growing amplitude and period, a natural tendency due to the weakening attractive force.
The distinct characteristics of these relativistic motions are more easily seen by comparing the exact trajectory with those of the motions in three approximations :
(1) the non-relativistic motion described by the Hamiltonian
$$H=2m+\frac{p^2}{m}+\frac{q^2}{2}|r|+\frac{\kappa m^2}{4}|r|,$$
(105)
(2) the linear approximation in $`\kappa `$, its Hamiltonian being
$`H`$ $`=`$ $`2\sqrt{p^2+m^2}+{\displaystyle \frac{1}{2}}q^2|r|+{\displaystyle \frac{\kappa }{4}}\{(2p^2+m^22ϵ\stackrel{~}{p}\sqrt{p^2+m^2})|r|`$ (106)
$`+{\displaystyle \frac{1}{2}}q^2(\sqrt{p^2+m^2}ϵ\stackrel{~}{p})r^2+{\displaystyle \frac{1}{24}}q^4|r|^3\},`$
(3) the $`\kappa 0`$ limit theory, namely, special-relativistic electrodynamics in (1+1) flat space-time described by the Hamiltonian
$$H=2\sqrt{p^2+m^2}+\frac{q^2}{2}|r|.$$
(107)
In Fig.5 the trajectories of the exact and the three approximate solutions under identical total energy $`H_0=3`$ are drawn for $`q=0.5,1,5`$ and $`10`$. For small $`q=0.5`$ both the exact solution (solid curve) and the linear approximation (dashed curve) follow the $`S`$-shaped trajectories, while the trajectories of the non-relativistic solution (dotted curve) and flat electrodynamics (dot-dashed curve) have symmetrical oval forms. As $`|q|`$ increases all trajectories tend to coincide with the trajectory of flat electrodynamics (since the effect of gravity becomes relatively weak) though the initial value of $`p_0=\sqrt{m(H2m)}=1`$ in the non-relativistic case is slightly different from the value of others ($`p_0=\sqrt{(H/2)^2m^2}=\sqrt{5}/2`$). If the motion in the non- relativistic case starts from the same initial value $`p_0=\sqrt{5}/2`$ the oval trajectory for large $`|q|`$ region as shown in Fig.6 becomes larger than others, reflecting the difference between the relativistic and non- relativistic effects.
### 5.2 The repulsive case: the small $`|q|`$ regime ($`e_1=e_2=q,|q|q_c`$)
For the case where the charges are of the same sign, the electric force is repulsive and competes with the attractive gravitational force. Depending on the strengths of the two forces the solutions become either tanh-type and/or tan-type. The transition from tanh-type to tan-type is given by the zeroes of $`𝒥=\sqrt{(H2ϵ\stackrel{~}{p})^28q^2/\kappa }`$, which lead to a critical value of the charge
$$q_c=\sqrt{\frac{\kappa }{8}}\left(H\sqrt{H^24m^2}\right),$$
(108)
which separates two qualitatively different kinds of motion. In the regime of $`|q|<q_c`$, $`(𝒥)^2`$ takes both positive and negative values and $`r(\tau )`$ obeys both tanh-type and tan-type solutions, representing bounded and unbounded motions respectively. Alternatively, (108) gives the critical value of $`H`$ for fixed $`\kappa `$ and $`q`$, or the critical value of $`\kappa `$ for fixed $`H`$ and $`q`$, both corresponding to the transition from bounded to unbounded motion.
In Fig.7 we show $`r(\tau )`$ plots for $`H_0=3`$ and five different values of $`q`$. As $`|q|`$ increases the period becomes large. When $`|q|`$ exceeds the critical value $`q_c=0.2700907567`$, the motion becomes unbounded and the separation of the two particles diverges at finite $`\tau `$. A similar transition is found in Fig.8, where $`r(\tau )`$ plots are depicted for $`|q|=0.1`$ and five different values of $`H_0`$ and the transition occurs at $`H_0=7.21249`$.
Before proceeding to the analysis of the phase space trajectories, it is instructive to compare the general structure of the determining equation for the repulsive case with that in no-charge case. In the $`q=0`$ case the determining equation (59) of the Hamiltonian for equal masses becomes
$$Y^2e^{2Y}=Z^2e^{2Z},$$
(109)
where $`Y`$ and $`Z`$ are defined by
$`H`$ $`=`$ $`\sqrt{p^2+m^2}+ϵ\stackrel{~}{p}{\displaystyle \frac{8}{\kappa |r|}}Y(p,r),`$ (110)
$`Z`$ $`=`$ $`{\displaystyle \frac{\kappa |r|}{8}}(\sqrt{p^2+m^2}ϵ\stackrel{~}{p}).`$ (111)
Equation (109) has three formal solutions shown in Fig.9:
sol 1: line F-O ; trivial solution $`Y=Z`$,
sol 2: curve A-B-O-C-D; $`Y=W(Ze^Z),Z<W^1(e^1)=0.278`$,
sol 3: curve E-F-G; $`Y=W(Ze^Z),Z<0`$.
Here $`W(x)`$ is the Lambert $`W`$ function defined via
$$ye^y=xy=W(x).$$
(112)
Since $`Z>0`$ for real $`p`$, only sol.2 represents a physical solution. (Sol.1 holds for the special case of massless particles with no interactions.)
Of the four types of the determining equations (76)-(79) in which $`𝒥_\mathrm{\Lambda }`$ is replaced by $`𝒥`$, tanh-type A is a generalization of sol.2 and tanh-type B is that of sol.3. Tan-type A and tan-type B are new equations appearing specifically in the repulsive case and sol.1 is a special case of tan-type A and B solutions with $`𝒥=0`$ and $`q=0`$. Tan-type A and tan-type B trajectories yield a countably infinite series of unbounded motions of the particles. This is in strong contrast with the Newtonian case, in which only one trajectory exists for fixed $`H`$ and $`|q|`$.
Fig.10 is a diagram of the physical region of $`(|q|,p)`$ parameter space in the case of $`H_0=3`$. The shaded area of $`𝒥^2>0`$ and $`B>0`$ is the region where tanh-type A and tanh-type B give the actual trajectories. The boundary of this area is fixed by $`p=\pm p_0=\pm \sqrt{(H_0/2)^2m^2}`$ and $`p=\sqrt{2/\kappa }|q|+H_0/2`$. The values of $`|q|`$ at the intersections of these boundary lines are denoted as $`q_c`$ and $`q_d`$, of which $`q_c`$ is the critical value (108). In this area the tanh-type B solution is realized in a quite narrow region between $`p=\sqrt{2/\kappa }|q|+H_0/2`$ and $`𝒥B=0`$ (dashed curve). The motions of tan-type A, B are realized in the area of $`𝒥^2<0`$ whose boundaries are $`p=\pm \sqrt{2/\kappa }|q|+H_0/2`$. Consider a $`|q|=const`$ line (the dotted line in Fig.10) and define by $`p_1`$ and $`p_2`$ the momenta of the intersections of the line with $`p=\pm \sqrt{2/\kappa }|q|+H_0/2`$ .
For the case of $`0<|q|<q_c`$ the allowed value of $`p`$ is divided into two parts: $`p_0<p<p_0`$ and $`p_2<p<p_1`$. The solution in the former region is tanh-type A and the solutions in the latter region are both tan-type A and B. We show in Fig.11 the phase space trajectories for $`H_0=3`$ and $`|q|=0.25`$, in which the solid curve ($`N`$) denotes the bounded motion given by tanh-type A, and the dotted ($`A_n:n=0,1,\mathrm{}`$) and the dashed ($`B_n:n=1,2,\mathrm{}`$) curves represent the infinite series of unbounded motions specified by tan-type A and tan-type B, respectively. For the unbounded motions $`p_1`$ and $`p_2`$ are the asymptotic values of the momentum and the two particles simply approach one another at some minimal value of $`|r|`$ and then reverse direction toward infinity. In the figure we added also the trajectory of flat-space electrodynamics (a dot-dashed curve), which is the only solution of the theory for given values of $`H_0`$ and $`|q|`$. The motion of tan-type A, B has a specific feature that $`r(\tau )`$ becomes infinite at a finite proper time (but an infinite coordinate time). In Appendix C we shall present a simple model in flat space-time that has this feature. Fig.12 shows the $`r(\tau )`$ plots for $`H_0=3`$ and $`|q|=0.25`$.
As $`|q|`$ approaches $`q_c`$ the trajectories “$`N`$ ” and “$`A_0`$ ” come close to one another, meeting at $`|q|=q_c`$, and then for $`|q|`$ beyond the critical value they form new unbounded trajectories “$`N1`$” and “$`N2`$” as shown in Fig.13. This latter case will be discussed in the next subsection.
It should be stressed that the existence of two types of motion for fixed $`H`$ and $`q`$ is a new aspect of the relativistic gravitation theory, and has no non-relativistic analogue.
### 5.3 The repulsive case: the large $`|q|`$ regime ($`|q|>q_c`$)
For $`|q|`$ larger than $`q_c`$ the electric repulsive force overwhelms the attractive gravitational one and only unbounded motion is allowed. For the case of $`q_c<|q|<q_d`$ the allowed value of $`p`$ is $`p_0<p<p_1`$. Fig.14 shows the phase space trajectories of $`H_0=3`$ and $`|q|=0.3`$. Here the new unbounded trajectories $`N1`$ and $`N2`$ are realized instead of $`N`$ and $`A_0`$. The solution corresponding to the shaded area $`p_0<p<p_2`$ is $`N2`$, while the solutions for $`p_2<p<p_1`$ are $`N1`$, $`A_n`$ and $`B_n(n=1,2,\mathrm{})`$ and they are all described by tan-type A and B. The trajectories in $`r<0`$ region are obtained from those in $`r>0`$ by replacing the signs of both $`r`$ and $`p`$. By comparing all these trajectories with the analogous trajectory in flat space electrodynamics(a dot-dashed curve), we can see how the effects of gravity deform the flat-space trajectory. (Remember that there exist also time-reversed trajectories with $`ϵ=1`$.)
The phase space trajectories for the case of $`|q|=2>q_d`$ are depicted in Fig.15. Since $`p_2`$ is smaller than $`p_0`$ all solutions are tan-type A and B, and a characteristic cusp appears at $`r=0`$ in the trajectories $`N1`$ and $`N2`$. In the figure the trajectory specified by the symbol $`B01`$ is a combination of $`B_0+B_1`$, indicating that the solution switches between $`B_0`$ and $`B_1`$, namely, $`B_0`$ for $`p_0pp_0`$ and $`B_1`$ for $`p_2pp_0`$ and $`p_0pp_1`$. Similarly $`B12`$ is composed of a combination of $`B_1+B_2`$.
### 5.4 $`H<2m`$ case for repulsive charges
In 1+1 flat-space electrodynamics we know that for attractive charges the total energy of the system is restricted to $`H2m`$, but for repulsive charges no restriction on $`H`$ exists. Unbounded motion is also realized for $`H<2m`$ (the explicit solution is presented in Appendix B). It is expected that in a general relativistic theory the restriction on $`H`$ is identical. ¿From the $`(|q|,p)`$ diagram in Fig.10 we know that for $`H<2m`$ the shaded area disappears and only the region of $`𝒥^2<0`$ remains for the unbounded motions. In Fig.16 we show the phase space trajectories for $`H=1`$ and $`|q|=1`$. All types of unbounded motions $`A_n`$ and $`B_n`$ are realized. The unbounded trajectories $`N1`$ and $`N2`$ that appeared in $`H2m`$ turn to $`A_0`$ again. As compared with the flat-space trajectory (a dot-dashed curve) all trajectories are curved more toward the $`r`$-axis (due to the additional effect of gravitational attraction) and are shifted in the direction of the positive $`p`$-axis. This shift is caused by the $`p`$-dependence of the gravitational potential. As remarked previously there exist corresponding trajectories with $`ϵ=1`$ and invariance under time-reversal is retained.
## 6 ELECTRODYNAMICS WITH $`\mathrm{\Lambda }_e0`$
In a recent paper we presented a detailed analysis of two-body motion with no charge, namely particle dynamics in lineal gravity. The effects of a cosmological constant ($`\mathrm{\Lambda }_e=\mathrm{\Lambda }`$) may be incorporated into a momentum-dependent potential between particles, as shown in the two parameters expansion in terms of $`\kappa `$ and $`\mathrm{\Lambda }_e/\kappa ^2`$
$`H`$ $`=`$ $`2\sqrt{p^2+m^2}+{\displaystyle \frac{\kappa }{4}}(\sqrt{p^2+m^2}ϵ\stackrel{~}{p})^2|r|+{\displaystyle \frac{\kappa ^2}{4^2}}(\sqrt{p^2+m^2}ϵ\stackrel{~}{p})^3r^2`$ (113)
$`+{\displaystyle \frac{7\kappa ^3}{6\times 4^3}}(\sqrt{p^2+m^2}ϵ\stackrel{~}{p})^4|r|^3{\displaystyle \frac{\mathrm{\Lambda }_e}{2\kappa }}{\displaystyle \frac{ϵ\stackrel{~}{p}}{\sqrt{p^2+m^2}}}|r|{\displaystyle \frac{\mathrm{\Lambda }_e}{16}}{\displaystyle \frac{ϵ\stackrel{~}{p}m^2}{p^2+m^2}}r^2`$
$`+{\displaystyle \frac{\mathrm{\Lambda }_e^2}{4\kappa ^3}}{\displaystyle \frac{ϵ\stackrel{~}{p}}{(p^2+m^2)^{3/2}}}|r|+\mathrm{}.`$
The exact phase space trajectories $`(r(\tau ),p(\tau ))`$ indicate that a negative cosmological constant $`\mathrm{\Lambda }<0`$ acts effectively as an attractive force leading to bounded (periodic) motions, which are specified by the tanh-type A equation (76), whereas a positive cosmological constant acts effectively as a repulsive force. One noteworthy special situation takes place for a particular range of negative $`\mathrm{\Lambda }`$ and small $`m`$: both $`r(\tau )`$ and the phase space trajectory have a double peak structure. An example is shown in Figs.17 and 18 for $`H_0=10,m=0.02`$ and $`\mathrm{\Lambda }=0.5`$. Two particles starting at $`r=0`$ with initial momenta in opposite directions depart one another, reach a maximum separation, and then reverse direction due to the attractive force. However at some point they reverse direction again, reaching a second maximum before returning to the starting point. This peculiar behavior takes place due to the induced $`p`$-dependent $`\mathrm{\Lambda }`$ potential combined with the gravitational attraction and the relativistic effect.
For $`0<\mathrm{\Lambda }<\mathrm{\Lambda }_c=\frac{\kappa ^2m^4}{2(H^24m^2)}`$, both bounded and unbounded motions are realized for a fixed value of $`H`$, as with the motions depicted in Fig.11. For $`\mathrm{\Lambda }>\mathrm{\Lambda }_c`$ only unbounded motion is realized analogous to the motions shown in Fig.14.
In the general case of electrodynamics with a cosmological constant the dynamics of particles is governed by a combination of four factors: gravitational attraction, the electric force between charges, the effect of the cosmological constant and relativistic effects. The solution is characterized by the signs of $`\gamma _m`$ and $`\kappa ^2\left(H_0mf(\tau )+\frac{m}{f(\tau )}\right)^28\kappa e_1e_28\mathrm{\Lambda }_e`$. We shall investigte the motion in four combinations of the signs of $`\mathrm{\Lambda }_e`$ and the charge, separately.
### 6.1 $`\mathrm{\Lambda }_e<0`$ and attractive charges
In this case all interactions (gravitational, cosmological and electric) are attractive and $`𝒥_\mathrm{\Lambda }^2`$ is positive.The motion becomes necessarily bounded and described by the tanh-type solution. The period of the bounded motion is
$$T=\{\begin{array}{cc}\frac{16}{\kappa m\sqrt{\gamma _m}}\mathrm{tanh}^1\left(\frac{2p_0\sqrt{\gamma _m}(1+\sqrt{\gamma _H})H}{[H(1+\sqrt{\gamma _H})\gamma _e\sqrt{p_0^2+m^2}]^2\gamma _e^2p_0^2\gamma _mm^2}\right)\hfill & \gamma _m>0\text{,}\hfill \\ \frac{32p_0(1+\sqrt{\gamma _H})H}{\kappa m\left\{[H(1+\sqrt{\gamma _H})\gamma _e\sqrt{p_0^2+m^2}]^2\gamma _e^2p_0^2\right\}}\hfill & \gamma _m=0\text{,}\hfill \\ \frac{16}{\kappa m\sqrt{\gamma _m}}\mathrm{tan}^1\left(\frac{2p_0\sqrt{\gamma _m}(1+\sqrt{\gamma _H})H}{[H(1+\sqrt{\gamma _H})\gamma _e\sqrt{p_0^2+m^2}]^2\gamma _e^2p_0^2\gamma _mm^2}\right)\hfill & \gamma _m<0\text{.}\hfill \end{array}$$
(114)
For negative $`\mathrm{\Lambda }_e`$ motion is allowed for total energy larger than $`\sqrt{|\mathrm{\Lambda }_e|/\kappa ^2}`$. As the total energy increases both the period and the amplitude of the motion become large and the trajectory $`r(\tau )`$ deforms just as shown in Fig.1 in $`\mathrm{\Lambda }_e=0`$ case.
The main purpose in this sub-section is to investigate the effects of $`\mathrm{\Lambda }_e`$ on the motion, especially on how the double-peak structure appears in the system of charged particles. We find that the double peak is caused by the interplay among the $`\mathrm{\Lambda }`$ potential, gravitational attraction and relativistic effects, and is suppressed as the attractive force between charges becomes strong. Fig.19 shows the $`r(\tau )`$ plots for $`H_0=10,m=0.02,|q|=0.1`$ and four different values of negative $`\mathrm{\Lambda }_e`$. The double peak appears for $`\mathrm{\Lambda }_e=0.5`$. As $`\mathrm{\Lambda }_e`$ approaches its lower bound $`\kappa ^2H^2/8`$, the form of the phase space trajectories changes from an $`S`$-shaped curve to a double peaked one and then to a diamond shape as depicted in Fig.20. In Fig.21 we trace how the double peak in Fig.19 is affected by the value of charge $`|q|`$. We see that for small values of $`|q|`$ the double peak structure survives, but it disappears for large $`|q|`$.
### 6.2 $`\mathrm{\Lambda }_e<0`$ and repulsive charges
For the case of repulsive charges $`𝒥_\mathrm{\Lambda }^2`$ may become negative. We can classify the solutions in terms of the $`(|q|,p)`$ diagram as was used in the sub-section 5.2. The boundary of $`𝒥_\mathrm{\Lambda }^2=0`$ is given by
$$p=\sqrt{\left(\frac{H}{2}\right)^2\frac{2|\mathrm{\Lambda }_e|}{\kappa ^2}}\pm \sqrt{\frac{2}{\kappa }\left(q^2\frac{|\mathrm{\Lambda }_e|}{\kappa }\right)}$$
(115)
There are two types of $`(|q|,p)`$ diagram depending on whether the vertex $`(q_0=\sqrt{\frac{|\mathrm{\Lambda }_e|}{\kappa }},\sqrt{\left(\frac{H}{2}\right)^2\frac{2|\mathrm{\Lambda }_e|}{\kappa ^2}})`$ of $`𝒥_\mathrm{\Lambda }^2=0`$ is within the region $`p_0pp_0:p_0=\sqrt{(H/2)^2m^2}`$, or not. Fig.22 is the diagram for $`m>\sqrt{2|\mathrm{\Lambda }_e|}/\kappa `$ where the vertex is outside the region $`p_0pp_0`$. The physical region consists of the shaded area of $`𝒥_\mathrm{\Lambda }^2>0`$ and $`B>0`$, and the area of $`𝒥_\mathrm{\Lambda }^2<0`$. The former area is the region of tanh-type solutions and the latter is the region of tan-type solutions. A narrow region between $`𝒥_\mathrm{\Lambda }B=0`$ (dashed line) and $`𝒥_\mathrm{\Lambda }=0`$ in the shaded area is the region corresponding to tanh-type B solution. The $`q_c`$ and $`q_d`$ are the values of $`|q|`$ at the intersections of $`𝒥_\mathrm{\Lambda }^2=0`$ with $`p=\pm p_0`$, namely, $`q_c=\sqrt{\frac{\kappa }{2}\left\{\sqrt{(H/2)^22|\mathrm{\Lambda }_e|/\kappa ^2}\sqrt{(H/2)^2m^2}\right\}+|\mathrm{\Lambda }_e|/\kappa }`$ and $`q_d=\sqrt{\frac{\kappa }{2}\left\{\sqrt{(H/2)^22|\mathrm{\Lambda }_e|/\kappa ^2}+\sqrt{(H/2)^2m^2}\right\}+|\mathrm{\Lambda }_e|/\kappa }`$.
The motions corresponding to this diagram are classified into three categories according to the $`|q|`$ value:
(i) $`|q|q_0`$ : bounded motion,
(ii) $`q_0<|q|<q_c`$ : both bounded and unbounded motions,
(iii) $`q_c|q|`$ : unbounded motion.
For a small $`|q|`$ in the category (i), the attractive effect of $`\mathrm{\Lambda }_e<0`$ is stronger than the repulsive effect between charges $`(|\mathrm{\Lambda }_e|>\kappa q^2)`$ and $`𝒥_\mathrm{\Lambda }^2`$ is positive. The phase space trajectories resemble the trajectories in Fig.4, but they become more $`S`$-shaped as $`|q|`$ increases. For the categories (ii) and (iii) the $`(|q|,p)`$ diagram is nearly the same as the diagram in Fig.10. The phase space trajectories for (ii) are just like those in Fig.11. The trajectories of the motion for the category (iii) are divided into two cases of $`q_c|q|q_d`$ and $`q_d|q|`$. In the case of $`q_c|q|q_d`$ the trajectories are analogous to those in Fig.14 in which the $`N_2`$ trajectory is a tanh-type, while in the case of $`q_d|q|`$ all trajectories are tan-type and like those in Fig.15 where $`N_1`$ and $`N_2`$ have cusps at $`r=0`$. In Fig.23 the $`r(\tau )`$ plot for each category is depicted for the parameters $`H_0=3,m=1.2`$ and $`\mathrm{\Lambda }_e=0.1`$ with $`q_0=0.316`$ and $`q_c=0.491`$: $`|q|=0.1,0.4`$ and $`0.6`$ for the categories (i), (ii) and (iii), respectively. As $`|q|`$ becomes large, both the period and the amplitude of the motion increase and finally the motion becomes unbounded, because the repulsive force between charges prevails over the attractive forces of gravity and the cosmological constant. Fig.24 shows the corresponding phase space trajectories.
Since the analysis in sub-section 6.1 indicates that the double peak structure appears for negative $`\mathrm{\Lambda }_e`$, small $`m`$, and small attractive $`|q|`$, it can be inferred that for the repulsive charges, if $`|q|`$ is sufficiently small, the double peak still survives. The smallness of $`m`$ corresponds to the condition $`m\sqrt{2|\mathrm{\Lambda }_e|}/\kappa `$ for which the $`(|q|,p)`$ diagram is given in Fig.25 and the vertex of $`𝒥_\mathrm{\Lambda }^2=0`$ is within the region $`p_0pp_0`$. The motions are classified into two categories:
(i) $`|q|q_0`$ : bounded motion or unbounded motion,
(ii) $`q_0<|q|`$ : unbounded motion.
The physical region of the category (i) belongs to the shaded area of $`𝒥_\mathrm{\Lambda }^2>0`$ and $`B>0`$.
We can find the double peak structure for a certain range of the parameters as shown in Fig.26, in which $`H_0=10,m=0.02,\mathrm{\Lambda }_e=0.5`$ and $`|q|=0,\mathrm{\hspace{0.33em}0.1},\mathrm{\hspace{0.33em}0.14}`$.
The above parameters of $`m,\mathrm{\Lambda }_e`$ and $`|q|`$ correpond to $`\gamma _m<0`$. As $`|q|`$ exceeds $`q_m\sqrt{m\sqrt{2|\mathrm{\Lambda }_e|}\kappa m^2/2}`$, the solution to $`\gamma _m=0`$, we encounter a new situation. The maximum turning point of $`r`$ extends to infinity and the trajectory splits into two nonperiodic motions as shown in Fig.27 and Fig.28.
For $`|q|=q_m`$ the asymptotic value of $`p`$ is $`m\sqrt{(H^2|\mathrm{\Lambda }_e|)/8|\mathrm{\Lambda }_e|}`$ and in the $`(|q|,p)`$ diagram of Fig.25 this corresponds to the vertex point of a dashed-curve of $`𝒥_\mathrm{\Lambda }B=0`$. For $`q_m<|q|<q_0`$ the two asymptotic values of $`p`$ are the intersections of $`|q|=const`$ line with $`𝒥_\mathrm{\Lambda }B=0`$ curve. The region between these asymptotic values originally belongs to tanh-type B solution but no trajectory exists because $`𝒥_\mathrm{\Lambda }>H|mfm/f|`$ for the parameters in this region.
The trajectories of the motion for category (ii) are diveded into three cases of $`q_0<|q|<q_c`$, $`q_c|q|q_d`$ and $`q_d<|q|`$. For $`q_0<|q|<q_c`$ under the condition $`m\sqrt{2|\mathrm{\Lambda }_e|}/\kappa `$ , a $`𝒥_\mathrm{\Lambda }^2<0`$ region is sandwiched by the shaded areas, in contrast to $`m>\sqrt{2|\mathrm{\Lambda }_e|}/\kappa `$ case where the region for $`q_0<|q|<q_c`$ is separated into two parts, namely, a shaded area and a $`𝒥_\mathrm{\Lambda }^2<0`$ area. A typical phase space trajectory for $`q_0<|q|<q_c`$ is shown in Fig.29 where the parameters are $`H_0=3,m=0.6,\mathrm{\Lambda }_e=1`$ and $`|q|=1.1`$. Here between two split nonperiodic trajectories (solid curves) there appear an infinite series of tan-type A, B solutions of the unbounded motions. For the motions of the cases $`q_c|q|q_d`$ and $`q_d<|q|`$ the outlines of the unbounded trajectories are easily inferred from Figs.14, 15 and 29.
### 6.3 $`\mathrm{\Lambda }_e>0`$ and attractive charges
For the attractive charges the boundary curve of $`𝒥_\mathrm{\Lambda }^2=0`$ is
$$p=\sqrt{\left(\frac{H}{2}\right)^2+\frac{2\mathrm{\Lambda }_e}{\kappa ^2}}\pm \sqrt{\frac{2}{\kappa }\left(\frac{\mathrm{\Lambda }_e}{\kappa }q^2\right)}$$
(116)
which opens out to the direction of $`p`$-axis. Fig.30 is the $`(|q|,p)`$ diagram for
$`H<\sqrt{\kappa ^2m^2/2\mathrm{\Lambda }_e+4m^2}`$. The motions for this diagram are classified into two categories:
(i) $`0|q|<q_0`$ : both bounded and unbounded motions,
(ii) $`q_0<|q|`$ : bounded motion.
The phase space trajectories for category (i) are just like the trajectories in Fig.11, while for category (ii) the motions become simply bounded as the attractive charges overwhelm the repulsive effect of the effective cosmological constant.
Under the condition of $`H\sqrt{\kappa ^2m^2/2\mathrm{\Lambda }_e+4m^2}`$ the situation is slightly different from the previous case, as shown in the $`(|q|,p)`$ diagram of Fig.31. In this case the motions are classified into three categories according to the $`|q|`$ value:
(i) $`0<|q|q_c`$ : unbounded motion,
(ii) $`q_c<|q|q_0`$ : both bounded and unbounded motions,
(iii) $`q_0<|q|`$ : bounded motion,
where $`q_c=\sqrt{\mathrm{\Lambda }_e/\kappa (\kappa /2)\left\{\sqrt{(H/2)^2+2\mathrm{\Lambda }_e/\kappa ^2}\sqrt{(H/2)^2m^2}\right\}^2}`$. Since this $`(|q|,p)`$ diagram is like the inverse of the diagram in Fig.22, categories (i), (ii) and (iii) correspond to the previous categories (iii), (ii) and (i) of $`m>\sqrt{2|\mathrm{\Lambda }_e|}/\kappa `$ in sub-section 6.2, respectively. In Fig.32 the $`r(\tau )`$ plot for each category is drawn for the parameters $`H_0=3,m=1`$ and $`\mathrm{\Lambda }_e=1`$ with $`q_0=1`$ and $`q_c=0.745`$: $`|q|=0.5,0.8`$ and $`1.5`$ for the category (i), (ii) and (iii), respectively. As $`|q|`$ increases, the motion changes from unbounded to bounded and then both the period and the amplitude decrease.The phase space trajectory for each $`r(\tau )`$ is easily inferred from Fig.24.
### 6.4 $`\mathrm{\Lambda }_e>0`$ and repulsive charges
In this case, since the cosmological repulsion is commensurate with electromagnetic repulsion, classification of the motion is analogous to that of repulsive charges with $`\mathrm{\Lambda }_e=0`$ in sub-section 5.2. Under the condition $`H<\sqrt{\kappa ^2m^2/2\mathrm{\Lambda }_e+4m^2}`$ the $`(|q|,p)`$ diagram is given by Fig.33 which has the same pattern as Fig.10. The physical region is classified into two categories:
(i) $`|q|<q_c`$ : both bounded and unbounded motions,
(ii) $`q_c|q|`$ : unbounded motion.
The trajectories of the motion for the category (ii) are divided into two cases of $`q_c|q|q_d`$ and $`q_d<|q|`$. The phase space trajectories and $`r(\tau )`$ plots for $`|q|<q_c`$ are analogous to those in Fig.11 and 12. For a larger $`H`$ of $`H\sqrt{\kappa ^2m^2/2\mathrm{\Lambda }_e+4m^2}`$ the motion is always unbounded as inferred from the diagram of Fig.34.
## 7 MOTION OF UNEQUAL MASSES
What time variable is adequate for the analysis of the motion of the two particles with unequal masses? The proper time (83) of each particle is different as
$`d\tau _1`$ $`=`$ $`dt{\displaystyle \frac{16YK_0K_1m_1}{JKM_1\sqrt{p^2+m_1^2}}},`$
$`d\tau _2`$ $`=`$ $`dt{\displaystyle \frac{16YK_0K_2m_2}{JKM_2\sqrt{p^2+m_2^2}}},`$
where $`KK_+=K_{}`$ and $`YY_+=Y_{}`$. As with our analysis of the electrically neutral case , we propose using
$$d\stackrel{~}{\tau }dt\frac{16YK_0}{JK}\left(\frac{K_1K_2m_1m_2}{M_1M_2\sqrt{p^2+m_1^2}\sqrt{p^2+m_1^2}}\right)^{1/2},$$
(118)
which is symmetric with respect to $`12`$ and reduces to the proper time (84) when $`m_1=m_2`$.
In terms of this variable the canonical equations are expressed as
$`{\displaystyle \frac{dp}{d\stackrel{~}{\tau }}}`$ $`=`$ $`{\displaystyle \frac{1}{4\kappa }}\left({\displaystyle \frac{K_1K_2M_1M_2\sqrt{p^2+m_1^2}\sqrt{p^2+m_2^2}}{m_1m_2}}\right)^{1/2},`$ (119)
$`{\displaystyle \frac{dz_i}{d\stackrel{~}{\tau }}}`$ $`=`$ $`(1)^{i+1}\left({\displaystyle \frac{M_1M_2\sqrt{p^2+m_1^2}\sqrt{p^2+m_2^2}}{K_1K_2m_1m_2}}\right)^{1/2}\left\{{\displaystyle \frac{ϵJ}{16K_0}}+{\displaystyle \frac{K_i}{M_i}}\left({\displaystyle \frac{p}{\sqrt{p^2+m_i^2}}}ϵ{\displaystyle \frac{Y}{K}}\right)\right\},`$
$`{\displaystyle \frac{dr}{d\stackrel{~}{\tau }}}`$ $`=`$ $`\left({\displaystyle \frac{M_1M_2\sqrt{p^2+m_1^2}\sqrt{p^2+m_2^2}}{K_1K_2m_1m_2}}\right)^{1/2}`$ (121)
$`\times \left\{{\displaystyle \frac{ϵJ}{8K_0}}+{\displaystyle \frac{K_1}{M_1}}\left({\displaystyle \frac{p}{\sqrt{p^2+m_1^2}}}ϵ{\displaystyle \frac{Y}{K}}\right)+{\displaystyle \frac{K_2}{M_2}}\left({\displaystyle \frac{p}{\sqrt{p^2+m_2^2}}}ϵ{\displaystyle \frac{Y}{K}}\right)\right\}.`$
Note that $`r`$ still describes the proper distance between the particles at any fixed instant.
Unlike the equal mass case, the integration $`𝑑p(K_1K_2M_1M_2\sqrt{p^2+m_1^2}\sqrt{p^2+m_2^2})^{1/2}`$ can not be performed within the framework of elementary calculus. We resort to numerical calculation for solving above equations.
As in Sec.VI we analyze the motions by plotting $`r(\tau )`$ in four combinations of the signs of $`\mathrm{\Lambda }_e`$ and the charges. In the case of a negative cosmological constant and attractive charges, the $`r(\tau )`$ plots are shown in Fig.35 for various mass ratios $`m_1/m_2`$ in the fixed $`H_0=10,m_2=1,\mathrm{\Lambda }_e=1`$ and $`|q|=1`$. When the mass ratio gets larger than unity (the case of $`m_1=m_2=1`$), the gravitational attraction is stronger and the proper distance between particles as well as the period becomes shorter. As the mass ratio gets smaller gravity becomes weak; however for quite small mass ratios the attractive effect due to the cosmological constant prevails and the period changes to being short again. The effect of the cosmological constant is seen by changing $`|q|`$ small but preserving the values of other parameters. Fig.36 shows the $`r(\tau )`$ plots for $`|q|=0.1`$. While both the proper distance and the period becomes large as a whole, the double peak structures appear for a small mass ratio.
In the case of $`\mathrm{\Lambda }_e<0`$ and repulsive charges unbounded trajectories appear as $`|q|`$ increases. An example is shown in Fig.37, in which $`r(\tau )`$ for a small mass ratio $`m_1/m_2=0.1`$ becomes unbounded because the repulsive force between charges exceeds the weak gravity and the $`\mathrm{\Lambda }`$ force.
In the case of $`\mathrm{\Lambda }_e>0`$ and attractive charges, as we analyzed in Sec.VI the trajectory changes from unbounded to bounded as $`|q|`$ increases. In Fig.38 we plot $`r(\tau )`$ for various $`m_1/m_2`$ in the fixed $`H_0=10,m_2=1,\mathrm{\Lambda }_e=0.1`$ and $`|q|=0.1`$, from which the strong repulsive effect of a positive $`\mathrm{\Lambda }_e`$ is perceived. The motions for the case of $`\mathrm{\Lambda }_e>0`$ and the repulsive charges are mostly unbounded except for the case of the large masses.
## 8 STATIC BALANCE IN (1+1) DIMENSIONS
In this section we treat the problem of static balance in (1+1) dimensions. This problem originated in attempts to find the exact solution of the $`N(N2)`$ body system in Einstein’s theory of general relativity . In $`(3+1)`$ dimensions gravitational radiation carries away energy from a system of particles, and so exact solutions have so far been unobtainable. One simple way to search an exact solution for $`N2`$ is to balance the gravitational attraction with some extra repulsive force, a natural candidate of which is the electric force. The first trial was achieved by Majumdar and Papapetrou for $`N=2`$ and afterwards it was generalized to $`N`$ bodies on a line . Their condition for balance is
$$e_i=\pm \sqrt{4\pi G}m_i,$$
(122)
and is much more strict than the corresponding condition in Newtonian theory
$$Gm_1m_2\frac{e_1e_2}{4\pi }=0.$$
(123)
Since then, the question has long been raised as to why the balance condition differs in the relativistic and non-relativistic cases. Some people conjectured that there should be an exact solution in general relativity under the condition (123). Others showed in the (2nd) post-Newtonian approximation that the condition (123) is incompatible with the static balance condition in general relativity. Clearly (122) is a sufficient condition for static balance but there exists no proof that (122) is a necessary condition. On the other hand a test particle analysis suggested that the condition (123) was neither necessary nor sufficient, but a separation-dependent balance position might exist in general relativity. Recently several numerical trials on the separation-dependence were reported , but so far no one has been able -in any relativistic theory of gravity- to succeed in finding analytically an exact solution under (123) or another (separation-dependent) condition.
In (1+1) dimensions the absence of radiation makes it easy to fix the balance condition in terms of the determining equation (59). From the relation
$$\frac{H}{r}=\frac{4K_0K_1K_2}{\kappa J},$$
(124)
the balance condition $`H/r=0`$ leads to
$$K_1K_2=(2K_0+2K\kappa \sqrt{p^2+m_1^2})(2K_0+2K\kappa \sqrt{p^2+m_2^2})=0,$$
(125)
where $`KK_\pm =\kappa X`$ and $`H=\frac{4}{\kappa }K`$. In deriving (125) we excluded the possibility that $`K_0=0`$ because it gives an unphysical solution.
It is evident from (59) that the condition (125) means at the same time $`M_1M_2=0`$, namely,
$$(2K_02K+\kappa \sqrt{p^2+m_1^2})(2K_02K+\kappa \sqrt{p^2+m_2^2})=0,$$
(126)
The equations (125) and (126) lead to
$$4K_0^2+(2K\kappa \sqrt{p^2+m_1^2})(2K\kappa \sqrt{p^2+m_2^2})=0$$
(127)
and
$$K_0(4K\kappa \sqrt{p^2+m_1^2}\kappa \sqrt{p^2+m_2^2})=0.$$
(128)
The solution to (128) is
$$H=\sqrt{p^2+m_1^2}+\sqrt{p^2+m_2^2}$$
(129)
and the insertion of this solution into (127) leads to
$$\frac{\kappa }{2}(\sqrt{p^2+m_1^2}ϵ\stackrel{~}{p})(\sqrt{p^2+m_2^2}ϵ\stackrel{~}{p})e_1e_2=0.$$
(130)
This condition is the force-balance condition and fixes the value of momentum $`p=p_c=\text{const}`$. Under the condition (130) the two particles move with a constant velocity. The condition (130) and the Hamiltonian (129) indicate that the full Hamiltonian must have the simple structure
$$H=\sqrt{p^2+m_1^2}+\sqrt{p^2+m_2^2}+\left\{\frac{\kappa }{2}(\sqrt{p^2+m_1^2}ϵ\stackrel{~}{p})(\sqrt{p^2+m_2^2}ϵ\stackrel{~}{p})e_1e_2\right\}F(|r|,p).$$
(131)
Actually, in the case of $`\mathrm{\Lambda }_e=0`$ the $`\kappa `$-expansion of the Hamiltonian leads to
$`H`$ $`=`$ $`\sqrt{p^2+m_1^2}+\sqrt{p^2+m_2^2}{\displaystyle \frac{1}{2}}e_1e_2|r|`$ (132)
$`+{\displaystyle \frac{\kappa }{4}}\{\sqrt{p^2+m_1^2}\sqrt{p^2+m_2^2}|r|ϵ\stackrel{~}{p}(\sqrt{p^2+m_1^2}+\sqrt{p^2+m_2^2})|r|+p^2|r|`$
$`{\displaystyle \frac{1}{4}}e_1e_2(\sqrt{p^2+m_1^2}+\sqrt{p^2+m_2^2})r^2+{\displaystyle \frac{1}{2}}ϵe_1e_2\stackrel{~}{p}r^2+{\displaystyle \frac{1}{24}}(e_1e_2)^2|r|^3\}`$
$`+{\displaystyle \frac{\kappa ^2}{4^2}}\{{\displaystyle \frac{1}{2\times 4}}r^2(\sqrt{p^2+m_1^2}+\sqrt{p^2+m_2^2}2ϵ\stackrel{~}{p}{\displaystyle \frac{1}{2}}e_1e_2|r|)^3`$
$`+{\displaystyle \frac{1}{3\times 4^2}}e_1e_2|r|^3(\sqrt{p^2+m_1^2}+\sqrt{p^2+m_2^2}2ϵ\stackrel{~}{p}{\displaystyle \frac{1}{2}}e_1e_2|r|)^2`$
$`{\displaystyle \frac{1}{6\times 4^2}}r^2\left[12(\sqrt{p^2+m_1^2}\sqrt{p^2+m_2^2})^2+(e_1e_2)^2r^2\right]`$
$`\times (\sqrt{p^2+m_1^2}+\sqrt{p^2+m_2^2}2ϵ\stackrel{~}{p}{\displaystyle \frac{1}{2}}e_1e_2|r|)`$
$`+{\displaystyle \frac{1}{3\times 4^2}}e_1e_2|r|^3(\sqrt{p^2+m_1^2}\sqrt{p^2+m_2^2})^2{\displaystyle \frac{1}{15\times 4^3}}(e_1e_2)^3|r|^5\}+𝒪(\kappa ^3).`$
From (132) we get explicitly
$$F(|r|,p)=\frac{1}{2}|r|+\frac{\kappa }{2}(\sqrt{p^2+m_1^2}+\sqrt{p^2+m_2^2}2ϵ\stackrel{~}{p})r^2+𝒪(|r|^3).$$
(133)
The solution to the condition (130) exists only for $`e_1e_2>0`$;
$$p_c=\pm \frac{|\left(\frac{\kappa }{2}\right)^2m_1^2m_2^2e_1^2e_2^2|}{\sqrt{2\kappa e_1e_2}\sqrt{(\frac{\kappa }{2}m_1^2+e_1e_2)(\frac{\kappa }{2}m_2^2+e_1e_2)}}.$$
(134)
When the particles are initially at rest $`(p_c=0)`$, the condition (130) becomes
$$\frac{\kappa }{2}m_1m_2e_1e_2=0.$$
(135)
This is the condition of static balance and is identical with the condition in Newtonian theory in (1+1) dimensions. Note that in Newtonian theory (135) is the force-balance condition which includes both the static case and a uniform motion. However in the relativistic case (135) represents only the static balance condition – the condition of force-balance (130) in general depends on the momentum and allows a uniform motion in the C.I. ststem in which two particles approach or recede with the designated constant momentum (134).
The above result may seem to provide suggestive evidence that in (3+1) dimensions (123) could be a sufficient solution. However,the post- Newtonian Hamiltonian for the system of two charged bodies in (3+1) dimensions is given as
$`H`$ $`=`$ $`m_1+m_2+{\displaystyle \frac{𝒑_1^2}{2m_1}}+{\displaystyle \frac{𝒑_2^2}{2m_2}}{\displaystyle \frac{(𝒑_1^2)^2}{8m_1^3}}{\displaystyle \frac{(𝒑_2^2)^2}{8m_2^3}}`$ (136)
$`{\displaystyle \frac{Gm_1m_2}{r}}\left\{1+{\displaystyle \frac{3}{2}}\left({\displaystyle \frac{𝒑_1^2}{m_1^2}}+{\displaystyle \frac{𝒑_2^2}{m_2^2}}\right){\displaystyle \frac{7}{2}}{\displaystyle \frac{(𝒑_1𝒑_2)}{m_1m_2}}{\displaystyle \frac{(𝒑_1𝒓)(𝒑_2𝒓)}{2m_1m_2r^2}}\right\}`$
$`+{\displaystyle \frac{e_1e_2}{4\pi r}}\left\{1{\displaystyle \frac{(𝒑_1𝒑_2)}{2m_1m_2}}{\displaystyle \frac{(𝒑_1𝒓)(𝒑_2𝒓)}{2m_1m_2r^2}}\right\}+{\displaystyle \frac{G^2m_1m_2(m_1+m_2)}{2r^2}}`$
$`{\displaystyle \frac{G(m_1+m_2)e_1e_2}{4\pi r^2}}+{\displaystyle \frac{G(m_1e_2^2+m_2e_1^2)}{8\pi r^2}},`$
which is derived from Bażański Lagrangian and also the ADM canonical formalism. The structure of the Hamiltonian (136) is rather complicated compared to (131). For example in (1+1) dimensional theory the charges appear always in a combination $`e_1e_2`$, while in (3+1) dimensions there is a combination of $`m_1e_2^2+m_2e_1^2`$ and in higher approximations more complicated combinations of mass and charge appear . It is clear that the Hamiltonian (136) supports the balance condition (122) for the static case, while it does not correspond to a uniform motion. What causes this difference between (1+1) and (3+1) dimensional theories? Is it intrinsic to the dimensionality, or is there still a possibility for satisfying the balance condition (123) in (3+1) dimensions? Or is there a uniform motion in the C.I. system in (3+1) dimensions? These are interesting open problems for further investigation.
## 9 CONCLUSIONS
Since in (1+1) dimensions the degrees of freedom of both gravitational and electromagnetic radiation are frozen, one expects the motion of a set of $`N`$ charged particles in curved-spacetime with a cosmological constant to be described by a conservative Hamiltonian. And this is what we find to be the case. We began by canonically reducing the charged $`N`$-body action to first-order form and then for a system of two charged particles derived the determining equation of the Hamiltonian from the matching conditions. The canonical equations of motion given by the Hamiltonian can be solved exactly when they are expressed in terms of the proper time, and we have given the explicit solution. To our knowledge this is the first non-perturbative relativistic solution of this problem.
We recapitulate the main results of this paper:
(1) In (1+1) dimensions the square of the electric field plays the same role as the consmological constant and an overall constant part is incorporated into the effective cosmological constant $`\mathrm{\Lambda }_e\mathrm{\Lambda }_0\kappa (_ae_a)^2/4`$, which induces the momentum-dependent potential in the Hamiltonian. Effectively $`\mathrm{\Lambda }_e>0`$ acts on the particles as a repulsive potential and $`\mathrm{\Lambda }_e<0`$ acts as an attractive potential.
(2) The theory in the case $`\mathrm{\Lambda }_e=0`$ is a general relativistic electrodynamics which is an extension of flat space Newtonian electrodynamics. For the attractive charges the motion becomes bounded similar to $`|q|=0`$ case with smaller amplitude and period. For repulsive charges the electric force between two particles competes with the gravitational attractive force. For $`|q|<q_c`$ and a fixed energy not only bounded motion but also an infinite series of unbounded motions is realized. Such multiple solutions do not exist in Newtonian theory. For $`|q|>q_c`$ all motions become unbounded.
(3) The condition for static balance in (1+1) dimensional general relativistic electrodynamics is identical with the Newtonian balance condition in the flat space. This simple result contrasts strikingly with the unresolved situation in (3+1) dimensions.
(4) In spacetimes with nonzero $`\mathrm{\Lambda }_e`$ the motions are more complicated. For a fixed total energy $`H_0`$ and $`\mathrm{\Lambda }_e`$ the solutions are classified in terms of $`(|q|,p)`$ diagram in which the region of the parameters for the bounded and/or unbounded motions are depicted in terms of the curves of $`p=\pm p_0,𝒥_\mathrm{\Lambda }^2=0`$ and $`𝒥_\mathrm{\Lambda }B=0`$. By drawing a $`q=\text{const}`$ line on the diagram we can easily find what types of the motion are realized.
(5) For a certain range of $`\mathrm{\Lambda }_e<0`$ and a small $`|q|`$ the double peak structure appears in $`r(\tau )`$ and the phase space trajectory, which is caused by a subtle interplay amongst the momentum-dependent $`\mathrm{\Lambda }`$ potential, the gravitational and the electric potentials, and relativistic effects.
(6) For unbounded motion the mutual separation of two particles diverges at finite proper time. This is common for the cases of $`\mathrm{\Lambda }_e=0`$ and $`\mathrm{\Lambda }_e0`$.
(7) In the unequal mass case the basic features of the motion are the same, but a double peak structure appears more clearly than in the equal mass case.
A number of interesting questions arise from this work. First, it would be of interest to know how the features of the motion we have found are modified as one increases the number of bodies in the system. For neutral bodies this situation is relevant in modelling stellar systems and galactic evolution, where non-relativsitic one-dimensional self-gravitating systems are employed as tool toward understanding such dynamics. For charged bodies the relevance is most likely in terms of the physics of the early universe, where charged black holes, strings and domain walls interact in a highly relativistic setting. Second, it would be interesting to study the circumstances under which black hole event horizons can form – we expect this will involve investigating the regime where eq. (89) is not satisfied. A third possibility would involve generalizing our work to other (1+1) dilatonic theories of gravity. Moreover, a full quantum treatment of the problem would also be of considerable interest. We intend to turn our attention to these problems in the future.
## APPENDIX A: SOLUTION OF THE METRIC TENSOR
Under the coordinate conditions (35) the field equations (27), (28), (31) and (32) become
$`\dot{\pi }+N_0\left\{{\displaystyle \frac{3\kappa }{2}}\pi ^2+{\displaystyle \frac{1}{8\kappa }}(\mathrm{\Psi }^{})^2+{\displaystyle \frac{1}{4}}\left(V{\displaystyle \frac{\mathrm{\Lambda }_e}{\kappa }}\right){\displaystyle \frac{p_1^2}{2\sqrt{p_1^2+m_1^2}}}\delta (xz_1(t)){\displaystyle \frac{p_2^2}{2\sqrt{p_2^2+m_2^2}}}\delta (xz_2(t))\right\}`$
$`+N_1\left\{\pi ^{}+p_1\delta (xz_1(t))+p_2\delta (xz_2(t))\right\}+{\displaystyle \frac{1}{2\kappa }}N_0^{}\mathrm{\Psi }^{}+N_1^{}\pi =0,`$ (137)
$`\kappa \pi N_0+N_1^{}=0,`$ (138)
$`_1({\displaystyle \frac{1}{2}}N_0\mathrm{\Psi }^{}+N_0^{})=0,`$ (139)
$`\dot{\mathrm{\Psi }}+2\kappa N_0\pi N_1\mathrm{\Psi }^{}=0.`$ (140)
The procedure to solve these equations is just the same as that in the case of $`V=0`$, which is given in the Appendix of the previous paper .
The solutions have the same form as those of $`V=0`$ when they are expressed in terms of $`K_\pm ,K_0,K_{1,2},_{1,2},Y_\pm `$ and $`J`$. For $`r=z_1z_2>0`$ the metric tensor is determined as
$`N_{0(+)}(x)`$ $`=`$ $`{\displaystyle \frac{8}{J}}\left({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}}\right){\displaystyle \frac{K_0K_1}{_1}}e^{K_+(xz_1)},`$
$`N_{0(0)}(x)`$ $`=`$ $`{\displaystyle \frac{1}{2K_0J}}\left({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}}\right)\left[(K_1_1)^{\frac{1}{2}}e^{\frac{1}{2}K_0(xz_1)}+(K_2_2)^{\frac{1}{2}}e^{\frac{1}{2}K_0(xz_2)}\right]^2,`$
$`N_{0()}(x)`$ $`=`$ $`{\displaystyle \frac{8}{J}}\left({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}}\right){\displaystyle \frac{K_0K_2}{_2}}e^{K_{}(xz_2)},`$
$`N_{1(+)}`$ $`=`$ $`ϵ{\displaystyle \frac{Y_+}{K_+}}\left\{{\displaystyle \frac{8}{J}}\left({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}}\right){\displaystyle \frac{K_0K_1}{_1}}e^{K_+(xz_1)}1\right\},`$
$`N_{1(0)}`$ $`=`$ $`ϵ\{{\displaystyle \frac{Y_0}{2JK_0^2}}({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}})[K_2M_2e^{K_0(xz_2)}K_1M_1e^{K_0(xz_1)}`$
$`+2K_0(K_1K_2M_1M_2)^{1/2}e^{\frac{1}{2}K_0(z_1z_2)}x]+D_0\},`$
$`N_{1()}`$ $`=`$ $`ϵ{\displaystyle \frac{Y_{}}{K_{}}}\left\{{\displaystyle \frac{8}{J}}\left({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}}\right){\displaystyle \frac{K_0K_2}{_2}}e^{K_{}(xz_2)}1\right\},`$
where
$`D_0`$ $`=`$ $`{\displaystyle \frac{1}{2}}({\displaystyle \frac{Y_+}{K_+}}{\displaystyle \frac{Y_{}}{K_{}}})+{\displaystyle \frac{1}{2J}}({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}})\{2[({\displaystyle \frac{Y_0}{K_0}}+{\displaystyle \frac{Y_+}{K_+}})K_1({\displaystyle \frac{Y_0}{K_0}}+{\displaystyle \frac{Y_{}}{K_{}}})K_2]`$ (142)
$`2[({\displaystyle \frac{Y_0}{K_0}}{\displaystyle \frac{Y_+}{K_+}}){\displaystyle \frac{1}{_1}}({\displaystyle \frac{Y_0}{K_0}}{\displaystyle \frac{Y_{}}{K_{}}}){\displaystyle \frac{1}{_2}}]K_1K_2{\displaystyle \frac{Y_0}{K_0}}K_1K_2(z_1+z_2)\}.`$
The metric tensor for $`r<0`$ is simply obtained by interchanging the suffix 1 and 2. With these metric tensors and the canonical equations, the field equations (137), (138) and (139) are proved to hold in a whole $`x`$ space.
As we showed in the previous paper, to satisfy (140) the dilaton field $`\mathrm{\Psi }`$ needs an extra $`x`$-independent function $`f(t)`$, which has no effect on the dynamics of particles. After lengthy calculation Eq.(140) leads to
$`\dot{f}(t)`$ $`=`$ $`{\displaystyle \frac{d}{dt}}(K_{01}z_1K_{02}z_2)+{\displaystyle \frac{2}{J}}({\displaystyle \frac{Y_+}{K_+}}+{\displaystyle \frac{Y_{}}{K_{}}})\{2K_0K_1{\displaystyle \frac{p_1}{\sqrt{p_1^2+m_1^2}}}2K_0K_2{\displaystyle \frac{p_2}{\sqrt{p_2^2+m_2^2}}}`$ (143)
$`+ϵ{\displaystyle \frac{Y_0}{K_0}}(K_0K_1+K_0K_2{\displaystyle \frac{K_0K_1K_2}{_1}}{\displaystyle \frac{K_0K_1K_2}{_2}})+4ϵY_0K_0({\displaystyle \frac{K_1}{_1}}+{\displaystyle \frac{K_2}{_2}})\}.`$
Thus $`f(t)`$ is uniquely determined.
## APPENDIX B: LINEAR APPROXIMATION OF THE EXACT SOLUTION IN $`\mathrm{\Lambda }_e=0`$
In this appendix we investigate how the exact solution is related to its corresponding solution in Newtonian theory. Take the tanh-type A solution, namely a periodic motion with $`e_1e_2<0`$ or small $`e_1e_2>0`$. For motion starting from $`r=0`$ at $`\tau =\tau _0=0`$, the relationship between the energy and the initial momentum is $`H_0=2\sqrt{p_0^2+m^2}`$.
In the $`\kappa `$ expansion, $`f_e(\tau )`$ is expressed as
$`f_e(\tau )`$ $`=`$ $`{\displaystyle \frac{\sqrt{p_0^2+m^2}+ϵp_0}{m}}e^{\frac{ϵ\tau }{2m}e_1e_2}`$ (144)
$`\kappa m\left[{\displaystyle \frac{\sqrt{p_0^2+m^2}}{e_1e_2}}\left(e^{\frac{ϵ\tau }{2m}e_1e_2}1\right)(\sqrt{p_0^2+m^2}+ϵp_0){\displaystyle \frac{ϵ\tau }{4m}}e^{\frac{ϵ\tau }{2m}e_1e_2}\right]+𝒪(\kappa ^2)`$
and $`p(\tau )`$ is
$`p`$ $`=`$ $`\sqrt{p_0^2+m^2}\text{sinh}{\displaystyle \frac{e_1e_2\tau }{2m}}+p_0\text{cosh}{\displaystyle \frac{e_1e_2\tau }{2m}}`$ (145)
$`{\displaystyle \frac{\kappa m^2}{2ϵ}}\left[1+{\displaystyle \frac{(\sqrt{p_0^2+m^2}ϵp_0)^2}{m^2}}e^{\frac{ϵ\tau }{m}e_1e_2}\right]`$
$`\times \left[{\displaystyle \frac{\sqrt{p_0^2+m^2}}{e_1e_2}}\left(e^{\frac{ϵ\tau }{2m}e_1e_2}1\right)(\sqrt{p_0^2+m^2}+ϵp_0){\displaystyle \frac{ϵ\tau }{4m}}e^{\frac{ϵ\tau }{2m}e_1e_2}\right]+𝒪(\kappa ^2).`$
The trajectory $`r(\tau )>0`$ in the linear approximation becomes
$`r(\tau )`$ $`=`$ $`{\displaystyle \frac{4}{e_1e_2}}\left[\sqrt{p_0^2+m^2}\left(1\text{cosh}{\displaystyle \frac{e_1e_2\tau }{2m}}\right)p_0\text{sinh}{\displaystyle \frac{e_1e_2\tau }{2m}}\right]`$ (146)
$`{\displaystyle \frac{4\kappa ϵ}{e_1e_2}}\left[\sqrt{p_0^2+m^2}\text{sinh}{\displaystyle \frac{e_1e_2\tau }{2m}}+p_0\text{cosh}{\displaystyle \frac{e_1e_2\tau }{2m}}\right]`$
$`\times \left[{\displaystyle \frac{\sqrt{p_0^2+m^2}(\sqrt{p_0^2+m^2}ϵp_0)}{e_1e_2}}\left(1e^{\frac{ϵ\tau }{2m}e_1e_2}\right){\displaystyle \frac{ϵm\tau }{4}}\right]`$
$`{\displaystyle \frac{4\kappa }{e_1e_2}}\left[\sqrt{p_0^2+m^2}\left(1\text{cosh}{\displaystyle \frac{e_1e_2\tau }{2m}}\right)p_0\text{sinh}{\displaystyle \frac{e_1e_2\tau }{2m}}\right]`$
$`\times \{{\displaystyle \frac{1}{2e_1e_2}}[\sqrt{p_0^2+m^2}(1ϵ\text{sinh}{\displaystyle \frac{e_1e_2\tau }{2m}})ϵp_0\text{cosh}{\displaystyle \frac{e_1e_2\tau }{2m}}]^2`$
$`{\displaystyle \frac{1}{6e_1e_2}}[\sqrt{p_0^2+m^2}(1\text{cosh}{\displaystyle \frac{e_1e_2\tau }{2m}})p_0\text{sinh}{\displaystyle \frac{e_1e_2\tau }{2m}}]^2\}.`$
Furthermore in the limit $`\kappa 0`$ the solution is
$`p`$ $`=`$ $`\sqrt{p_0^2+m^2}\text{sinh}{\displaystyle \frac{e_1e_2\tau }{2m}}+p_0\text{cosh}{\displaystyle \frac{e_1e_2\tau }{2m}},`$
$`r`$ $`=`$ $`{\displaystyle \frac{4}{e_1e_2}}\left[\sqrt{p_0^2+m^2}\left(1\text{cosh}{\displaystyle \frac{e_1e_2\tau }{2m}}\right)p_0\text{sinh}{\displaystyle \frac{e_1e_2\tau }{2m}}\right].`$
We can transform it into an expression in terms of the original time coordinate $`t`$, by using the relation
$$t=\frac{2\sqrt{p_0^2+m^2}}{e_1e_2}\text{sinh}\frac{e_1e_2\tau }{2m}+\frac{2p_0}{e_1e_2}\left(\text{cosh}\frac{e_1e_2\tau }{2m}1\right)$$
(148)
which is obtained by integrating (84). Then
$`p(t)`$ $`=`$ $`p_0+{\displaystyle \frac{e_1e_2}{2}}t,`$ (149)
$`r(t)`$ $`=`$ $`{\displaystyle \frac{4}{e_1e_2}}\left\{\sqrt{(p_0+{\displaystyle \frac{e_1e_2}{2}}t)^2+m^2}\sqrt{p_0^2+m^2}\right\}.`$ (150)
This solution is identical with that derived from the Hamiltonian for flat-space electrodynamics in (1+1) dimensions: $`H=2\sqrt{p^2+m^2}e_1e_2|r|/2`$. The same solution can be applied to the system of arbitrary charges as far as $`H_02m`$.
On the other hand, for the repulsive charges $`(e_1e_2>0)`$ there exists the solution for $`H_0<2m`$, which denotes an unbounded motion and is given by
$`p(t)`$ $`=`$ $`{\displaystyle \frac{e_1e_2}{2}}t,`$ (151)
$`r(t)`$ $`=`$ $`{\displaystyle \frac{4}{e_1e_2}}\left\{\sqrt{{\displaystyle \frac{(e_1e_2)^2}{4}}t^2+m^2}{\displaystyle \frac{H_0}{2}}\right\}.`$ (152)
This solution is also derived from the tan-type A solution in the limit $`\kappa 0`$.
Another limit of the solution (146) is to take $`e_a0`$, which leads to
$$r(\tau )=\frac{2p_0}{m}\tau \kappa \frac{\sqrt{p_0^2+m^2}}{4}\tau ^2.$$
(153)
In the non-relativistic approximation (keeping the terms to the lowest order of $`p_0/m`$) we get the solution for the motion in Newtonian gravity in (1+1) dimensions:
$$r(t)=\frac{2p_0}{m}t\frac{\kappa m}{4}t^2.$$
(154)
## APPENDIX C: CAUSAL RELATIONSHIPS BETWEEN PARTICLES
One of the striking features we found in the analysis of the two body motion is that in the repulsive trajectories $`N_1,N_2,A_n`$ and $`B_n`$ the particles reach the asymptotic regime ($`r\pm \mathrm{},p=\text{finite}`$) at some finite proper time $`\tau _{\mathrm{}}`$. For example for $`N_1`$ trajectory it is
$$\tau _{\mathrm{}}=\frac{4}{\kappa m\sqrt{\gamma _m}}\text{log}\left(\frac{H(1+\sqrt{\gamma _m})\left\{p_++\sqrt{p_+^2+m^2}\right\}(\sqrt{\gamma _m}+\gamma _e)}{\left[H(1+\sqrt{\gamma _m})+\left\{p_++\sqrt{p_+^2+m^2}\right\}(\sqrt{\gamma _m}\gamma _e)\right]\eta }\right)$$
(155)
with $`p_+=\frac{1}{2\kappa }\left\{\sqrt{\kappa ^2H^2+8\mathrm{\Lambda }_e}+\sqrt{8\mathrm{\Lambda }_e+8e_1e_2}\right\}`$.
We can understand this feature in a simple flat-space model. Consider the 2-velocity
$$u^\mu =(f(\sigma \tau ),\sqrt{f^2(\sigma \tau )1})\text{with}d\tau ^2=dt^2dx^2,$$
(156)
where $`f(\sigma \tau )`$ is some function. This 2-velocity means
$$\frac{dx}{dt}=\frac{\sqrt{f^21}}{f},$$
(157)
and leads to the 2-acceleration
$$a^\mu =\frac{du^\mu }{d\tau }=f^{}\sigma (1,\frac{f}{\sqrt{f^2(\sigma \tau )1}}),$$
(158)
where $`f^{}=df(\tau )/d\tau `$. There exist functions $`f`$ such that $`f\mathrm{}`$ as $`\tau \tau _{\mathrm{}}`$, $`\tau _{\mathrm{}}`$ being finite; the particle thus becomes lightlike in a finite amount of proper time, but an infinite amount of coordinate time $`t=_{\tau _i}^\tau 𝑑\tau f(\sigma \tau )`$. An example would be $`f=\mathrm{sec}(\sigma \tau )`$. The acceleration is not constant, but increases as a function of proper time, diverging at $`\tau =\tau _{\mathrm{}}\frac{\pi }{2\sigma }`$. This is the situation we encounter for the unbounded motions of two charged particles (or of two neutral particles with sufficiently large positive cosmological constant).
To see the causal relationships between particles and the space-time structure we try to pursue the path of light signal emitted from particle 2 in the metric given in Appendix A. The path $`x(t)`$ of the light is governed by $`d\tau =0`$, which reads
$$\left(\frac{dx}{dt}\right)^2+2N_1\frac{dx}{dt}(N_0^2N_1^2)=0.$$
(159)
The equation of the light signal emitted inward (directed to particle 1) is
$$\frac{dx}{dt}=N_0(x(t),z_1(t),z_2(t),p(t))N_1(x(t),z_1(t),z_2(t),p(t)),$$
(160)
and the light emitted outward is described by
$$\frac{dx}{dt}=N_0(x(t),z_1(t),z_2(t),p(t))N_1(x(t),z_1(t),z_2(t),p(t)).$$
(161)
Let’s take first a typical bounded motion in the case of $`\mathrm{\Lambda }_e=0`$ and the attractive charges for the parameters of $`H_0=3,m=1`$ and $`q=1`$. In Fig.39 the trajectories of light signals emitted from particle 2 at various times $`T`$ are plotted. The two particles are always in causal contact, because the inward light signal from particle 2 reaches particle 1. We see one striking feature for the case of $`\mathrm{\Lambda }_e=0`$, namely that the outward light pulse has a constant velocity $`c`$, since the equations (160) and (161) become $`dx/dt=\pm 1`$, where the plus and the minus signs correspond to the light emitted outward from the particle 1 and 2, respectively.
Next we look into the effect of the cosmological constant on the path of the light signal. Fig.40 shows the trajecories of light for the case of a negative cosmological constant $`\mathrm{\Lambda }_e=1`$ and the same values of the parameters $`H_0,m`$ and $`q`$. The outward light behaves as if it is subjected to a repulsive force, in significant contrast to the particles’ trajectories on which the cosmological constant acts effectively as an attractive force as we analyzed in Sec.VI. On the other hand, while the positive cosmological constant causes a repulsive force between particles, the outward light behaves as if it undergoes an attractive force and approaches a kind of horizon given by the line $`N_0(x,z_1(t),z_2(t),p(t))\pm N_1(x,z_1(t),z_2(t),p(t))=0`$ which is shown in Fig.41 as a narrow solid line for the case of $`\mathrm{\Lambda }_e=0.5`$.
When the cosmological constant exceeds a critical value, the particles’ motion becomes unbounded and the light signals emitted from the particle 2 exhibit new characteristics as shown in Fig.42 for the case of $`\mathrm{\Lambda }_e=2.5`$. For small $`T(T<0.9)`$, the particles are in causal contact (the inward dotted curve), but for $`T0.9`$ the signal just barely catches up with particle 1, which is almost in light-like motion (the inward dashed curve). For $`T=2`$ the inward world line $`x(t)`$ is parallel to $`z_1(t)`$ at large $`t`$ and in the outward direction it goes nearly on the same trajectory with the particle 2. For large $`T`$ ($`T>2`$) the particles are out of causal contact with each other : a light ray sent from particle 2 toward particle 1 receives a strong repulsive effect and ultimately reverses direction, following behind particle 2.
Acknowledgements
This work was supported in part by the Natural Sciences and Engineering Research Council of Canada.
|
warning/0005/math0005072.html
|
ar5iv
|
text
|
# Untitled Document
$`U(𝔤)`$-finite locally analytic representations
P. Schneider, J. Teitelbaum
In this paper we continue the study of locally analytic representations of a $`p`$-adic Lie group $`G`$ in vector spaces over a spherically complete non-archimedean field $`K`$. In \[ST\], we began with an algebraic approach to this type of representation theory based on a duality functor that replaces locally analytic representations by certain topological modules over the algebra $`D(G,K)`$ of locally analytic distributions. As an application, we established the topological irreducibility of generic locally analytic principal series representations of $`\mathrm{𝐆𝐋}_\mathrm{𝟐}(\text{ }Q_p)`$ by proving the algebraic simplicity of the corresponding $`D(\mathrm{𝐆𝐋}_\mathrm{𝟐}(\text{ }Q_p),K)`$-modules.
In this paper we further exploit this algebraic point of view. We introduce a particular category of “analytic” $`D(G,K)`$-modules that lie in the image of the duality functor and therefore correspond to certain locally analytic representations. For compact groups $`G`$, these are finitely generated $`D(G,K)`$-modules that allow a (necessarily uniquely determined) Fréchet topology for which the $`D(G,K)`$-action is continuous. For more general groups, one tests analyticity by considering the action of $`D(H,K)`$ for a compact open subgroup $`H`$ in $`G`$. The category of analytic modules has the nice property that any algebraic map between such modules is automatically continuous. The concept of analytic module is dual to the concept of strongly admissible $`G`$-representation introduced in \[ST\]. The actual definition can and will be given in a way that avoids any mention of a topology on the module.
Next, we study the modules dual to the traditional smooth representations of Langlands theory. We show that a smooth representation gives rise, under duality, to an analytic module precisely when it is “strongly admissible”; this is a condition on the multiplicities with which the irreducible representations of a compact open subgroup of $`G`$ appear in the representation. In particular, if $`L`$ is a finite extension of $`\text{ }Q_p`$ and $`G`$ is the group of $`L`$-points of a connected reductive algebraic group over $`L`$, then any smooth representation of finite length is strongly admissible. This is basically a theorem of Harish-Chandra (\[HC\]) although we must use in addition results of Vigneras (\[Vig\]) to deal with some complications arising from the fact that we do not assume that our coefficient field $`K`$ is algebraically closed.
Given these foundational results, suppose that $`G`$ is the group of $`L`$-points of a split, semisimple, and simply connected group over $`L`$. We completely determine the structure of analytic modules $`M`$ that are $`U(𝔤)`$-finite, i.e., that are annihilated by a 2-sided ideal of finite codimension in the universal enveloping algebra $`U(𝔤)`$ of the Lie algebra $`𝔤`$ of $`G`$. Such a module can be decomposed into a finite sum of modules of the form $`EHom(V,K)`$ where $`E`$ is irreducible, finite dimensional, and algebraic, and $`V`$ is smooth and strongly admissible. The dual representations $`E^{}V`$ are irreducible – in fact, simple as $`K[G]`$ modules – if and only if $`V`$ is irreducible. Some of the technical hypotheses on the group $`G`$ in this section are consequences of the fact that our coefficient field is not algebraically closed.
We conclude the paper by studying the reducible members of the locally analytic principal series of $`\mathrm{𝐒𝐋}_\mathrm{𝟐}(\text{ }Q_p)`$. The corresponding modules contain a simple submodule such that the quotient is $`U(𝔤)`$-finite, and we use our methods to determine the structure of this quotient. In particular, we obtain the result that the topological length of the locally analytic principal series is at most three – a fact that is due to Morita (\[Mor\]) by a different method.
1. Analytic modules
We fix fields $`\text{ }Q_pLK`$ such that $`L/\text{ }Q_p`$ is finite and $`K`$ is spherically complete with respect to a nonarchimedean absolute value $`||`$ extending the one on $`L`$. We let $`G`$ be a $`d`$-dimensional locally $`L`$-analytic group and $`D(G,K)`$ be the corresponding $`K`$-algebra of $`K`$-valued distributions on $`G`$. Recall (\[ST\] 2.3) that $`D(G,K)`$ is an associative unital $`K`$-algebra with a natural locally convex topology in which the multiplication $``$ is separately continuous. Unless this topology is explicitly mentioned $`D(G,K)`$ is treated as an abstract algebra. In the following we want to single out a certain class of (unital left) $`D(G,K)`$-modules which seems to provide a convenient framework for the representation theory of $`G`$ over $`K`$. Let $`M`$ be a $`D(G,K)`$-module.
Definition:
A $`K`$-linear form $`\mathrm{}`$ on $`M`$ is called locally analytic if, for any $`mM`$, the linear form $`\lambda \mathrm{}(\lambda m)`$ on $`D(G,K)`$ is continuous.
Clearly the locally analytic linear forms on $`M`$ form a vector subspace $`M^{}`$ of the full $`K`$-linear dual $`M^{}`$ of $`M`$. We first consider the case of a compact group $`G`$. Recall that then $`D(G,K)`$ is a $`K`$-Fréchet algebra and as a locally convex $`K`$-vector space is reflexive (\[ST\] 1.1, 2.1, and 2.3).
Definition:
Suppose $`G`$ to be compact; a $`D(G,K)`$-module $`M`$ is called analytic if it is finitely generated and if, for any $`mM`$, there is a locally analytic linear form $`\mathrm{}`$ on $`M`$ such that $`\mathrm{}(m)0`$.
Proposition 1.1:
Suppose $`G`$ to be compact; for a finitely generated $`D(G,K)`$-module $`M`$ the following assertions are equivalent:
i. $`M`$ is analytic;
ii. $`M`$ carries a Fréchet topology with respect to which it is a continuous $`D(G,K)`$-module.
Proof: We first assume that ii. holds true. Evidently any continuous linear form on $`M`$ then is locally analytic. Hence it follows from the Hahn-Banach theorem that $`M`$ is analytic. Assume now vice versa that $`M`$ is analytic. Choose an epimorphism $`\alpha :D(G,K)^rM`$ of $`D(G,K)`$-modules for some $`r1`$. Then the linear forms $`\mathrm{}\alpha `$ for any $`\mathrm{}M^{}`$ are continuous and their simultaneous kernel coincides with the kernel of $`\alpha `$. In particular the kernel of $`\alpha `$ is closed in $`D(G,K)^r`$ so that the quotient topology via $`\alpha `$ on $`M`$ has the required properties.
By the argument in the proof of \[ST\] 3.5 the above Fréchet topology on an analytic $`D(G,K)`$-module $`M`$ is unique and therefore will be called the canonical topology of $`M`$. The continuous dual of $`M`$ is $`M^{}`$ and given the strong topology it is a vector space of compact type carrying a locally analytic $`G`$-representation (\[ST\] §§1 and 3); in particular, the canonical topology on $`M`$ is reflexive. Again by \[ST\] 3.5 any $`D(G,K)`$-linear map between two analytic $`D(G,K)`$-modules is continuous in the canonical topologies.
Question: Is any $`D(G,K)`$-module of finite presentation analytic ?
Example: As a consequence of \[ST\] 4.4 the answer is yes for the group $`G=ZZ_p`$.
The above definition of an analytic $`D(G,K)`$-module for a compact group is extended to a general group $`G`$ in the following way. Note first that for any compact open subgroup $`HG`$ the algebra $`D(H,K)`$ is a subalgebra of $`D(G,K)`$ and that
$$D(G,K)=\underset{gG/H}{}\delta _gD(H,K)$$
where $`\delta _g`$ denotes the Dirac distribution in $`gG`$.
Definition:
A $`D(G,K)`$-module $`M`$ is called analytic if it is analytic as a $`D(H,K)`$-module for any compact open subgroup $`HG`$.
Lemma 1.2:
Fix a compact open subgroup $`HG`$; a $`D(G,K)`$-module $`M`$ is analytic if it is analytic as a $`D(H,K)`$-module.
Proof: This follows easily from the fact that for any two compact open subgroups $`H`$ and $`H^{}`$ in $`G`$ the intersection $`HH^{}`$ is of finite index in $`H`$ and in $`H^{}`$.
Suppose that $`M`$ is an analytic $`D(G,K)`$-module. One easily checks that the canonical topology of $`M`$ as a $`D(H,K)`$-module is independent of the choice of the compact open subgroup $`HG`$, that the $`D(G,K)`$-action on $`M`$ is separately continuous, and that $`M^{}`$ is the continuous dual of $`M`$ and equipped with the strong topology carries a locally analytic $`G`$-representation. Of course, any $`D(G,K)`$-linear map between two analytic $`D(G,K)`$-modules is continuous in the canonical topologies.
Definition:
An analytic $`D(G,K)`$-module is called quasi-simple if it has no nonzero proper $`D(G,K)`$-submodules which are closed in the canonical topology.
An analytic $`D(G,K)`$-module $`M`$ is trivially quasi-simple if it is (algebraically) simple. But, as a consequence of polarity, it also is quasi-simple (and usually not simple) if $`M^{}`$ is a simple $`D(G,K)`$-module. For a noncompact $`G`$ we will see examples of this later on. We don’t know whether such examples also exist for compact groups.
2. Smooth $`G`$-representations
In this section we want to see how the smooth representation theory of $`G`$ fits into our new framework. We recall that a smooth $`G`$-representation $`V`$ (over $`K`$) is a $`K`$-vector space $`V`$ with a linear $`G`$-action such that the stabilizer of each vector in $`V`$ is open in $`G`$. (Traditionally one considers smooth $`G`$-representations in $`\text{ }C`$-vector spaces; but since the topology of the coefficient field plays absolutely no role in the definition this makes a difference only insofar as we do not require $`K`$ to be algebraically closed.) Moreover, a smooth $`G`$-representation $`V`$ is called admissible if, for any compact open subgroup $`HG`$, the vector subspace $`V^H`$ of $`H`$-invariant vectors in $`V`$ is finite dimensional. Finally, irreducibility of a smooth representation is always meant in the algebraic sense.
The unit element in $`G`$ has a countable fundamental system of open compact neighborhoods. This implies that the finest locally convex topology on an admissible $`G`$-representation $`V`$ is of compact type (being the countable locally convex inductive limit of the finite dimensional spaces $`V^H`$). Since the orbit maps $`\rho _v(g):=gv`$, for $`vV`$, are locally constant on $`G`$ we see that any admissible $`G`$-representation $`V`$ becomes a locally analytic $`G`$-representation on a vector space of compact type once we equip $`V`$ with the finest locally convex topology; as such we denote it by $`V^c`$.
Let $`𝔤`$ denote the Lie algebra of $`G`$ and let $`U(𝔤)`$ be the universal enveloping algebra of $`𝔤`$. The latter is naturally included in $`D(G,K)`$ (\[ST\] §2). The action of an $`𝔵𝔤`$ on a locally analytic $`G`$-representation $`W`$ is given by
$$w𝔵w:=\frac{d}{dt}\mathrm{exp}(t𝔵)w|_{t=0}$$
$`(1)`$
where $`\mathrm{exp}:𝔤>G`$ denotes the exponential map defined locally around 0 (\[ST\] 3.2). In addition Taylor’s formula says that, for each fixed $`wW`$ there is a sufficiently small neighborhood $`U`$ of 0 in $`𝔤`$ such that, for $`𝔵U`$, we have a convergent expansion
$$\mathrm{exp}(𝔵)w=\underset{n=0}{\overset{\mathrm{}}{}}\frac{1}{n!}𝔵^nw.$$
$`(2)`$
The formulas $`(1)`$ and $`(2)`$ together imply that the orbit maps $`\rho _w`$, for $`wW`$, are locally constant if and only if the $`𝔤`$-action on $`W`$ is trivial or equivalently if and only if the closed 2-sided ideal $`I(𝔤)`$ in $`D(G,K)`$ generated by $`𝔤`$ annihilates $`W`$.
What can we say about the quotient algebra $`D^{\mathrm{}}(G,K):=D(G,K)/I(𝔤)`$ ?
$`D(G,K)`$ is the strong dual of the space $`C^{an}(G,K)`$ of $`K`$-valued locally analytic functions on $`G`$. Since the Dirac distributions generate a dense subspace in $`D(G,K)`$ (\[ST\] 3.1) the ideal $`I(𝔤)`$ is the orthogonal of the closed subspace in $`C^{an}(G,K)`$ which is the simultaneous kernel of all linear forms $`\delta _g𝔵\delta _h`$ with $`𝔵𝔤`$ and $`g,hG`$. This is precisely the subspace of those functions in $`C^{an}(G,K)`$ which are annihilated by the action of $`𝔤`$. And this in turn, by Taylor’s formula, is the subspace $`C^{\mathrm{}}(G,K)`$ of all $`K`$-valued locally constant functions on $`G`$ with the subspace topology. On the other hand as a direct product of spaces of compact type the space $`C^{an}(G,K)`$ is reflexive. In this situation the strong dual of a closed subspace is the quotient of the strong dual by the orthogonal subspace (\[B-TVS\] IV.16 Cor.). In other words we have
$$D^{\mathrm{}}(G,K)=C^{\mathrm{}}(G,K)_b^{}.$$
Moreover, if $`HG`$ is a fixed compact open subgroup, then
$$C^{\mathrm{}}(G,K)=\underset{gG/H}{}C^{\mathrm{}}(gH,K)$$
is the direct product of the spaces $`C^{\mathrm{}}(gH,K)`$ each of which is a locally convex inductive limit of finite dimensional spaces and hence carries the finest locally convex topology (compare \[ST\] 1.2.i). In particular $`C^{\mathrm{}}(H,K)`$ is the inductive limit
$$C^{\mathrm{}}(H,K)=\underset{\underset{N}{}}{lim}K[H/N]$$
of the algebraic group rings $`K[H/N]`$ with $`N`$ running through the open normal subgroups of $`H`$.
All of this applies to $`V^c`$ for any admissible $`G`$-representation $`V`$. In parti-cular, $`V^c`$ as well as its strong dual $`(V^c)_b^{}`$ are $`D^{\mathrm{}}(G,K)`$-modules. Clearly $`M:=(V^c)_b^{}`$ is an analytic $`D(G,K)`$-module if and only if $`M`$ is finitely generated as a $`D^{\mathrm{}}(H,K)`$-module for some fixed but arbitrary choice of a compact open subgroup $`HG`$. This condition can be expressed purely in terms of multiplicities as follows. Let $`\widehat{H}`$ denote the set of isomorphism classes of all irreducible smooth $`H`$-representations. Recall that any $`\pi \widehat{H}`$ is finite dimensional. We let
$$\mu (\pi ):=\text{multiplicity of}\pi \text{in}C^{\mathrm{}}(H,K)$$
so that we have
$$C^{\mathrm{}}(H,K)\underset{\pi \widehat{H}}{}\mu (\pi )\pi $$
and
$$D^{\mathrm{}}(H,K)\underset{\pi \widehat{H}}{}(\pi ^{})^{\times \mu (\pi )}$$
$`(3)`$
where $`\pi ^{}`$ denotes the contragredient of $`\pi `$. Any smooth $`G`$-representation $`V`$ is semisimple as an $`H`$-representation. Moreover $`V`$ is admissible if and only if the multiplicities
$$\mu (\pi ,V):=\text{multiplicity of}\pi \text{in}V$$
for any $`\pi \widehat{H}`$ are finite. We then have
$$V\underset{\pi \widehat{H}}{}\mu (\pi ,V)\pi $$
and
$$(V^c)_b^{}\underset{\pi \widehat{H}}{}(\pi ^{})^{\times \mu (\pi ,V)}$$
$`(4)`$
as $`D^{\mathrm{}}(H,K)`$-modules.
Definition:
A smooth $`G`$-representation is called strongly admissible if there is a natural number $`m`$ such that
$$\mu (\pi ,V)m\mu (\pi )$$
for any $`\pi \widehat{H}`$.
That the above definition does not depend on the particular choice of $`H`$ can be seen as follows. Let $`H_0H`$ be a pair of compact open subgroups in $`G`$. For any $`\pi \widehat{H}`$ and $`\sigma \widehat{H_0}`$ let $`\mu (\pi :\sigma )`$ denote the multiplicity of $`\sigma `$ in $`\pi |H_0`$. One easily checks that
$$[H:H_0]\mu (\sigma )=\underset{\pi \widehat{H}}{}\mu (\pi :\sigma )\mu (\pi )\text{and}\mu (\pi )=\underset{\sigma \widehat{H_0}}{}\mu (\pi :\sigma )\mu (\sigma ).$$
Assuming that $`\mu (\pi ,V)m\mu (\pi )`$, resp. $`\mu (\sigma ,V)n\mu (\sigma )`$, we compute
$$\begin{array}{ccc}\mu (\sigma ,V)& =& \underset{\pi \widehat{H}}{}\mu (\pi :\sigma )\mu (\pi ,V)\\ \\ & & m\underset{\pi \widehat{H}}{}\mu (\pi :\sigma )\mu (\pi )\\ \\ & =& m[H:H_0]\mu (\sigma ),\end{array}$$
resp.
$$\begin{array}{ccc}\mu (\pi ,V)& & \underset{\sigma \widehat{H_0}}{}\mu (\pi :\sigma )\mu (\sigma ,V)\\ \\ & & n\underset{\sigma \widehat{H_0}}{}\mu (\pi :\sigma )\mu (\sigma )\\ \\ & =& n\mu (\pi ).\end{array}$$
Proposition 2.1:
The functor $`V(V^c)_b^{}`$ is an (anti)equivalence of categories between the category of all strongly admissible $`G`$-representations and the category of all analytic $`D(G,K)`$-modules which are annihilated by $`I(𝔤)`$.
Proof: Comparing $`(3)`$ and $`(4)`$ it is obvious that $`(V^c)_b^{}`$ is finitely generated as a $`D^{\mathrm{}}(H,K)`$-module if and only if $`V`$ is strongly admissible. Hence the functor in question is well defined and fully faithful. Moreover, if $`M`$ is an analytic $`D(G,K)`$-module annihilated by $`I(𝔤)`$ then we have a topological surjection $`D^{\mathrm{}}(H,K)^rM`$ for some $`r1`$. The dual embedding $`M^{}C^{\mathrm{}}(H,K)^r`$ shows that $`V:=M_b^{}`$ carries the finest locally convex topology and therefore is a strongly admissible $`G`$-representation. By reflexivity we have $`M=(V^c)_b^{}`$ so that $`M`$ lies in the image of our functor.
Proposition 2.2:
If $`G`$ is the group of $`L`$-rational points of a connected reductive $`L`$-group $`𝐆`$ then any smooth $`G`$-representation of finite length is strongly admissible.
Proof: Let $`C`$ be a fixed algebraically closed field which contains $`K`$. We first want to reduce the assertion to the case where the coefficient field of the smooth representation is $`C`$. Denoting by $`(.)_C`$ the base extension functor from $`K`$ to $`C`$ we have
$$V_C\underset{\pi \widehat{H}}{}\mu (\pi ,V)\pi _C.$$
Let $`\mathrm{Irr}_C(H)`$ denote the set of isomorphism classes of all irreducible smooth $`H`$-representations over $`C`$. For each $`\sigma \mathrm{Irr}_C(H)`$ there is a unique $`\pi (\sigma )\widehat{H}`$ such that $`\sigma `$ occurs in $`\pi (\sigma )_C`$. The theory of the Schur index tells us the following (\[CR\] (70:15)):
1) The Schur index $`m_K(\sigma )`$ of $`\sigma \mathrm{Irr}_C(H)`$ with respect to $`K`$ only depends on $`\pi (\sigma )`$; we therefore put $`m_K(\pi ):=m_K(\sigma )`$ if $`\pi =\pi (\sigma )`$.
2) For any $`\pi \widehat{H}`$ we have the decomposition
$$\pi _Cm_K(\pi )\underset{\pi (\sigma )=\pi }{}\sigma .$$
3) If $`\pi =\pi (\sigma )`$ then $`\mu (\pi )m_K(\pi )=\mathrm{dim}_C\sigma `$.
By using 2) our above decomposition of $`V_C`$ becomes
$$V_C\underset{\sigma \mathrm{Irr}_C(H)}{}\mu (\pi (\sigma ),V)m_K(\pi (\sigma ))\sigma .$$
If we therefore show that there is an $`mIN`$ such that $`\mu (\sigma ,V_C)=\mu (\pi (\sigma ),V)m_K(\pi (\sigma ))m\mathrm{dim}_C\sigma `$ for any $`\sigma \mathrm{Irr}_C(H)`$ then it follows from 3) that $`\mu (\pi ,V)m\mu (\pi )`$ for any $`\pi \widehat{H}`$. According to \[Vig\] II.4.3.c with $`V`$ also $`V_C`$ is of finite length. This reduces us to proving our assertion for smooth $`G`$-representation over some algebraically closed field $`C`$ containing the field of complex numbers $`\text{ }C`$. We first look at the case when $`V`$ is irreducible supercuspidal. By a character twist we may assume that the central character of $`V`$ is of finite order. According to \[Vig\] II.4.9 the representation $`V`$ then is the base extension to $`C`$ of an irreducible supercuspidal $`G`$-representation over $`\text{ }C`$. For the latter our assertion is a theorem of Harish-Chandra (\[HC\] Cor. of Thm. 2), and it is obvious that base extension between two algebraically closed fields respects our assertion. Since a general irreducible $`V`$ is contained in a representation parabolically induced from a supercuspidal representation it remains to show that parabolic induction respects strong admissibility. Let $`P=P_LP_u`$ be a parabolic subgroup of $`G`$ with unipotent radical $`P_u`$ and Levi factor $`P_L`$ and let $`W`$ be a strongly admissible smooth representation of $`P_L`$. We have to check that $`V:=\mathrm{Ind}_P^G(W)`$ is again strongly admissible. Since $`V`$ is known to be admissible (\[Vig\] II.2.1) we can do this by proving that the full linear dual $`V^{}`$ of $`V`$ is finitely generated as an $`D^{\mathrm{}}(H,K)`$-module. Moreover, being completely free in the choice of the compact open subgroup $`H`$ of $`G`$ we may choose it in such a way that the Iwasawa decomposition $`G=HP`$ holds. Put $`H_P:=HP`$ and let $`H_L`$ denote the image of $`H_P`$ in $`P_L`$. As an $`H`$-representation we then have
$$\mathrm{Ind}_P^G(W)=\mathrm{Ind}_{H_P}^H(W|H_L).$$
By assumption $`(W|H_L)^{}`$ is a finitely generated $`D^{\mathrm{}}(H_L,K)`$-module. All we have to see therefore is that
$$\mathrm{Ind}_{H_P}^H(W|H_L)^{}=D^{\mathrm{}}(H,K)\underset{D^{\mathrm{}}(H_P,K)}{}(W|H_L)^{}$$
holds true. By semisimplicity this is an easy consequence of the analogous identity with $`W|H_L`$ replaced by $`C^{\mathrm{}}(H_P,K)`$.
As a consequence of these results we obtain that the functor $`V(V^c)_b^{}`$ induces a bijective correspondence between irreducible smooth $`G`$-representations and quasi-simple analytic $`D(G,K)`$-modules which are annihilated by $`I(𝔤)`$. It should be pointed out that $`(V^c)_b^{}`$ as a vector space is the full linear dual of $`V`$. The smooth linear forms form a in general proper $`D^{\mathrm{}}(G,K)`$-submodule of $`(V^c)_b^{}`$ so that the latter cannot be simple.
3. $`U(𝔤)`$-finite modules
In this section we let $`G`$ be the group of $`L`$-rational points of a connected reductive split $`L`$-group $`𝐆`$. We want to understand more generally those analytic $`D(G,K)`$-modules $`M`$ on which $`U(𝔤)`$ acts through a finite dimensional quotient. They will be called $`U(𝔤)`$-finite.
Let $`E`$ be the underlying $`L`$-vector space of an irreducible $`L`$-rational algebraic representation of $`𝐆`$. For any $`U(𝔤)`$-finite analytic $`D(G,K)`$-module $`M`$ we set
$$M^E:=Hom_{U(𝔤)}(E,M).$$
$`Hom_L(E,M)`$ and hence $`M^E`$ as a closed vector subspace both inherit a natural Fréchet topology from $`M`$. The group $`G`$ acts on $`M^E`$ via the continuous $`K`$-linear endomorphisms
$$f^gf(x):=g(f(g^1x))\text{for}gG\text{and}fM^E.$$
Moreover,
$$\begin{array}{ccc}E\times M^E& & M\\ (x,f)& & f(x)\end{array}$$
is a continuous $`G`$-equivariant bilinear map.
Let $`V:=M_b^{}`$ denote the strong dual of $`M`$ as a locally analytic $`G`$-representation. In order to determine the topology on $`V`$ we need the following result.
Proposition 3.1:
Let $`JU(𝔤)`$ be a 2-sided ideal of finite codimension and let $`HG`$ be a compact open subgroup; then the subspace topology on the subspace $`C^{an}(H,K)^{J=0}`$ of all vectors in $`C^{an}(H,K)`$ annihilated by $`J`$ is the finest locally convex topology.
Proof: Fix an ordered vector space basis for $`𝔤`$, and an exponential map for $`G`$. This data, together with a choice of disk of sufficiently small radius $`s`$ around the origin in $`L^{\mathrm{dim}𝔤}`$, determines a “canonical chart of the second kind” on $`H`$. Let $`H_r`$ be the family of standard compact open subgroups of $`H`$ obtained from this canonical chart (see \[Fea\] 4.3.3). The Banach space of analytic functions on $`H_r`$ is the standard Banach space $`_{0,r}(K)`$ of convergent series with coefficients in $`K`$ on the disk of radius $`r`$ for $`0<rs`$. Let
$$_r:=\underset{hH_r\backslash H}{}_{0,r}(K).$$
Following the proof of \[Fea\] 3.3.4 we see that this Banach space is an analytic $`H_s`$-representation and
$$\underset{}{lim}_r\stackrel{}{}C^{an}(H,K).$$
By \[Fea\] 4.7.3, there is a non-degenerate pairing between $`U(𝔤)`$ and the factor $`_{0,r}`$ of the product defining $`_r`$ corresponding to the trivial coset $`H_r`$. This pairing is given by evaluation at the identity element:
$$\begin{array}{ccc}U(𝔤)\times _{0,r}& & K\\ (𝔷,f)& & (𝔷f)(1).\end{array}$$
The ideal $`J`$ is of finite codimension in $`U(𝔤)`$, and given the non-degeneracy of the pairing it follows that the space $`_{0,r}^{J=0}`$ is finite dimensional. Furthermore, because the $`U(𝔤)`$-action from the left commutes with the right translation action of $`H`$ it follows immediately that $`_r^{J=0}`$ is finite dimensional. Then $`C^{an}(H,K)^{J=0}`$, being the direct limit of these finite dimensional spaces (\[ST\] 1.2.i), has the finest locally convex topology.
Since $`M`$ is analytic we have a surjection $`D(H,K)^mM`$ of $`D(H,K)`$-modules for some $`mIN`$ and some (or any) compact open subgroup $`HG`$. After dualizing we obtain an injection $`VC^{an}(H,K)^m`$ which certainly is $`U(𝔤)`$-linear (\[ST\] 3.2). Moreover, by assumption there is a 2-sided ideal $`JU(𝔤)`$ of finite codimension which annihilates $`M`$ and hence $`V`$. Hence we actually have an injection $`V(C^{an}(H,K)^{J=0})^m`$. Applying Prop. 3.1 we now see that the topology on $`V`$ necessarily is the finest locally convex one.
For general reasons $`E_LV`$ with $`G`$ acting diagonally also is a locally analytic $`G`$-representation on a $`K`$-vector space of compact type (\[Fea\] 2.4.3 and \[ST\] 1.2.ii). For our particular $`V`$ the topology on $`E_LV`$ is, according to the above discussion, the finest locally convex one. We let $`E_{U(𝔤)}V`$ denote the $`G`$-equivariant quotient of $`E_LV`$ by the (automatically closed) $`K`$-vector subspace generated by all vectors of the form $`𝔵xv+x𝔵v`$ for $`𝔵𝔤`$, $`xE`$, and $`vV`$. By \[ST\] 1.2.i this quotient is a locally analytic $`G`$-representation on a $`K`$-vector space of compact type whose topology is the finest locally convex one and whose strong dual evidently is $`M^E`$. In particular, both $`E_{U(𝔤)}V`$ and $`M^E`$ are separately continuous $`D(G,K)`$-modules.
By continuity and \[ST\] 3.1 the above bilinear map $`E\times M^EM`$ induces a continuous $`D(G,K)`$-module homomorphism
$$E_LM^EM.$$
By construction the $`𝔤`$-action on $`E_{U(𝔤)}V`$ derived from the $`G`$-action is trivial. Hence $`I(𝔤)`$ annihilates $`E_{U(𝔤)}V`$ and by duality also $`M^E`$. Provided that $`M^E`$ is finitely generated as a $`D(H,K)`$-module for some compact open subgroup $`HG`$ it follows from Prop. 2.1 that $`M^E`$ is the dual of the strongly admissible $`G`$-representation $`E_{U(𝔤)}V`$.
Let $`\widehat{𝐆}`$ denote the set of isomorphism classes of all irreducible $`L`$-rational algebraic representations of $`𝐆`$. We have the continuous $`D(G,K)`$-module module homomorphism
$$\underset{E\widehat{𝐆}}{}E_LM^EM.$$
The direct sum on the left hand side in fact is finite since the number of $`E\widehat{𝐆}`$ which are annihilated by a given 2-sided ideal of finite codimension in $`U(𝔤)`$ is finite.
Proposition 3.2:
Assume that $`𝐆`$ is split semisimple and simply connected; for any $`U(𝔤)`$-finite analytic $`D(G,K)`$-module $`M`$ the natural map
$$\underset{E\widehat{𝐆}}{}E_LM^E\stackrel{}{}M$$
is an isomorphism of $`D(G,K)`$-modules, each $`M^E`$ is the linear dual of a strongly admissible $`G`$-representation over $`K`$, and $`M^E=0`$ for all but finitely many $`E\widehat{𝐆}`$.
Proof: We have already noted that the map in question is a homomorphism of $`D(G,K)`$-modules and that the direct sum on the left hand side is finite. To establish the bijectivity we set $`𝔤_K:=𝔤_LK`$ and we let $`\widehat{𝔤}`$, resp. $`\widehat{𝔤_K}`$, denote the set of isomorphism classes of all finite dimensional simple $`𝔤`$-modules, resp. $`𝔤_K`$-modules. By assumption $`M`$ is an $`U(𝔤_K)/J`$-module for some 2-sided ideal $`JU(𝔤_K)`$ of finite codimension. Since $`𝔤`$ is semisimple the algebra $`U(𝔤_K)/J`$ is semisimple (\[Dix\] 1.6.4). Hence $`M`$ is a semisimple $`𝔤_K`$-module and we have its isotypic decomposition
$$M=\underset{E\widehat{𝔤_K}}{}M_E$$
(compare \[Dix\] 1.2.8). Moreover, since $`\mathrm{End}_{U(𝔤_K)}(E)=K`$ (\[Dix\] 2.6.5 and 7.2.2(i)) we have the natural isomorphism
$$E_KHom_{U(𝔤_K)}(E,M)\stackrel{}{}M_E$$
for any $`E\widehat{𝔤_K}`$. Since the functor $`EE_LK`$ induces a bijection $`\widehat{𝔤}\stackrel{}{}\widehat{𝔤_K}`$ (both sides are classified by the dominant weights) the above isotypic decomposition can be rewritten as a bijection
$$\underset{E\widehat{𝔤}}{}E_LHom_{U(𝔤)}(E,M)\stackrel{}{}M.$$
But since $`𝐆`$ is assumed to be simply connected we have, by derivation, the natural bijection $`\widehat{𝐆}\stackrel{}{}\widehat{𝔤}`$ so that the last bijection coincides with the isomorphism in the assertion.
With $`M`$ also its direct summand $`E_LM^E`$ is finitely generated as a $`D(H,K)`$-module for any compact open subgroup $`HG`$. It follows that $`M^E`$ is a finitely generated $`D(H,K)`$-module as well: Take finitely many tensors which generate $`E_LM^E`$; their $`M^E`$-components form a generating set for $`M^E`$. We have explained above that then $`M^E`$ is the linear dual of a strongly admissible $`G`$-representation.
Example: The assumptions on the group $`𝐆`$ in the above Proposition cannot be weakened as the following example shows. Let $`L=K:=\text{ }Q_2`$, $`𝐆:=\mathrm{𝐏𝐆𝐋}_\mathrm{𝟑}`$, and $`𝐆_𝐨:=\mathrm{𝐒𝐋}_\mathrm{𝟑}`$. Then $`G_\mathrm{o}:=\mathrm{𝐒𝐋}_\mathrm{𝟑}(\text{ }Q_2)`$ is an open normal subgroup of index three in $`G=\mathrm{𝐏𝐆𝐋}_\mathrm{𝟑}(\text{ }Q_2)`$. Let $`E_\mathrm{o}`$ denote the three dimensional standard representation of $`G_\mathrm{o}`$ and let $`M:=\mathrm{Ind}_{G_\mathrm{o}}^G(E_\mathrm{o})`$ be the induced $`G`$-representation (in the sense of abstract groups). It is clear that $`M`$ is an $`U(𝔤)`$-finite analytic $`D(G,K)`$-module. One checks that as a $`G_\mathrm{o}`$-representation $`M`$ is isomorphic to $`E_\mathrm{o}E_\mathrm{o}E_\mathrm{o}`$. Since $`\widehat{𝐆}`$ is a subset of $`\widehat{𝐆_𝐨}=\widehat{𝔤}`$ to which $`E_\mathrm{o}`$ does not belong we see that $`Hom_{U(𝔤)}(E,M)=0`$ for any $`E\widehat{𝐆}`$.
For the sake of completeness we remark that vice versa any finite direct sum $`E_1_LHom_K(V_1,K)\mathrm{}E_r_LHom_K(V_r,K)`$ with $`E_i\widehat{𝐆}`$ and strongly admissible smooth $`G`$-representations $`V_i`$ over $`K`$ is an $`U(𝔤)`$-finite analytic $`D(G,K)`$-module. Apart from the finite generation which is contained in the subsequent lemma this is clear.
Lemma 3.3:
Let $`HG`$ be a compact open subgroup; for any finitely generated $`D^{\mathrm{}}(H,K)`$-module $`N`$ and any $`E\widehat{𝐆}`$ the $`D(H,K)`$-module $`E_LN`$ is finitely generated.
Proof: We begin with a general observation. Let $`\mathrm{𝒪}(𝐆)`$ denote the space of $`L`$-rational functions on $`G`$. Then the map
$$\begin{array}{ccc}\mathrm{𝒪}(𝐆)\underset{L}{}C^{\mathrm{}}(H,K)& & C^{an}(H,K)\\ (\psi ,f)& & \psi |Hf\end{array}$$
is injective. This can be seen as follows. Let $`_{j=1}^m\psi _jf_j`$ be an element in the left hand side such that $`_j\psi _j|Hf_j=0`$. We may assume that $`\psi _1,\mathrm{},\psi _m`$ are linearly independent. Choose a disjoint covering $`H=\dot{}_{i=1}^nU_i`$ by nonempty open subsets $`U_iH`$ such that the restrictions $`f_j|U_i`$, for any $`1jm`$ and $`1in`$, are constant. Since each $`U_i`$ is Zariski dense in $`𝐆`$ (this can be deduced, e.g., from \[DG\] II.5.4.3 and II.6.2.1) it follows that $`\psi _1|U_i,\mathrm{},\psi _m|U_i`$ viewed in $`C^{an}(U_i,K)`$ still are linearly independent. Hence $`f_j|U_i=0`$ for any $`i`$ and $`j`$ and therefore $`f_j=0`$ for any $`j`$.
Coming back to our assertion it suffices, of course, to consider the case $`N=D^{\mathrm{}}(H,K)`$. On the other hand, if $`JU(𝔤)`$ denotes the annihilator ideal of $`E^{}`$ then we find some $`G`$-equivariant embedding $`E^{}\mathrm{𝒪}(𝐆)^{J=0}`$. Combining this with the above map leads, using the Leibniz rule, to an $`H`$-equivariant embedding $`E^{}_LC^{\mathrm{}}(H,K)C^{an}(H,K)^{J=0}C^{an}(H,K)`$. As a consequence of Prop. 3.1 the topology induced by $`C^{an}(H,K)`$ on the left hand side is the finest locally convex topology. By dualizing we therefore obtain a surjection $`D(H,K)E_LD^{\mathrm{}}(H,K)`$ of $`D(H,K)`$-modules.
We finally study the question when an $`U(𝔤)`$-finite analytic $`D(G,K)`$-module is quasi-simple.
Proposition 3.4:
If $`E\widehat{𝐆}`$ and $`V`$ is an irreducible smooth $`G`$-representation over $`K`$ then $`E\underset{L}{}V`$ with the diagonal $`G`$-action is a simple module over the group ring $`K[G]`$.
Proof: We show that each nonzero element $`xE_LV`$ generates $`E_LV`$ as a $`K[G]`$-module. But first we recall a few facts from rational representation theory (compare \[Jan\] II §§1 and 2). Fix a Borel subgroup $`PG`$ and a maximal split torus $`TP`$, and let $`N`$ denote the unipotent radical of $`P`$.
1. The subspace $`E^N`$ of $`N`$-invariants in $`E`$ is one dimensional and coincides with the weight space $`E_\lambda `$ where $`\lambda `$ is the highest weight of $`E`$ (w.r.t. $`T`$ and $`B`$).
2. If $`eE_\mu `$ has weight $`\mu `$ then $`Nee+_{\mu <\nu }E_\nu `$.
The fact 1. holds true similarly on the level of Lie algebras. This shows that whenever $`U_\mathrm{o}N`$ is an open subgroup then
1’. $`E^{U_\mathrm{o}}=E^N=E_\lambda `$ is one dimensional.
Since $`E`$ is also an irreducible module for the induced action of the Lie algebra of $`G`$ it follows that whenever $`UG`$ is an open subgroup we have
3. $`L[U]e=E`$ for any nonzero $`eE`$.
Consider now a fixed nonzero element
$$x=e_1v_1+\mathrm{}+e_rv_r$$
with $`0e_iE`$ and $`0v_iV`$. We may assume that each $`e_i`$ is a weight vector. In order to show that $`K[G]x=E_LV`$ we may replace $`x`$ when convenient by any other nonzero element in $`K[G]x`$. In a first step we will show that for this reason we may assume in fact that $`r=1`$.
By the smoothness of $`V`$ we find an open subgroup $`UG`$ which fixes each of the vectors $`v_1,\mathrm{},v_r`$. Put $`U_\mathrm{o}:=UN`$. If $`U_\mathrm{o}`$ fixes $`x`$ we are immediately reduced to the case $`r=1`$ since, by 1’., we then have $`x(E_LV^U)^{U_\mathrm{o}}=E^{U_\mathrm{o}}_LV^U=E_\lambda _LV^U`$. Otherwise there is a $`gU_\mathrm{o}`$ such that $`gxx0`$ and we replace $`x`$ by
$$gxx=(ge_1e_1)v_1+\mathrm{}+(ge_re_r)v_r.$$
The point to note is that, by 2., each $`ge_ie_i`$ lies in a sum of weight spaces where the occuring weights are strictly bigger than the weight of $`e_i`$. This means that one way or another after finitely many steps we have replaced $`x`$ by a nonzero element in $`E_\lambda _LV^U`$ for which $`r`$ can be assumed to be one.
Let therefore, for the second step of the proof, $`xE_LV`$ be an element of the form $`x=ev`$ with $`0eE`$ and $`0vV`$. Denote by $`UG`$ the stabilizer of $`v`$. Using 3. and the irreducibility of $`V`$ we obtain
$$K[G]x=K[G]((L[U]e)v)=K[G](Ev)=EK[G]v=EV.$$
Corollary 3.5:
Assume that $`𝐆`$ is split semisimple and simply connected and let $`M`$ be any $`U(𝔤)`$-finite analytic $`D(G,K)`$-module; then $`M`$ is quasi-simple if and only if it is of the form $`ME_LHom_K(V,K)`$ for some $`E\widehat{𝐆}`$ and some irreducible smooth $`G`$-representation $`V`$ over $`K`$.
Proof: If $`E\widehat{𝐆}`$ and $`V`$ is irreducible smooth then $`E^{}_LV`$ is a simple $`D(G,K)`$-module by Prop. 3.4. Hence $`E_LHom_K(V,K)=(E^{}_LV)^{}`$ is quasi-simple.
If on the other hand $`M`$ is quasi-simple then there is, by Prop. 3.2, an $`E\widehat{𝐆}`$ and a strongly admissible $`G`$-representation $`V`$ such that $`M=E_LHom_K(V,K)`$. With $`M`$ also $`Hom_K(V,K)`$ is quasi-simple. Hence $`V`$ is irreducible.
The results of this section have more or less obvious counterparts for $`G`$ being a compact open subgroup in $`𝐆(L)`$. We leave precise formulations to the reader.
4. An example
In this last section we will analyze the reducible members of the locally analytic principal series of the group $`\mathrm{𝐒𝐋}_\mathrm{𝟐}(\text{ }Q_p)`$ and we will show that they contain tensor product representations of the kind considered in the last section.
Throughout this section let $`G:=\mathrm{𝐒𝐋}_\mathrm{𝟐}(\text{ }Q_p)`$. Furthermore, let $`P`$ denote the Borel subgroup of lower triangular matrices in $`G`$ and $`T`$ the subgroup of diagonal matrices. We actually will view $`T`$ as a quotient of $`P`$. Assuming that $`K`$ is contained in the completion of an algebraic closure of $`\text{ }Q_p`$ we fix a $`K`$-valued locally analytic character
$$\chi :TK^\times .$$
The corresponding principal series representation is
$$\mathrm{Ind}_P^G(\chi ):=\{fC^{an}(G,K):f(gp)=\chi (p^1)f(g)\text{for any}gG,pP\}$$
with $`G`$ acting by left translation. This is a locally analytic $`G`$-representation on a vector space of compact type and its strong dual
$$M_\chi :=\mathrm{Ind}_P^G(\chi )_b^{}$$
is a $`D(G,K)`$-module which is finitely generated, e.g., as a $`D(B,K)`$-module where $`B`$ is the Iwahori subgroup of $`G`$ (\[ST\] §§5 and 6). By Prop. 1.1 the $`D(G,K)`$-module $`M_\chi `$ therefore is analytic.
The basic numerical invariant of the character $`\chi `$ which governs the irreducibility properties of $`\mathrm{Ind}_P^G(\chi )`$ is the number $`c(\chi )K`$ defined by the expansion
$$\chi (\left(\begin{array}{cc}t^1& 0\\ 0& t\end{array}\right))=\mathrm{exp}(c(\chi )\mathrm{log}(t))$$
for $`t`$ sufficiently close to $`1`$ in $`ZZ_p`$. It is shown in \[ST\] 6.1 that $`M_\chi `$ is a simple $`D(G,K)`$-module if $`c(\chi )IN_0`$. We therefore assume for the rest of this section that $`m:=c(\chi )IN_0`$. According to \[ST\] 6.2 we then have a nonzero homomorphism of $`D(G,K)`$-modules $`M_\chi ^{}M_\chi `$ where $`\chi ^{}:=ϵ^{m+1}\chi `$ and $`ϵ(\left(\begin{array}{cc}t^1& 0\\ 0& t\end{array}\right)):=t^2`$ is the positive root of $`G`$ with respect to $`P`$. Since $`c(\chi ^{})=m+2`$ the module $`M_\chi ^{}`$ is simple and the above map consequently is injective. It therefore remains to discuss the quotient module
$$M_\chi ^{loc}:=M_\chi /M_\chi ^{}$$
which, of course, is finitely generated. On the other hand, the above map is exhibited in the proof of \[ST\] 6.2 as the dual $`I^{}`$ of a $`G`$-equivariant continuous linear map
$$I:\mathrm{Ind}_P^G(\chi )\mathrm{Ind}_P^G(\chi ^{})$$
whose actual construction we will recall further below. By the argument in \[ST\] 3.5 the kernel of $`I`$ again is a locally analytic $`G`$-representation on a vector space of compact type. We will see that $`I`$ is a quotient map or equivalently that the image $`I^{}(M_\chi ^{})`$ is closed in $`M_\chi `$. The module $`M_\chi ^{loc}`$ therefore is the continuous dual of the kernel of $`I`$ and in particular is analytic.
Write $`\chi =\chi _{alg}\chi _{lc}`$ where $`\chi _{alg}(\left(\begin{array}{cc}t^1& 0\\ 0& t\end{array}\right)):=t^m`$ is a $`\text{ }Q_p`$-rational character and $`\chi _{lc}`$ is a $`K`$-valued locally constant character of $`T`$. The character $`\chi _{alg}`$ is dominant for the Borel subgroup $`P^{}`$ opposite to $`P`$; hence the algebraic induction $`\mathrm{ind}_P^G(\chi _{alg})`$ is the irreducible $`\text{ }Q_p`$-rational representation of highest weight $`\chi _{alg}`$ (w.r.t. $`P^{}`$) of $`G`$ (compare \[Jan\] II.2 and II.8.23). On the other hand, since the character $`\chi _{lc}`$ is locally constant we may form the smooth induced $`G`$-representation
$$\mathrm{Ind}_{P,\mathrm{}}^G(\chi _{lc}):=\{fC^{\mathrm{}}(G,K):f(gp)=\chi _{lc}(p^1)f(g)\text{for any}gG,pP\}$$
over $`K`$ with $`G`$ acting by left translation. It is known (\[Vig\] II.5.13) to be a smooth $`G`$-representation of finite length which, by Prop. 2.2, implies that it is strongly admissible. There is the obvious $`G`$-equivariant linear map
$$\begin{array}{cccc}\tau :& \mathrm{ind}_P^G(\chi _{alg})\underset{Q_p}{}\mathrm{Ind}_{P,\mathrm{}}^G(\chi _{lc})& & \mathrm{Ind}_P^G(\chi )\\ & (\psi ,f)& & \psi f.\end{array}$$
We claim that
$$0\mathrm{ind}_P^G(\chi _{alg})\underset{\text{ }Q_p}{}\mathrm{Ind}_{P,\mathrm{}}^G(\chi _{lc})\stackrel{\tau }{}\mathrm{Ind}_P^G(\chi )\stackrel{I}{}\mathrm{Ind}_P^G(\chi ^{})0$$
$`()`$
is an exact sequence of locally convex $`K`$-vector spaces (where the left hand term carries the finest locally convex topology). This means that it is exact as a sequence of vector spaces and that the maps involved are strict. By dualizing and observing that the rational representations of $`G`$ are selfdual this leads to the following result.
Proposition 4.1:
If $`c(\chi )IN_0`$ then the $`D(G,K)`$-module $`M_\chi ^{loc}`$ is analytic and $`U(𝔤)`$-finite and is isomorphic to the tensor product of the $`\text{ }Q_p`$-rational $`G`$-representation $`\mathrm{ind}_P^G(\chi _{alg})`$ and the full $`K`$-linear dual of the smooth representation $`\mathrm{Ind}_{P,\mathrm{}}^G(\chi _{lc})`$ of finite length.
We begin by recalling the construction of $`I`$ from \[ST\] 6.2. The group $`G`$ acts on $`C^{an}(G,K)`$ by left and by right translations. Both actions derive into an action of the Lie algebra $`𝔤=𝔰𝔩_2(\text{ }Q_p)`$. Whereas the actions coming from left translation are denoted, as usual, by $`fgf`$ for $`gG`$ and $`f𝔵f`$ for $`𝔵𝔤`$ we write $`f𝔵_rf`$ for the $`𝔤`$-action derived from right translation. Then
$$I(f)=(𝔲^{})_r^{1+m}f$$
where $`𝔲^{}:=\left(\begin{array}{cc}0& 1\\ 0& 0\end{array}\right)𝔤`$.
Corresponding to the decomposition $`G=BP\dot{}BwP`$ where $`BG`$ is the Iwahori subgroup and $`w:=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)`$ the sequence $`()`$ is the direct sum of the sequences
$$0\mathrm{ind}_P^G(\chi _{alg})\underset{\text{ }Q_p}{}\mathrm{Ind}_{P,\mathrm{}}^{BP}(\chi _{lc})\stackrel{\tau }{}\mathrm{Ind}_P^{BP}(\chi )\stackrel{I}{}\mathrm{Ind}_P^{BP}(\chi ^{})0$$
and
$$0\mathrm{ind}_P^G(\chi _{alg})\underset{\text{ }Q_p}{}\mathrm{Ind}_{P,\mathrm{}}^{BwP}(\chi _{lc})\stackrel{\tau }{}\mathrm{Ind}_P^{BwP}(\chi )\stackrel{I}{}\mathrm{Ind}_P^{BwP}(\chi ^{})0.$$
The superscripts $`BP`$ and $`BwP`$ indicate the subspaces of those functions in the induced representation which are supported in $`BP`$ and $`BwP`$, respectively. Both these sequences can be computed explicitly as follows. Let $`U`$, resp. $`U^{}`$, be the unipotent radical of $`P`$, resp. $`P^{}`$, and define $`U_\mathrm{o}:=UB`$ and $`U_\mathrm{o}^{}:=U^{}B`$. Denoting by $`u`$, resp. $`u^{}`$, the function on $`U_\mathrm{o}`$, resp. $`U_\mathrm{o}^{}`$, which sends a matrix to its left lower, resp. right upper, entry we introduce the finite dimensional $`\text{ }Q_p`$-vector spaces $`Pol^m(U_\mathrm{o})`$ and $`Pol^m(U_\mathrm{o}^{})`$ of polynomials of degree $`m`$ in $`u`$ and $`u^{}`$, respectively, with coefficients in $`\text{ }Q_p`$. By restricting, resp. translating by $`w`$ and restricting, functions the above two sequences become isomorphic to
$$0Pol^m(U_\mathrm{o}^{})\underset{\text{ }Q_p}{}C^{\mathrm{}}(U_\mathrm{o}^{},K)\stackrel{\tau }{}C^{an}(U_\mathrm{o}^{},K)\stackrel{(\frac{d}{du^{}})^{1+m}}{}C^{an}(U_\mathrm{o}^{},K)0$$
and
$$0Pol^m(U_\mathrm{o})\underset{\text{ }Q_p}{}C^{\mathrm{}}(U_\mathrm{o},K)\stackrel{\tau }{}C^{an}(U_\mathrm{o},K)\stackrel{(\frac{d}{du})^{1+m}}{}C^{an}(U_\mathrm{o},K)0.$$
In these sequences the injectivity of the first map as well as the exactness in the middle are obvious. By Prop. 3.1 the subspace topology on the kernel of the second map is the finest locally convex topology. The surjectivity and strictness of the second map can either be checked directly or can be seen as a special case of the more general statement in \[Fea\] 2.5.4. This finishes the proof of the exactness of $`()`$.
The smooth $`G`$-representation $`\mathrm{Ind}_{P,\mathrm{}}^G(\chi _{lc})`$ is of length at most 2. More precisely one has (\[GGP\] p.173) that $`\mathrm{Ind}_{P,\mathrm{}}^G(\chi _{lc})`$ is irreducible except in the following cases:
A) $`\chi _{lc}=1`$ is the trivial character. Then $`\mathrm{Ind}_{P,\mathrm{}}^G(1)`$ contains the one dimensional trivial representation on the subspace of constant functions. The corresponding quotient is the socalled Steinberg representation which is irreducible.
B) $`\chi _{lc}`$ is the character $`\chi _{lc}(\left(\begin{array}{cc}t^1& 0\\ 0& t\end{array}\right))=|t|^2`$ where $`||`$ denotes the normalized absolute value of $`\text{ }Q_p`$. Then $`\mathrm{Ind}_{P,\mathrm{}}^G(\chi _{lc})`$ contains the Steinberg representation and the corresponding quotient is the one dimensional trivial representation.
C) $`\chi _{lc}`$ is of the form $`\chi _{lc}(\left(\begin{array}{cc}t^1& 0\\ 0& t\end{array}\right))=|t|\delta (t)`$ for some non-trivial quadratic character $`\delta :\text{ }Q_{p}^{}{}_{}{}^{\times }K^\times `$. Then $`\mathrm{Ind}_{P,\mathrm{}}^G(\chi _{lc})`$ either is irreducible (but not absolutely irreducible) or is the direct sum of two infinite dimensional non-equivalent irreducible $`G`$-representations.
If we combine this information with Prop. 4.1 and Cor. 3.5 we obtain a complete list of the quasi-simple constituents of $`M_\chi ^{loc}`$ up to isomorphism. In particular, each of them is isomorphic to the tensor product of $`\mathrm{ind}_P^G(\chi _{alg})`$ and the full $`K`$-linear dual of one of the irreducible smooth representations in the above list. At this point it should be mentioned that the length of a Jordan-Hölder series for the kernel of $`I`$ on $`\mathrm{Ind}_P^G(\chi )`$ was already determined in \[Mor\].
References
\[B-TVS\] Bourbaki, N.: Topological Vector Spaces. Berlin-Heidelberg-New York: Springer 1987
\[CR\] Curtis, C.W., Reiner, I.: Representation theory of finite groups and associative algebras. New York-London: Wiley 1962
\[DG\] Demazure, M., Gabriel, P.: Groupes Algébriques. Amsterdam: North-Holland 1970
\[Dix\] Dixmier, J.: Enveloping Algebras. AMS 1996
\[Fea\] Féaux de Lacroix, C. T.: Einige Resultate über die topologischen Darstellungen $`p`$-adischer Liegruppen auf unendlich dimensionalen Vektorräumen über einem $`p`$-adischen Körper. Thesis, Köln 1997, Schriftenreihe Math. Inst. Univ. Münster, 3. Serie, Heft 23, pp. 1-111 (1999)
\[GGP\] Gel’fand, I.M., Graev, M.I., Pyatetskii-Shapiro, I.I.: Representation Theory and Automorphic Functions. San Diego: Academic Press 1990
\[HC\] Harish-Chandra, van Dijk: Harmonic Analysis on Reductive $`p`$-adic Groups. Lect. Notes Math., vol. 162. Berlin-Heidelberg-New York: Springer 1970
\[Jan\] Jantzen, J.C.: Representations of Algebraic Groups. Orlando: Academic Press 1987
\[Mor\] Morita, Y.: Analytic Representations of $`SL_2`$ over a $`p`$-Adic Number Field, III. In Automorphic Forms and Number Theory, Adv. Studies Pure Math. 7, pp.185-222. Tokyo: Kinokuniya 1985
\[ST\] Schneider, P., Teitelbaum, J.: Locally analytic distributions and $`p`$-adic representation theory, with applications to $`GL_2`$. Preprint 1999
\[Vig\] Vigneras, M.-F.: Représentations $`l`$-modulaires d’un groupe réductifs $`p`$-adique avec $`lp`$. Progress in Math., vol. 137. Boston-Basel-Stuttgart: Birkhäuser 1996
Peter Schneider Mathematisches Institut Westfälische Wilhelms-Universität Münster Einsteinstr. 62 D-48149 Münster, Germany pschnei@math.uni-muenster.de http://www.uni-muenster.de/math/u/schneider
Jeremy Teitelbaum Department of Mathematics, Statistics, and Computer Science (M/C 249) University of Illinois at Chicago 851 S. Morgan St. Chicago, IL 60607, USA jeremy@uic.edu http://raphael.math.uic.edu/$``$jeremy
|
warning/0005/math0005178.html
|
ar5iv
|
text
|
# Normalizers of operator algebras and reflexivity
## 1. Introduction and preliminaries
This paper is devoted to the study of a class of linear spaces of operators on Hilbert space which arise as normalizers of operator algebras; we call them *normalizing* subspaces.
Normalizers of von Neumann algebras (in particular, of maximal abelian selfadjoint algebras - masas for short) are known to play an important role in various contexts. The non-selfadjoint generalizations of von Neumann algebras are the *reflexive algebras* first introduced by Halmos in . We will primarily be concerned with normalizers of such algebras. Recall that an operator $`T`$ normalizes a reflexive (not necessarily selfadjoint) algebra $`𝒜`$ if $`T`$ and its adjoint satisfy $`T^{}𝒜T𝒜`$. The set of all normalizers of $`𝒜`$ is of course not a linear space in general. However, it turns out that the action of a normalizer $`T`$ and its adjoint $`T^{}`$ on the invariant projections of $`𝒜`$ defines a linear space of operators all of which normalize $`𝒜`$. Moreover, this space is *reflexive* in the terminology of Loginov-Shulman and Erdos and is closed under the ‘triple product’ $`AB^{}C`$. Thus the set of normalizers of a reflexive (not necessarily selfadjoint) algebra (and indeed, the set of *semi-normalizers* between two reflexive algebras - see section 5) appears as the union of reflexive linear spaces which have additional algebraic structure. It is the interplay between linearity and normalization that forms the subject matter of the present work. For example, we show that every normalizer is the norm-limit of linear combinations of normalizing partial isometries, and every compact normalizer is the limit of finite rank normalizers. We also show that the sum of two normalizers of CSL algebras is again a normalizer only when both are contained in a single reflexive masa bimodule consisting of normalizers, and obtain generalizations of the results of Coates on normalizers of nest algebras.
These observations lead us to introduce the class of *normalizing* spaces, namely, those linear spaces of operators which are closed under the ‘triple product’ $`AB^{}C`$. These subspaces are interesting in their own right; they generalize selfadjoint algebras of operators, and share many properties in common with such algebras. For instance, they satisfy an analogue of the bicommutant theorem: they are reflexive whenever they are ultraweakly closed (unlike general non-selfadjoint algebras). Alternatively given a set $``$ of projections and a map $`\varphi `$ defined on $``$, the set of all operators $`T`$ satisfying $`TL=\varphi (L)T`$ for all $`L`$ is a normalizing subspace, and every ultraweakly closed normalizing subspace is of this form.
A normalizing subspace $`𝒰`$ is a bimodule over the selfadjoint algebras $`[𝒰^{}𝒰]`$ and $`[𝒰𝒰^{}]`$ and induces a complete lattice isomorphism $`\chi `$ between the invariant projections of the ‘non-degenerate parts’ of these algebras. Also, $`𝒰`$ normalizes the first algebra into the second (that is, $`T(𝒰^{}𝒰)T^{}𝒰𝒰^{}`$ and $`T^{}(𝒰𝒰^{})T𝒰^{}𝒰`$ for each $`T𝒰`$). Conversely, we are able to characterize when $`𝒰`$ normalizes a pair of reflexive (not necessarily selfadjoint) algebras $`𝒜`$ and $``$. Apart from the obvious relations $`𝒰^{}𝒰𝒜`$ and $`𝒰𝒰^{}`$, the map $`\chi `$ must induce a bijection between the invariant projection lattices of the ‘non-degenerate parts’ of these algebras.
Thus, if $`𝒰`$ is a normalizing space then the non-degenerate parts of the von Neumann algebras generated by $`[𝒰^{}𝒰]`$ and $`[𝒰𝒰^{}]`$ are Morita equivalent in the sense of Rieffel . Conversely, if $`𝒜`$ and $``$ are Morita equivalent W\*-algebras, then there are faithful representations of $`𝒜`$ and $``$ such that the bimodule which establishes the equivalence is represented as a normalizing space $`𝒰`$ of operators between the respective Hilbert spaces. In this paper our concern is not with the notion of Morita equivalence of (abstract) W\*-algebras, but rather with the properties of normalizers between (concrete) reflexive algebras and especially with the interplay between normalizers and reflexivity. Notice, however, that this connection between normalizers and Morita equivalence might not have been observed had we considered normalizers of a single algebra.
We prove that normalizing subspaces which are bimodules over two maximal abelian selfadjoint algebras consists of operators ‘supported’ on sets of the form $`[f=g]`$ where $`f`$ and $`g`$ are appropriate Borel functions. This includes the case of normalizing subspaces which are generated by rank one operators. In case one of the algebras $`[𝒰^{}𝒰]`$, $`[𝒰𝒰^{}]`$ is abelian, the support of $`𝒰`$ turns out to be the ‘graph’ or the ‘reverse graph’ of a Borel function. We also show that normalizing masa-bimodules satisfy spectral synthesis in the sense of Arveson . This gives a clear geometric description of the normalizers of a CSL algebra in terms of generalized graphs or partial graphs. These partial graphs are analogous to the ones appearing in the work of Feldman and Moore and others. In these papers, only partial isometries normalizing certain Cartan masas are considered, while in our work the emphasis is on the whole reflexive linear space generated by each generalised graph. Also, we deal with arbitrary (nonabelian and non-selfadjoint) CSL algebras.
The notation we use is standard; see for example . We review some definitions and facts from and . Let $`_1`$ and $`_2`$ be complex Hilbert spaces, $`𝒫_i`$ the lattice of all (orthogonal) projections on $`_i`$, $`i=1,2`$. We let $`(𝒫_1,𝒫_2)`$ denote the set of all maps $`\phi :𝒫_1𝒫_2`$ which are 0-preserving and $``$-continuous (i.e. preserve arbitrary suprema). Erdos shows that each $`\phi (𝒫_1,𝒫_2)`$ uniquely defines semi-lattices $`𝒮_{1\phi }𝒫_1`$ and $`𝒮_{2\phi }=\phi (𝒫_1)𝒫_2`$ such that $`\phi `$ is a bijection between $`𝒮_{1\phi }`$ and $`𝒮_{2\phi }`$ and is uniquely determined by its restriction to $`𝒮_{1\phi }`$. Moreover, $`𝒮_{1\phi }`$ is meet-complete and contains the identity projection while $`𝒮_{2\phi }`$ is join-complete and contains the zero projection.
Note that the set
$$Op\phi =\{T(_1,_2):\phi (P)^{}T(P)=0\text{for each }P𝒫_1\}$$
is also uniquely determined by $`\phi |_{𝒮_{1\phi }}`$: if $`T`$ satisfies $`\phi (P)^{}T(P)=0`$ for each $`P𝒮_{1\phi }`$, then $`TOp\phi `$.
Given a subspace $`𝒰(_1,_2)`$, we define its map $`Map𝒰:𝒫_1𝒫_2`$ by
$$(Map𝒰)(P)=\overline{[𝒰(P)]}(P𝒫_1)$$
(where, here and in the sequel, the symbol $`\overline{[𝒰(P)]}`$ will stand for the projection onto the closed subspace spanned by $`\{Sx:xP(_1),S𝒰\}`$).
If $`\phi ^{}=Map𝒰^{}`$, then $`𝒮_{1\phi }=\{P^{}:P\phi ^{}(𝒫_2)\}`$ .
The reflexive hull $`Ref𝒰`$ of $`𝒰`$ is defined to be the space
$$Ref𝒰=\{T(_1,_2):Tx\overline{𝒰x},\text{for each}x_1\}$$
. A subspace $`𝒰`$ is called reflexive if $`𝒰=Ref𝒰`$. It is easily seen that $`Ref𝒰=OpMap𝒰`$ . A *unital algebra* $`𝒜()`$ is reflexive if and only if it is of the form $`𝒜=Alg=\{A:L^{}AL=0\text{ for all }L\}`$, where $`=Lat𝒜`$ is the complete lattice of all invariant projections of $`𝒜`$.
Given $`\phi (𝒫_1,𝒫_2)`$ the subspace
$$𝒱=\{T(_1,_2):\phi (L)T(L^{})=0\text{for each }L𝒮_{1\phi }\}$$
is clearly reflexive. We denote its map by $`\phi ^{}`$. Thus $`𝒱=Op\phi ^{}`$ and $`\phi ^{}(𝒫_1,𝒫_2)`$ satisfies $`\phi ^{}(L^{})\phi (L)^{}`$ for each $`L𝒮_{1\phi }`$.
The following simple observations, whose proofs are routine, will be used repeatedly.
###### Lemma 1.1.
Let $`𝒰(_1,_2)`$ be a subspace and $`𝒜_i(_i)(i=1,2)`$ be unital algebras such that $`𝒜_2𝒰𝒜_1𝒰`$. If $`\phi =Map𝒰`$, then $`𝒮_{i\phi }Lat𝒜_i`$. Thus $`𝒜_1Alg𝒮_{1\phi }`$ and $`𝒜_2Alg𝒮_{2\phi }`$. If additionally $`𝒰`$ is reflexive, then the algebra $`Alg𝒮_{1\phi }`$ (resp. $`Alg𝒮_{2\phi }`$) is the largest algebra over which $`𝒰`$ is a right (resp. left) module.
## 2. Normalizing spaces of operators
The notion of reflexivity for subspaces generalizes the corresponding notion defined by Halmos for unital algebras. Among reflexive algebras, the selfadjoint ones, namely the von Neumann algebras, have of course a distinguished place. Note that a unital algebra $`𝒜`$ is selfadjoint if and only if $`𝒜𝒜^{}𝒜𝒜`$. As the results of this paper show, the generalization of this property to subspaces is particularly fruitful.
###### Definition 2.1.
A subspace $`𝒰(_1,_2)`$ is said to be normalizing if it is closed under the ‘triple product’ $`(A,B,C)AB^{}C`$.
###### Remark 2.2.
For a subspace $`𝒰(_1,_2)`$, the following are equivalent:
(i) $`𝒰`$ is normalizing.
(ii) There is a unital \*-algebra $`𝒜_1(_1)`$ such that $`𝒰𝒜_1𝒰`$ and $`𝒰^{}𝒰𝒜_1`$.
(iii) There is a subspace $`𝒜_1(_1)`$ such that $`𝒰𝒜_1𝒰`$ and $`𝒰^{}𝒰𝒜_1`$.
(iv) There is a unital \*-algebra $`𝒜_2(_2)`$ such that $`𝒜_2𝒰𝒰`$ and $`𝒰𝒰^{}𝒜_2`$.
(v) There is a subspace $`𝒜_2(_2)`$ such that $`𝒜_2𝒰𝒰`$ and $`𝒰𝒰^{}𝒜_2`$.
Proof. (i) $``$(ii) Let $`𝒜_1(_1)`$ be the linear span of $`𝒰_1=\{S^{}T:S,T𝒰\}\{I\}`$. Since $`𝒰𝒰^{}𝒰𝒰`$, one verifies that $`𝒰_1`$ is a unital \*-semigroup and so $`𝒜_1`$ is a unital \*-subalgebra. The properties $`𝒰^{}𝒰𝒜_1`$ and $`𝒰𝒜_1𝒰`$ are immediate.
(ii)$``$(iii) is trivial.
(iii)$``$(i) If $`S,T,R𝒰`$ then $`T^{}R𝒜_1`$ and hence $`S(T^{}R)𝒰`$.
The implications (i)$``$(iv)$``$(v)$``$(i) are equally easy. $`\mathrm{}`$
Remark Let $`𝒰`$ be a normalizing subspace, and write $`𝒜`$ and $``$ for the algebras generated by $`𝒰^{}𝒰`$ and $`𝒰𝒰^{}`$ respectively. Then $`𝒰`$ *normalizes $``$ into $`𝒜`$*, in the sense that $`T^{}T𝒜`$ and $`T𝒜T^{}`$ for all $`T𝒰`$. Conversely, if $`𝒰`$ is a subspace of operators, which normalizes an algebra into another, then, as we shall prove (see Proposition 5.10), $`𝒰`$ is contained in a normalizing space of operators.
If $`𝒜`$ is a unital selfadjoint algebra, then its invariant subspace lattice $``$ is orthocomplemented; thus its map $`\chi =Map(𝒜)`$ (namely, the identity map of $``$) preserves orthogonality. This property characterizes maps of normalizing subspaces.
###### Definition 2.3.
A map $`\chi (𝒫_1,𝒫_2)`$ is said to be an ortho-map if $`\chi (L)\chi (L^{})`$ for each $`L𝒮_{1\chi }`$.
The following theorem shows the connection between ortho-maps and normalizing spaces. Statement (b)(ii) corresponds to the von Neumann Bicommutant Theorem.
###### Theorem 2.4.
(a) Let $`\mathrm{\phi }(𝒫_1,𝒫_2)`$ and
$$𝒰=\{T(_1,_2):TL=\phi (L)T\text{for each }L𝒮_{1\phi }\}.$$
Then $`𝒰=Op\phi Op\phi ^{}`$ is a normalizing space.
(b) Let $`𝒰(_1,_2)`$ be a normalizing subspace. Then
(i) $`Map𝒰`$ is an ortho-map and
(ii) $`Ref(𝒰)=\text{cl}_{WOT}(𝒰)=\text{cl}_{uw}(𝒰)`$.
(c) Let $`\mathrm{\chi }(𝒫_1,𝒫_2)`$ be an ortho-map. Then
$$Op\chi =\{T(_1,_2):TL=\chi (L)T\text{for each }L𝒮_{1\chi }\}.$$
Proof. (a) The equality $`𝒰=Op\phi Op\phi ^{}`$ is easily verified.
Let $`S,T,R𝒰`$. Then for each $`L𝒮_{1\phi }`$ we have
$$(ST^{}R)L=ST^{}\phi (L)R=SLT^{}R=\phi (L)(ST^{}R)$$
and so $`ST^{}R𝒰`$ which shows that $`𝒰`$ is normalizing.
(b) Let $`𝒜_i(_i)`$ be unital \*-algebras such that $`𝒰^{}𝒰𝒜_1`$, $`𝒰𝒰^{}𝒜_2`$ and $`𝒜_2𝒰𝒜_1𝒰`$ (Remark 2.2).
(i) Let $`\chi =Map𝒰`$. We show that $`\chi (L)\chi (L^{})`$ for each $`LLat𝒜_1`$ (this will suffice since $`𝒮_{1\chi }Lat𝒜_1`$ by Lemma 1.1). Indeed, for each $`S,T𝒰`$ we have $`S^{}T𝒜_1`$ and so, if $`\xi L`$ and $`\eta L^{}`$ then
$$T\xi ,S\eta =S^{}T\xi ,\eta =0.$$
This shows that $`T\xi \chi (L^{})`$ for each $`T𝒰`$ and hence $`\chi (L)\chi (L^{})`$.
(ii) The properties $`𝒰^{}𝒰𝒜_1,𝒰𝒰^{}𝒜_2`$ and $`𝒜_2𝒰𝒜_1𝒰`$ ensure that the set
$$𝒞=\left(\begin{array}{cc}𝒜_2& 𝒰\\ 𝒰^{}& 𝒜_1\end{array}\right)=\{\left(\begin{array}{cc}B& T\\ S^{}& A\end{array}\right):A𝒜_1,B𝒜_2,S,T𝒰\}$$
is a unital \*-subalgebra of $`(_2_1)`$ and so $`\text{cl}_{WOT}(𝒞)=\text{cl}_{uw}(𝒞)=𝒞^{\prime \prime }`$ by the von Neumann bicommutant theorem. But, since $`𝒞`$ is a unital \*-algebra, it is easy to verify that $`𝒞^{\prime \prime }=AlgLat𝒞=Ref(𝒞)`$. This implies in particular that $`\text{cl}_{WOT}(𝒰)=\text{cl}_{uw}(𝒰)=Ref(𝒰)`$.
(c) If $`T(_1,_2)`$ satisfies $`TL=\chi (L)T`$ for each $`L𝒮_{1\chi }`$, then $`TL=\chi (L)TL`$ and so $`TOp\chi `$ since the latter is determined by $`𝒮_{1\chi }`$. If $`\chi `$ is an ortho-map and $`TOp\chi `$, then for each $`L𝒮_{1\chi }`$ the relation $`TL^{}=\chi (L^{})TL^{}`$ gives $`\chi (L)TL^{}=\chi (L)\chi (L^{})TL^{}=0`$ since $`\chi (L)\chi (L^{})`$. Adding to this the relation $`TL=\chi (L)TL`$ gives $`TL=\chi (L)T`$ as required. $`\mathrm{}`$
We isolate two consequences of this theorem for emphasis.
###### Corollary 2.5.
(i) The w\*-closure of a normalizing subspace is reflexive and coincides with its WOT-closure.
(ii) A reflexive subspace is normalizing if and only if its map is an ortho-map.
Remarks We do not know whether a WOT-closed subspace whose map is an ortho-map must be normalizing. Note that the map of a unital algebra is an ortho-map if and only if it its invariant subspace lattice is orthocomplemented. Thus the question, within the class of unital algebras, reduces to the well-known reductive algebra problem : must a WOT-closed algebra whose invariant lattice is orthocomplemented be selfadjoint?
A subspace whose map is an ortho-map need not be normalizing. Indeed there exist nonselfadjoint transitive algebras (even triangular ones - see ). However, as we show below (Corollary 4.3) an ultraweakly closed subspace whose map is an ortho-map must be normalizing, provided it is a masa bimodule.
A crucial property of von Neumann algebras is that they are generated by their projections. Of course, normalizing spaces need not contain any (nontrivial) projections; their role is played by the partial isometries.
###### Proposition 2.6.
If $`𝒰`$ is an ultraweakly closed normalizing space and $`A=U|A|`$ is the polar decomposition of an element of $`𝒰`$, then $`U`$ is a partial isometry in $`𝒰`$ and $`Uf(|A|)𝒰`$, for every Borel function $`f`$ on the spectrum $`sp(|A|)`$ of $`|A|`$. Moreover, $`𝒰`$ is the norm-closed linear span of the partial isometries it contains.
Proof. Let $`𝒜_1`$ be a von Neumann algebra with the property $`𝒰^{}𝒰𝒜_1`$ and $`𝒰𝒜_1𝒰`$. Then $`A^{}A𝒜_1`$ and so $`|A|𝒜_1`$. We have $`U=`$w-$`lim_{\epsilon 0}A(|A|+\epsilon )^1`$ and so $`U𝒰`$. Since $`|A|𝒜_1`$, for every Borel function $`f`$ on $`sp(|A|)`$, the operator $`f(|A|)`$ is in $`𝒜_1`$ as well and since $`𝒰`$ is a right $`𝒜_1`$-module, it follows that $`Uf(|A|)𝒰`$.
Given $`A𝒰`$ and $`ϵ>0`$, there are spectral projections $`P_1,\mathrm{},P_n`$ of $`|A|`$ and scalars $`c_1,\mathrm{},c_n`$ such that $`|A|c_iP_i<ϵ`$. Thus $`Ac_iUP_i<ϵ`$ and each $`UP_i`$ is a partial isometry in $`𝒰`$, since the initial projection of $`U`$ is the range projection of $`|A|.\mathrm{}`$
A w\*-closed normalizing subspace need not contain (nonzero) finite rank operators. We show that then it cannot contain compact operators, contrary to the situation in general reflexive subspaces .
###### Corollary 2.7.
Let $`𝒰`$ be an ultraweakly closed normalizing space and suppose that $`K`$ is a compact operator in $`𝒰`$. Then $`K`$ can be approximated in the norm topology by finite rank operators in $`𝒰`$. Moreover, if $`K`$ belongs to some Schatten class $`𝒞_p`$ then it can be approximated by finite rank operators in $`𝒰`$ in the $`p`$-norm topology.
Proof. Immediate from Proposition 2.6. $`\mathrm{}`$
Let us note that the last corollary can be inferred in a different way. Namely, consider the “matrix” algebra $`𝒞`$ defined in the proof of Theorem 2.4. Then $`𝒞`$ is a von Neumann algebra (Remark 2.2) and the result is clear for $`𝒞`$.
If $`𝒰`$ is a normalizing space, then its rank one subspace (i.e. the linear span of the rank one operators contained in $`𝒰`$) is also normalizing. The next proposition characterizes this subspace in terms of the map of $`𝒰`$. We denote the rank one operator sending $`\xi `$ to $`\xi ,xy`$ by the symbol $`yx^{}`$.
###### Proposition 2.8.
Let $`\chi `$ be an ortho-map. A rank one operator $`yx^{}`$ belongs to $`Op\chi `$ if and only if, for each $`L𝒮_{1\chi }`$,
$$Lx0L^{}x=0\chi (L)^{}y=0\chi (L)y0.$$
Proof. By Theorem 2.4, $`Op\chi =\{T(_1,_2):TL=\chi (L)T,L𝒮_{1\chi }\}`$. Suppose that the rank one operator $`yx^{}`$ belongs to $`Op\chi `$. Then, for each $`L𝒮_{1\chi }`$, $`(\chi (L)^{}y)(Lx)^{}=\chi (L)^{}(yx^{})L=0`$ and $`(\chi (L)y)(L^{}x)^{}=\chi (L)(yx^{})L^{}=0`$. So for each $`L𝒮_{1\chi }`$, (either $`\chi (L)^{}y=0`$ or $`Lx=0`$) and (either $`\chi (L)y=0`$ or $`L^{}x=0`$). If $`Lx0`$ and $`L^{}x0`$, then by the above conditions, we conclude that $`\chi (L)^{}y=\chi (L)y=0`$, so $`y=0`$, which is impossible. Because $`Lx`$ and $`L^{}x`$ cannot both be zero, we conclude that $`Lx0L^{}x=0`$. In the same way $`\chi (L)^{}y0\chi (L)y=0`$. Similarly, if $`Lx0`$, then $`\chi (L)^{}y=0`$ and, conversely, if $`\chi (L)^{}y=0`$, then $`\chi (L)y0`$ and so $`L^{}x=0`$. The converse is trivial. $`\mathrm{}`$
We would now like to show how the general theory of Erdos specializes in the case of ortho-maps.
###### Definition 2.9.
For a map $`\chi `$, set $`I_{}=\chi (I)𝒫_2`$ and $`0_+=\{P:\chi (P)=0\}𝒫_1`$. The map $`\chi `$ will be called essential if $`0_+=0`$ and $`I_{}=I`$. A subspace of operators $`𝒰`$ will be called essential, if $`Map𝒰`$ is essential.
###### Theorem 2.10.
Suppose that $`𝒰`$ is a normalizing space of operators and let $`\chi =Map𝒰`$. Then the semi-lattices $`𝒮_1`$ and $`𝒮_2`$ of $`\chi `$ are complete ortho-lattices and $`\chi `$ is a complete ortho-lattice isomorphism of $`𝒮_1`$ onto $`𝒮_2`$. Moreover, if $`𝒯_1=𝒮_1|_{0_+^{}_1}`$ and $`𝒯_2=𝒮_2|_{I_{}_2}`$, then $`𝒯_1`$ and $`𝒯_2`$ are the projection lattices of von Neumann algebras and
(i) $`Alg𝒮_1=\{\left(\begin{array}{cc}A& 0\\ C& D\end{array}\right):A𝒯_{1}^{}{}_{}{}^{},C(0_+^{}_1,0_+_1),D(0_+_1,0_+_1)\}`$ and $`(𝒰^{}𝒰)^{\prime \prime }=\{\left(\begin{array}{cc}A& 0\\ 0& \lambda I\end{array}\right):A𝒯_{1}^{}{}_{}{}^{},\lambda \}`$
(ii) $`Alg𝒮_2=\{\left(\begin{array}{cc}B& C\\ 0& D\end{array}\right):B𝒯_{2}^{}{}_{}{}^{},C(I_{}^{}_2,I_{}_2),D(I_{}^{}_2,I_{}^{}_2)\}`$ and $`(𝒰𝒰^{})^{\prime \prime }=\{\left(\begin{array}{cc}B& 0\\ 0& \lambda I\end{array}\right):B𝒯_{2}^{}{}_{}{}^{},\lambda \}`$.
(iii) $`Lat(𝒰^{}𝒰)=\{L𝒫_1:\chi (L)\chi (L^{})\}=\{L_1L_2:L_1𝒯_1\}`$.
###### Proof.
Put $`\chi ^{}=Map𝒰^{}`$. Recall that $`𝒮_1=\{P^{}:P\chi ^{}(𝒫_2)\}`$ and $`𝒮_2=\chi (𝒫_1)`$. First assume that $`𝒰`$ is essential. Let $`𝒳=Lat(𝒰^{}𝒰)`$ and $`𝒴=Lat(𝒰𝒰^{})`$. Since $`𝒰^{}𝒰`$ and $`𝒰𝒰^{}`$ are selfadjoint sets of operators, $`𝒳`$ and $`𝒴`$ are complete orthocomplemented projection lattices.
We will show that $`\chi `$ is an ortho-isomorphism from $`𝒳`$ onto $`𝒴`$.
First, if $`L𝒳`$, then given $`T,S𝒰`$ we have $`(TS^{})(R(L_1))=(TS^{}R)(L_1)\chi (L)_2`$ for all $`R𝒰`$. Hence $`(TS^{})(\chi (L)_2)\chi (L)_2`$ and so $`\chi (L)𝒴`$. Similarly, $`\chi ^{}(𝒴)𝒳`$.
Next, we show that, if $`L𝒳`$, then $`\chi ^{}\chi (L)=L`$. For $`T,S𝒰`$ we have $`(T^{}S)L=L(T^{}S)`$ since $`L(𝒰^{}𝒰)^{}`$ and so $`L^{}(T^{}S)L=0`$ and $`L(T^{}S)L^{}=0`$. The first relation gives $`L^{}T^{}(\chi (L))=0`$ for each $`T𝒰`$ and so $`\chi ^{}\chi (L)L`$. Similarly the second gives $`\chi ^{}\chi (L^{})L^{}`$. But $`\chi ^{}\chi (LL^{})=\chi ^{}\chi (I)=I`$ and so $`\chi ^{}\chi (L)\chi ^{}\chi (L^{})=I`$. The relation $`\chi ^{}\chi (L)=L`$ now follows readily. Applying the same arguments to the space $`𝒰^{}`$, it follows that, if $`M𝒴`$, then $`\chi \chi ^{}(M)=M`$.
So we have shown that $`\chi `$ maps $`𝒳`$ bijectively to $`𝒴`$ and its inverse is $`\chi ^{}`$. But $`\chi `$ preserves orthogonality (indeed if $`T,S𝒰`$ and $`L𝒳`$ then $`TLx,SL^{}y=`$ $`S^{}TLx,L^{}y=0`$ since $`L^{}S^{}TL=0`$) and since it is $``$-continuous and preserves $`0`$ and $`I`$, it follows that it is $``$-continuous as well and hence an ortho-isomorphism.
Finally, notice that $`𝒮_2=𝒴`$ and $`𝒮_1=𝒳`$. Indeed, given $`P𝒫_1`$ and $`T,S𝒰`$ we have $`\chi (P)^{}(TS^{})RP=0`$ for all $`R𝒰`$ (since $`TS^{}R𝒰`$ and $`\overline{[𝒰(P)]}=\chi (P)`$), hence $`\chi (P)^{}(TS^{})\chi (P)=0`$. Thus $`\chi (P)𝒴`$, for each $`P𝒫_1`$ and so $`𝒮_{2\chi }=\chi (𝒫_1)𝒴`$. Also if $`M𝒴`$ then $`M=\chi (L)𝒮_{2\chi }`$ where $`L=\chi ^{}(M)`$. This shows that $`𝒴=𝒮_2`$. The proof that $`𝒳=𝒮_1`$ is similar.
Now relax the assumption that $`𝒰`$ is essential. Note that $`0_+^{}`$ acts as the identity on $`𝒰^{}𝒰`$ and hence $`0_+(𝒰^{}𝒰)^{\prime \prime }`$. Similarly $`I_{}(𝒰𝒰^{})^{\prime \prime }`$ acts as the identity on $`𝒰𝒰^{}`$ and $`𝒰(_1)=𝒰(0_+^{}_1)I_{}_2`$. Let $`𝒦_1=0_+^{}_1`$, $`𝒦_2=I_{}_2`$ and $`𝒰_o=𝒰|_{𝒦_1}(𝒦_1,𝒦_2)`$. Then $`𝒰_o`$ is normalizing and essential. If $`\chi _o`$ is the map of $`𝒰_o`$ then clearly its semilattices are $`𝒮_{1\chi _o}=𝒮_{1\chi }|_{𝒦_1}=𝒯_1`$ and $`𝒮_{2\chi _o}=𝒮_{2\chi }|_{𝒦_2}=𝒯_2`$. By what was shown above, $`𝒯_1=Lat(𝒰_o^{}𝒰_o)`$, $`𝒯_2=Lat(𝒰_o𝒰_o^{})`$ and $`\chi _o:𝒯_1𝒯_2`$ is an isomorphism. Also $`(𝒰^{}𝒰)^{\prime \prime }=\overline{[𝒰_o^{}𝒰_o]}I_{𝒦_1}`$ and $`(𝒰𝒰^{})^{\prime \prime }=\overline{[𝒰_o𝒰_o^{}]}I_{𝒦_2}`$.
It follows that $`Lat(𝒰^{}𝒰)=\{L_1L_2:L_1𝒯_1\}`$, while the other equality of (iii) is readily verified, in fact for every subspace $`𝒰`$ with map $`\chi `$.
Now a projection $`M𝒫_2`$ is in $`𝒮_{2\chi }=\chi (𝒫_1)`$ if and only if it is of the form $`M=M_10`$ with respect to the decomposition $`_2=𝒦_2𝒦_2^{}`$, where $`M_1𝒮_{2\chi _o}`$. Similarly working with $`𝒰^{}`$ we obtain that $`\chi ^{}(𝒫_2)=\{L_10:L_1𝒮_{1\chi _o}\}`$ with respect to the decomposition $`_1=𝒦_1𝒦_1^{}`$. Thus
$$𝒮_{1\chi }=\{L_1I:L_1𝒯_1\},𝒮_{2\chi }=\{M_10:M_1𝒯_2\}.$$
From this we see that $`XAlg𝒮_{1\chi }`$ if and only if $`(L_1I)^{}X(L_1I)=0`$ for each $`L_1𝒯_1`$, and claim (i) follows easily. The proof of (ii) is similar. $`\mathrm{}`$
Remarks (i) It is not difficult to show that the von Neumann algebras $`𝒯_1^{\prime \prime }`$ and $`𝒯_2^{\prime \prime }`$ are in fact \*-isomorphic, and so the strongly closed algebras generated by $`𝒰^{}𝒰`$ and $`𝒰𝒰^{}`$ are Morita equivalent . We have preferred the direct approach above, which is sufficient for our needs.
(ii) The first part of the above theorem remains valid even if $`\chi `$ is not given as a map of some normalizing subspace, but is an arbitrary ortho-map. That is, the following lattice-theoretic result holds: If $`\chi `$ is an ortho-map, then its semi-lattices $`𝒮_1`$ and $`𝒮_2`$ are in fact complete ortho-lattices and $`\chi :𝒮_1𝒮_2`$ is a complete ortho-isomorphism .
The next corollary shows that the semilattices of the map of a normalizing space become *reflexive* lattices, if the extreme elements $`0`$ and $`I`$ are adjoined.
###### Corollary 2.11.
Suppose that $`𝒮_1`$ and $`𝒮_2`$ are the (semi-)lattices of the map $`\chi `$ of some normalizing subspace. Then $`LatAlg𝒮_1=𝒮_1\{0\}`$ and $`LatAlg𝒮_2=𝒮_2\{I\}`$. Moreover, in the notation of Theorem 2.10, the orthocomplemented projection lattice generated by $`𝒮_1`$, is $`𝒮_1\{P_10:P_1𝒯_1\}`$, while the orthocomplemented projection lattice generated by $`𝒮_2`$ is $`𝒮_2\{Q_1I:Q_1𝒯_2\}.`$
Proof. Let $`\mathrm{\Sigma }_1=LatAlg𝒮_1`$, $`\mathrm{\Sigma }_2=LatAlg𝒮_2`$ and $`\stackrel{~}{\mathrm{\Sigma }_1}`$ and $`\stackrel{~}{\mathrm{\Sigma }_2}`$ be the orthocomplemented lattices generated by $`\mathrm{\Sigma }_1`$ and $`\mathrm{\Sigma }_2`$ respectively. Suppose that a projection $`Q`$ is left invariant by the algebra $`Alg𝒮_1`$. Then by Theorem 2.10 it must be contained in $`Lat(𝒰^{}𝒰)`$; thus $`Q`$ is of the form $`Q=Q_1Q_2`$, where $`Q_1𝒯_1`$. Since $`0_+Alg𝒮_1|_{0_+_1}=(0_+_1,0_+_1)`$, it follows that either $`Q_2=I`$ or $`Q_2=0`$. If $`Q_2=I`$ then $`Q𝒮_1`$. If $`Q_2=0`$, then $`CQ_1=0`$ for each $`C(0_+^{}_1,0_+_2)`$ and it follows that $`Q_1=0`$. Thus we have that $`\mathrm{\Sigma }_1=𝒮_1\{0\}`$.
It is easy to check that the set $`𝒮_1\{Q_10:Q_1𝒯_1\}`$ is an orthocomplemented lattice. Since it contains $`𝒮_1`$, it must equal $`\stackrel{~}{\mathrm{\Sigma }_1}`$.
The other identities are proved by the same arguments. $`\mathrm{}`$
Recall that a subspace $`𝒰`$ is called strongly reflexive , if there is a set of rank one operators $``$ such that $`𝒰=Ref`$. We wish to describe the strongly reflexive normalizing subspaces. The unital algebra case might be known; we include a proof for completeness.
###### Lemma 2.12.
Every strongly reflexive selfadjoint unital algebra contains a totally atomic maximal abelian selfadjoint algebra (a masa).
###### Proof.
If $`𝒜()`$ is a strongly reflexive selfadjoint unital algebra, then it is a von Neumann algebra, since it is reflexive. If $`=`$ $`Lat𝒜`$, then $`=Lat`$, where $`𝒜`$ is the rank one subalgebra of $`𝒜`$, by strong reflexivity. It follows from that $``$ is completely distributive. But $``$ is orthocomplemented, hence, by Tarski’s theorem (see , p.119), it must be a complete atomic Boolean lattice. Thus, as is well-known, $``$ is commutative, hence $`𝒜`$. Therefore there exists a totally atomic masa $``$ with $`𝒜`$. $`\mathrm{}`$
###### Proposition 2.13.
Every strongly reflexive normalizing subspace $`𝒰(_1,_2)`$ is a bimodule over two totally atomic masas.
###### Proof.
Let $`x\overline{𝒰^{}_2}𝒦_1`$. If $`S,T𝒰`$, then, since $`𝒰`$ is strongly reflexive, $`S^{}Tx`$ can be approximated by vectors of the form $`S^{}Rx`$, where $`R`$ belongs to the rank one subspace of $`𝒰`$. The strong closure of the linear span of $`𝒰^{}𝒰|_{𝒦_1}`$ is therefore a strongly reflexive von Neumann algebra acting on the space $`𝒦_1`$. Thus it must contain a totally atomic masa by Lemma 2.12. Since $`(𝒦_1^{},𝒦_1^{})`$ also contains a totally atomic masa, it follows from Theorem 2.10 that $`Alg𝒮_1`$ contains a totally atomic masa.
The proof that $`Alg𝒮_2`$ contains a totally atomic masa is identical. Since $`𝒰`$ is an $`(Alg𝒮_2,Alg𝒮_1)`$-bimodule, we are done. $`\mathrm{}`$
We will in fact prove (Theorem 3.5) that normalizing strongly reflexive subspaces are of the form $`_\lambda (_\lambda ,𝒦_\lambda )`$, for some mutually orthogonal families of subspaces $`_\lambda _1`$ and $`𝒦_\lambda _2`$.
## 3. Normalizing masa bimodules
In this and the next section, we specialize to subspaces $`𝒰(_1,_2)`$ which are bimodules over maximal abelian selfadjoint algebras (for short masas) $`𝒟_1(_1)`$ and $`𝒟_2(_2)`$ in the sense that $`𝒟_2`$$`𝒰`$$`𝒟_1`$$`𝒰`$. As we will be using measure-theoretic arguments, we will make the blanket assumption that from now on *all Hilbert spaces will be separable*. If $`_1`$ and $`_2`$ are complete projection lattices, we let
$$(_1,_2)=\{\phi (𝒫_1,𝒫_2):𝒮_{i\phi }_i,i=1,2\}.$$
###### Theorem 3.1.
A w\*-closed masa-bimodule $`𝒰`$ is normalizing if and only if there exist nests $`𝒩_1,𝒩_2`$ and a map $`\phi (𝒩_1,𝒩_2)`$ such that $`𝒰=Op(\phi )Op(\phi ^{})`$.
Proof. Every space of the form $`𝒰=Op(\phi )Op(\phi ^{})`$ is normalizing (Theorem 2.4); and if $`\phi `$ is a nest map, then $`𝒰`$ is a masa bimodule.
For the converse, write $`𝒰`$ as $`Op\chi `$ for an ortho-map $`\chi `$. Recall (Theorem 2.10) that the lattice $`𝒮_{1o}=𝒮_{1\chi }|_{(0_+_1)^{}}`$ is orthocomplemented, hence, since $`𝒮_{1\chi }𝒟_1`$, a (complete) Boolean lattice. There exists a complete nest $`𝒩_o𝒮_{1o}`$ on the space $`(0_+_1)^{}`$, generating $`𝒮_{1o}`$ as a complete Boolean lattice . Then the nest $`𝒩_1=\{QI:Q𝒩_o\}\{0\}`$ on $`_1`$ will generate the complete Boolean lattice $`\stackrel{~}{\mathrm{\Sigma }_1}`$ generated by $`𝒮_{1\chi }`$. We define the map $`\phi `$ to be the restriction of $`\chi `$ to $`𝒩_1`$. From the fact that $`𝒩_1`$ is contained in $`𝒮_{1\chi }\{0\}`$ it is clear that the left semi-lattice of $`\phi `$, $`𝒮_{1\phi }`$, equals $`𝒩_1\{0\}`$ (or $`𝒩_1`$, if $`0_+=0`$). If $`XOp(\chi )`$, then certainly $`XOp(\phi )Op(\phi ^{})`$. Conversely, let $`XOp(\phi )Op(\phi ^{})`$. For each $`N𝒩_1`$, we have $`XN_1\phi (N)_2=\chi (N)_2`$ and $`XN^{}_1\phi (I)\phi (N)^{}_2=\chi (I)\chi (N)^{}_2=\chi (N^{})_2`$. Because $`𝒩_1`$ generates $`\stackrel{~}{\mathrm{\Sigma }_1}`$ and $`\chi `$ preserves arbitrary unions and intersections, it follows that $`XOp\chi `$. $`\mathrm{}`$
To state the next results, we need to recall some terminology from . Let $`(X,\mu )`$, $`(Y,\nu )`$ be standard Borel spaces and let $`𝒟_1,𝒟_2`$ be the multiplication masas on the corresponding $`L^2`$ spaces $`_1,_2`$, with projections denoted by $`E(\alpha )𝒟_1`$ and $`F(\beta )𝒟_2`$, where $`\alpha X`$ and $`\beta Y`$ are Borel sets. A bounded operator $`T:_1_2`$ is said to be supported by a set $`\kappa X\times Y`$ if $`F(\beta )TE(\alpha )=0`$ whenever $`(\alpha \times \beta )\kappa =\mathrm{}`$. It is shown in that a masa bimodule $`𝒰`$ is reflexive precisely when there exists a set $`\kappa X\times Y`$ such that $`𝒰`$ consists of all operators supported by $`\kappa `$. This set is uniquely defined up to marginal equivalence, and is called the $`\omega `$-support of $`𝒰`$; its complement is (marginally equivalent to) a countable union of Borel rectangles.
###### Theorem 3.2.
Let $`_1=L^2(X,\mu )`$, $`_2=L^2(Y,\nu )`$ and let $`𝒰(_1,_2)`$ be a w\*-closed normalizing space which is a bimodule over the multiplication masas $`𝒟_i(_i)`$. Then there exist Borel functions $`f:X[0,1]`$ and $`g:Y[0,1]`$ such that the $`\omega `$-support of $`𝒰`$ is the set $`\kappa =\{(x,y)X\times Y:f(x)=g(y)\}`$.
Conversely, the set of all operators supported by a set of the above form is a w\*-closed normalizing masa bimodule.
Proof Let $`\chi `$ be the map of $`𝒰`$, and let $`X_oX`$ and $`Y_oY`$ be Borel sets such that $`E(X_o)=\overline{[𝒰^{}_2]}`$ and $`F(Y_o)=\overline{[𝒰_1]}`$. Theorem 2.10 shows that $`𝒮_{1o}=𝒮_{1\chi }|_{E(X_o)}`$ and $`𝒮_{2o}=𝒮_{2\chi }|_{F(Y_o)}`$ are complete orthocomplemented lattices, hence (since they are commutative) complete Boolean lattices, and that $`\chi `$ induces a complete Boolean lattice isomorphism $`\chi _o`$ between them. Denote by $`𝔄_i`$ the w\*-closed algebra generated by $`𝒮_{io}`$ ($`i=1,2`$), and let $`\psi :𝔄_1𝔄_2`$ be the \*-isomorphism induced by $`\chi _o`$. Now the $`𝔄_i`$ are abelian separably acting von Neumann algebras, hence there exist standard Borel probability spaces $`(X_1,\mu _1)`$ and $`(Y_1,\nu _1)`$ such that the $`𝔄_i`$ are \*-isomorphic to the corresponding $`L^{\mathrm{}}`$ spaces. Moreover, we can take both $`X_1`$ and $`Y_1`$ to be compact intervals, which we take to be $`[\frac{1}{3},\frac{2}{3}]`$ for notational convenience. Now the inclusion $`𝔄_1𝒟_1`$ maps the identity of $`𝔄_1`$ to $`E(X_o)`$, and hence induces an injective unital \*-homomorphism $`\theta _1:L^{\mathrm{}}(X_1,\mu _1)L^{\mathrm{}}(X_o,\mu )`$. This map is implemented (see ) by a Borel function $`f_1:X_oX_1`$; thus every projection $`E𝔄_1`$ is of the form $`E=E(f_1^1(\sigma ))`$ for some Borel subset $`\sigma X_1`$. Since $`\theta _1`$ is injective, $`f_1`$ can be taken to be onto $`X_1`$. Similarly, there exists a Borel onto function $`g_1:Y_oY_1`$ such that every projection $`F𝔄_2`$ is of the form $`F=F(g_1^1(\tau ))`$ for some Borel subset $`\tau Y_1`$. Now the \*-isomorphism $`\psi :𝔄_1𝔄_2`$ is also implemented by a Borel bijection $`h:Y_1X_1`$; that is, for every projection $`E(f_1^1(\sigma ))`$ we have $`\psi (E(f_1^1(\sigma )))=F(g_1^1(h^1(\sigma )))`$. Define $`f:X[0,1]`$ by $`f(x)=f_1(x)`$ for $`xX_o`$ and $`f(x)=0`$ otherwise, and define $`g:Y[0,1]`$ by $`g(y)=h(g_1(y))`$ for $`yY_o`$ and $`g(y)=1`$ otherwise. Thus $`𝒮_{1\chi }=\{E(f^1(\sigma )):\sigma [0,1]`$ Borel, $`0\sigma \}`$ and $`\chi (E(f^1(\sigma ))=F(g^1(\sigma ))`$.
Let $`\kappa =\{(x,y)X\times Y:f(x)=g(y)\}`$. Pick a countable dense subset $`\{s_n\}[0,1)`$ and note that $`\kappa =\{(x,y)X\times Y:`$for each $`n,x\alpha _ny\beta _n\}`$ where $`\alpha _n=f^1([0,s_n])`$ and $`\beta _n=g^1([0,s_n])`$, so that $`\chi (E(\alpha _n))=F(\beta _n)`$ and $`\chi (E(\alpha _n^c))F(\beta _n^c)`$. Thus, the complement of $`\kappa `$ can be written $`\kappa ^c=_k\gamma _k\times \delta _k`$ where $`\chi (E(\gamma _k))F(\delta _k^c)`$ for each $`k`$.
It is shown in the proof of Theorem 4.2 of that an operator $`T`$ is supported by $`\kappa `$ if and only if $`F(\delta _k)TE(\gamma _k)=0`$ for all $`k`$. Thus if $`T𝒰`$ then $`T`$ is supported by $`\kappa `$.
Conversely, suppose $`T`$ is supported by $`\kappa `$. Then for each $`E=E(f^1(\sigma ))𝒮_{1\chi }`$ we have $`\chi (E)^{}TE=F(g^1(\sigma )^c)TE(f^1(\sigma ))`$ since $`\chi (E)=F(g^1(\sigma ))`$. But the rectangle $`f^1(\sigma )\times g^1(\sigma )^c`$ is disjoint from $`\kappa `$, hence $`\chi (E)^{}TE=0`$ so that $`T𝒰`$.
Finally, if $`𝒰`$ consists of all operators supported by a set $`\kappa `$ of the above form, then writing $`\kappa =\{(x,y)X\times Y:\text{ for each }n,x\alpha _ny\beta _n\}`$ as above we have that an operator $`T`$ is in $`𝒰`$ if and only if $`F(\beta _n)^{}TE(\alpha _n)=0`$ and $`F(\beta _n)TE(\alpha _n)^{}=0`$ for each $`n`$. Thus $`𝒰=\{T(_1,_2):TE(\alpha _n)=F(\beta _n)T`$ for each $`n\}`$, and this is clearly closed under the triple product. $`\mathrm{}`$
###### Corollary 3.3.
In the notation of the last Theorem, if the algebra $`[𝒰𝒰^{}]`$ (resp. $`[𝒰^{}𝒰]`$) is abelian, then the $`\omega `$-support of $`𝒰`$ is the graph (resp. ‘reverse graph’) of a Borel function $`f:X_oY`$ (resp. $`g:Y_oX`$) for a suitable Borel subset $`X_oX`$ (resp. $`Y_oY`$).
Proof If $`[𝒰^{}𝒰]`$ is abelian, then its w\*-closure is maximal abelian on $`\overline{[𝒰^{}_2]}`$. Hence it must be unitarily equivalent to the multiplication algebra of $`L^{\mathrm{}}(X_o,\mu )`$. It follows as in the above proof (essentially by replacing $`f_1`$ by the identity $`X_oX_o`$) that there exists a Borel function $`g:Y_oX`$ such that the $`\omega `$-support of $`𝒰`$ is $`\{(x,y)X\times Y_o:x=g(y)\}`$.
Similarly, if $`[𝒰𝒰^{}]`$ is abelian, there exists a Borel function $`f:X_oY`$ such that the $`\omega `$-support of $`𝒰`$ is $`\{(x,y)X_o\times Y:f(x)=y\}`$. $`\mathrm{}`$
Remarks (i) The previous Theorem is based on an idea of V.S. Shulman. We take this opportunity to thank him.
(ii) Based on the characterization of Theorem 3.1 a constructive proof of Theorem 3.2 can be given, avoiding the use of the implementability of \*-homomorphisms between $`L^{\mathrm{}}`$-spaces (see ).
(iii) In general, one function is not enough to describe the support of a normalizing masa-bimodule. For an example, consider the von Neumann algebra $`𝒜=M_2()`$, where $``$ is the multiplication algebra of $`L^{\mathrm{}}(0,1)`$. It is not hard to see that the support of $`𝒜`$ is a set of the form $`\{(x,y):f(x)=f(y)\}`$, where $`f`$ is a certain Borel function on $`[0,2]`$ with period $`1`$.
It is easy to see that every finite rank operator in a von Neumann algebra with abelian commutant is a sum of rank one operators in the algebra. Hence, for normalizing masa bimodules Corollary 2.7 can be improved as follows:
###### Proposition 3.4.
Let $`𝒰`$ be an ultraweakly closed normalizing masa bimodule and suppose that $`K`$ is a compact operator in $`𝒰`$. Then $`K`$ can be approximated in the norm topology by sums of rank one operators in $`𝒰`$. Moreover, if $`K𝒞_p`$ then it can be approximated by sums of rank one operators in $`𝒰`$ in the $`p`$-norm topology. Finally, every operator of rank $`n`$ in $`𝒰`$ is the sum of $`n`$ rank one operators in $`𝒰`$.
Next we want to identify the normalizing masa-bimodules which are strongly reflexive. If a rank one operator $`R`$ belongs to the masa-bimodule $`𝒰`$ and if $`E_R`$ and $`F_R`$ are the smallest projections in the masas such that $`F_RR=R`$ and $`RE_R=R`$, then $`F_R(_1,_2)E_R𝒰`$. It was proved in that, if $`𝒰`$ is a strongly reflexive masa-bimodule, the space
$$\text{span }\{F_R(_1,_2)E_R:R𝒰\}$$
is weakly dense in $`𝒰`$. Note that $`(_1,_2)`$ is a strongly reflexive normalizing masa bimodule. More generally, if $`\{E_n\}𝒟_1`$ and $`\{F_n\}𝒟_2`$ are countable families of mutually orthogonal projections, it is easy to verify that the direct sum $`_nF_n(_1,_2)E_n`$ is closed under the triple product, and is clearly a strongly reflexive normalizing masa bimodule. In fact, there are no others:
###### Theorem 3.5.
A strongly reflexive subspace $`𝒰`$ is normalizing if and only if there are countable families $`\{E_n\}`$ and $`\{F_n\}`$ consisting of mutually orthogonal projections, such that
$$𝒰=F_n(_1,_2)E_n.$$
Proof. Let $`𝒰`$ be a normalizing strongly reflexive subspace and let $`𝒰_o`$ be its essential part acting on $`𝒦_1=0_+^{}_1`$. If $`𝒜(𝒦_1)`$ is the von Neumann algebra $`(𝒰_o^{}𝒰_o)^{\prime \prime }`$, we have shown (Lemma 2.12) that $`=Lat𝒜`$ is a totally atomic commutative Boolean lattice. Let $`=\{E_n:n=1,2,\mathrm{}\}`$ be the set of atoms of $``$ considered as projections in $`_1`$ (necessarily countable since $`_1`$ is separable). Since $`E_n=0_+^{}`$ in the strong operator topology, each $`T𝒰`$ can be written $`T=TE_n`$. Setting $`F_n=[𝒰E_n]=\chi (E_n)`$ (where $`\chi =Map(𝒰)`$), we see that the $`F_n`$ are orthogonal since $`\chi `$ is an ortho-map. Since $`TE_n=F_nTE_n`$, it follows that $`T=F_nTE_n`$. On the other hand, for each $`n`$, each $`T`$ $`F_n(_1,_2)E_n`$ satisfies $`\chi (L)^{}TL=0`$ when $`L`$ is some $`E_m`$, and hence when $`L`$ is in $``$. Thus $`F_n(_1,_2)E_n𝒰`$ for each $`n`$, and this completes the proof. $`\mathrm{}`$
The proof of the above theorem actually gives more.
###### Corollary 3.6.
The ultraweak closure of the rank one subspace of a normalizing subspace $`𝒰`$ $`(,𝒦)`$ is of the form $`_n(_n,𝒦_n)`$, for some mutually orthogonal families of subspaces $`_n`$ and $`𝒦_n𝒦`$.
Proof. Obviously the rank one subspace of $`𝒰`$ is also a normalizing space and hence its ultraweak closure is a reflexive (Theorem 2.4) normalizing masa bimodule. Thus the last theorem applies. $`\mathrm{}`$
According to , a masa bimodule is strongly reflexive if and only if its $`\omega `$-support $`\kappa `$ is marginally equivalent to a countable union of Borel rectangles. In the terminology of , $`\kappa `$ is the $`\omega `$-closure of its $`\omega `$-interior. In this terminology, Corollary 3.6 says that the $`\omega `$-interior of the $`\omega `$-support of a normalizing masa-bimodule can be written as a countable union of Borel rectangles $`\alpha _n\times \beta _n`$, such that the families $`\{\alpha _n\}`$ and $`\{\beta _n\}`$ consist of disjoint Borel sets.
Let us recall that in it was shown that although the rank one subspace of a general strongly reflexive masa-bimodule is dense in the bimodule in the weak operator topology, it need not be dense in the ultraweak topology. By Theorem 3.5, this cannot occur for normalizing masa-bimodules.
###### Corollary 3.7.
If a normalizing subspace $`𝒰`$ is strongly reflexive, then its rank one subspace is ultraweakly dense in $`𝒰`$.
## 4. Normalizing masa-bimodules and synthesis
Now we turn our attention to the question of spectral synthesis. Spectral synthesis for operator algebras was introduced by Arveson . It can be generalized for masa-bimodules as follows. If $`_1=𝒫(𝒟_1),_2=𝒫(𝒟_2)`$ are the projection lattices of two masas, a map $`\phi (_1,_2)`$ (a *commutative subspace map* in the terminology of ) is said to be synthetic, if the only ultraweakly closed masa-bimodule $`𝒮`$ with the property $`Map𝒮=\phi `$ is $`Op\phi `$. There is a non trivial fact hidden behind this definition. As was proved by Arveson (in fact, for the case of CSL algebras, but it easily follows for masa-bimodules as well ), given $`\phi (_1,_2)`$ there exists an ultraweakly closed masa-bimodule $`_{min}`$, minimal with respect to the property that its reflexive hull equals $`Op\phi `$. Thus a reflexive masa-bimodule $``$ has synthetic map if and only if $`=_{min}`$. Von Neumann algebras with abelian commutant have this property. The same holds for their generalization, normalizing masa-bimodules.
###### Theorem 4.1.
Commutative subspace ortho-maps are synthetic.
This theorem will follow from a general fact about masa-bimodules (Proposition 4.2 below), which states that for every family of ultraweakly closed masa-bimodules with the same reflexive hull, there is a certain natural normalizing masa-bimodule contained in each member of the family. This bimodule corresponds to the diagonal of a CSL algebra.
###### Proposition 4.2.
Let $`𝒰=Op\phi (_1,_2)`$ be a reflexive masa bimodule, and let $`𝒰_0=Op\phi Op\phi ^{}`$. Then $`𝒰_0`$ is contained in $`𝒰_{min}`$, the minimal ultraweakly closed masa-bimodule with reflexive cover $`𝒰`$ .
Proof. We consider the Hilbert space $`=_2_1`$ and the set of projections
$$𝒮=\{\phi (L)L:L𝒮_{1\phi }\}$$
on it. We put $`𝒜=Alg𝒮_{1\phi }`$, $`=Alg\phi (𝒮_{1\phi })=Alg𝒮_{2\phi }`$ and $`𝒱=Op\phi ^{}`$.
It is an easy verification to show that $`Alg𝒮=\left(\begin{array}{cc}& 𝒰\\ 𝒱^{}& 𝒜\end{array}\right)`$. A similar calculation shows also that the diagonal $`𝒞𝒞^{}`$ of $`𝒞=Alg𝒮`$ (i.e. the commutant of $`𝒮`$) is $`𝒮^{}=\left(\begin{array}{cc}𝒮_{2\phi }^{}& 𝒰_0\\ 𝒰_0^{}& 𝒮_{1\phi }^{}\end{array}\right)`$. But Arveson has shown that, for a CSL algebra $`𝒞`$, the diagonal $`𝒞𝒞^{}`$ is contained in $`𝒞_{\mathrm{min}}`$. Therefore, to show that $`𝒰_0𝒰_{\mathrm{min}}`$, it suffices to prove that $`𝒞_{\mathrm{min}}`$ is contained in
$$=\left(\begin{array}{cc}_{\mathrm{min}}\hfill & 𝒰_{\mathrm{min}}\hfill \\ 𝒱_{\mathrm{min}}^{}\hfill & 𝒜_{\mathrm{min}}\hfill \end{array}\right).$$
Since $``$ is a w\*-closed subspace of $`(H)`$ containing the masa $`𝒟_2𝒟_1`$, to show that $`𝒞_{\mathrm{min}}`$ it suffices to show that $`[\xi ]=[𝒞\xi ]`$ for all $`\xi `$ (see Theorem 22.19 of ). But this can be immediately verified using the facts that $`=Ref(_{\mathrm{min}})`$, $`𝒰=Ref(𝒰_{\mathrm{min}})`$, $`𝒱=Ref(𝒱_{\mathrm{min}})`$ and $`𝒜=Ref(𝒜_{\mathrm{min}}).\mathrm{}`$
Proof of Theorem 4.1. Suppose that $`\chi `$ is a commutative subspace map, which is also an ortho-map. The space $`𝒰=Op\chi `$ is normalizing. But then $`𝒰=Op\chi Op\chi ^{}`$ (Theorem 2.4), hence $`𝒰𝒰_{min}`$ by Proposition 4.2. Thus $`𝒰=𝒰_{\mathrm{min}}`$, which means that $`\chi `$ is synthetic. $`\mathrm{}`$
###### Corollary 4.3.
If the map of an ultraweakly closed masa-bimodule $`𝒰`$ is an ortho-map, then $`𝒰`$ is normalizing.
Proof. Since $`𝒰`$ is a masa-bimodule, its map $`\chi `$ is commutative, hence by Theorem 4.1 it is synthetic. Hence $`𝒰=Ref𝒰=Op\chi `$. But since $`\chi `$ is an ortho-map, $`Op\chi `$ must be normalizing (Theorem 2.4). $`\mathrm{}`$
We conclude this section with another result on synthesis. Let us call an ultraweakly closed masa bimodule *synthetic* if its map is synthetic. Thus a synthetic masa-bimodule is automatically reflexive. As we will easily see, normalizing masa-bimodules are not only synthetic, but in a sense hereditarily synthetic.
###### Proposition 4.4.
Let $`𝒰`$ be an ultraweakly closed normalizing masa bimodule and let $`𝒮𝒰`$ be an ultraweakly closed masa-bimodule. If $`𝒮`$ is a $`𝒰^{}𝒰`$-submodule (or a $`𝒰𝒰^{}`$-submodule), then $`𝒮`$ is synthetic.
Proof. It suffices to show that $`𝒮`$ is normalizing. Let $`𝒰=Op\phi `$, $`_1=𝒰^{}𝒰`$ and $`_2=𝒰𝒰^{}`$. Suppose that $`R,S,T𝒮`$. Then $`S,T𝒰`$ and hence $`S^{}T_1`$. If $`𝒮_1𝒮`$, it follows that $`RS^{}T𝒮`$. If $`_2𝒮𝒮`$, we have $`RS^{}_2`$ and hence $`RS^{}T𝒮`$ again. $`\mathrm{}`$
An immediate application of the last proposition is the fact that, if $`𝒜`$ is a von Neumann algebra with abelian commutant, every ultraweakly closed left or right ideal of $`𝒜`$ is synthetic. Note that this can also be inferred from the results of .
## 5. Normalizers of reflexive algebras
In this section we present the relation between normalizing reflexive subspaces and normalizers of reflexive algebras of operators. Let $`𝒜`$ and $``$ be reflexive algebras of operators on $`_1`$ and $`_2`$ respectively. An operator $`T(_1,_2)`$ is called a semi-normalizer of $``$ into $`𝒜`$ if
$$T^{}T𝒜.$$
A normalizer of the algebra $``$ into the algebra $`𝒜`$ is a semi-normalizer of $``$ into $`𝒜`$ whose adjoint is a semi-normalizer of $`𝒜`$ into $``$. We denote by $`SN(,𝒜)`$ the set of semi-normalizers of $``$ into $`𝒜`$ and by $`N(,𝒜)`$ the set of normalizers of $``$ into $`𝒜`$.
In it is shown that the set of normalizers of a nest algebra into itself is closed in the weak operator topology. We show that, for arbitrary reflexive algebras $`𝒜`$ and $``$, the sets $`SN(,𝒜)`$ and $`N(,𝒜)`$ are closed in the strong operator topology, but not always in the weak operator topology; nevertheless, Coates’ result remains valid whenever the algebras $`𝒜`$ and $``$ are strongly reflexive.
###### Proposition 5.1.
If $`𝒜`$ and $``$ are reflexive algebras, the sets $`SN(,𝒜)`$ and $`N(,𝒜)`$ are strongly closed. If $``$ is strongly reflexive (resp. $`𝒜`$ and $``$ are strongly reflexive) then $`SN(,𝒜)`$ (resp. $`N(,𝒜)`$) is weakly closed. The sets $`SN(I,I)`$ and $`N(I,I)`$ are not weakly closed.
Proof. Suppose that $`T_\nu T`$ strongly and that $`T_\nu SN(,𝒜)`$. Then, for each operator $`B`$ and vectors $`x`$ and $`y`$, $`BT_\nu x,T_\nu yBTx,Ty`$. This means, of course, that $`T_\nu ^{}BT_\nu T^{}BT`$ in the weak operator topology. Since $`T_\nu ^{}BT_\nu 𝒜`$ for each $`\nu `$ and $`𝒜`$ is weakly closed, being reflexive, it follows that $`T^{}BT𝒜`$. The proof that $`N(,𝒜)`$ is strongly closed is similar.
Now suppose that $`T_\nu T`$ weakly and that $`T_\nu SN(,𝒜)`$. First note that for each *finite rank* operator $`F`$, $`T_\nu ^{}FT_\nu T^{}FT`$ weakly (indeed, writing $`F=F_1^{}F_2`$ where $`F_1,F_2`$ have finite rank, we have $`F_iT_\nu F_iT`$ strongly for $`i=1,2`$). Thus $`T^{}FT𝒜`$. Now if $`B`$ is arbitrary and $`x_2`$, by strong reflexivity there is a net $`(F_i)`$ of finite rank operators in $``$ such that $`F_iTxBTx`$ and so $`T^{}F_iTxT^{}BTx`$. Since each $`T^{}F_iT`$ is in $`𝒜`$ and $`𝒜`$ is reflexive, it follows that $`T^{}BT𝒜`$. This shows that $`SN(,𝒜)`$ is weakly closed. Therefore, if $`𝒜`$ is also strongly reflexive, the same holds for $`N(,𝒜)=SN(,𝒜)(SN(𝒜,))^{}`$.
The final statement is a consequence of the fact that the sets of isometries and unitaries are not weakly closed. For a specific example, we construct a sequence $`T_nN(I,I)`$ converging weakly to an operator $`T`$ which is not in $`SN(I,I)`$. Let $`\{e_k:k\}\{e\}`$ be an orthonormal basis of $``$ and let $`U`$ be the bilateral shift defined on $`_o=\overline{[e_k:k]}`$ by $`Ue_k=e_{k+1}`$. Define $`T_n=U^nI()`$ and $`T=0I()`$. Each $`T_n`$ is a unitary operator on $``$, and so $`T_n^{}(I)T_nI`$ and $`T_n(I)T_n^{}I`$, i.e. $`T_nN(I,I)`$. Since $`U^n0`$ weakly, $`T_nT`$ weakly. But $`T^{}(I)T=[T]⫅̸I.\mathrm{}`$
Let $`_1=Lat𝒜`$ and $`_2=Lat`$. If $`\phi (_1,_2)`$, we put
$$𝒰_\phi =\{T(_1,_2):TL=\phi (L)T\text{ for all }L_1\}$$
which is a reflexive normalizing subspace, and also a bimodule over the diagonals $`^{}`$ and $`𝒜𝒜^{}`$, as is readily verified.
###### Theorem 5.2.
The set $`SN(,𝒜)`$ of semi-normalizers of $``$ into $`𝒜`$ is a union of reflexive normalizing subspaces. More precisely,
$$SN(,𝒜)=\{𝒰_\phi :\phi (_1,_2)\}.$$
Proof. Suppose that $`\phi (_1,_2)`$ and $`T𝒰_\phi `$. Then, for each $`L_1`$ and $`B`$,
$$L^{}T^{}BTL=L^{}T^{}B\phi (L)T=L^{}T^{}\phi (L)B\phi (L)T=L^{}LT^{}B\phi (L)T=0$$
since $`\phi (L)_2`$, so that $`T^{}BT𝒜`$, which shows that $`T`$ is a semi-normalizer of $``$ into $`𝒜`$. Thus $`𝒰_\phi SN(,𝒜)`$.
Conversely, let $`TSN(,𝒜)`$. Define $`\phi (_1,_2)`$ by
$$\phi (L)=\overline{[T(L)]},L_1.$$
It is obvious that $`\phi (L)_2`$. It is also easy to see that $`\phi `$ is 0-preserving and join-continuous. We will show that $`T𝒰_\phi `$. Let $`L_1`$. Since $`T(L_1)T(L_1)`$, it is clear that $`\phi (L)^{}TL=0`$. On the other hand, for each $`B`$ we have $`T^{}BT𝒜`$ and so $`L^{}T^{}BTL=0`$; hence $`BTLx,TL^{}y=0`$ for each $`x,y`$. Since the closure of $`\{BTLx:B,x_1\}`$ is $`\phi (L)(_2)`$, it follows that $`\phi (L)(_2)TL^{}(_1)`$. Thus $`\phi (L)z,TL^{}y=0`$ for all $`y_1,z_2`$ and so $`\phi (L)TL^{}=0`$. Adding the relation $`\phi (L)TL=TL`$, we obtain $`\phi (L)T=TL`$ so that $`T𝒰_\phi `$ as required. $`\mathrm{}`$
It is obvious that the intersection of several normalizing reflexive subspaces is again a normalizing reflexive subspace. If, for each $`\psi (_2,_1)`$, we put
$$𝒱_\psi =\{S(_2,_1):SM=\psi (M)S\text{ for all }M_2\},$$
the above theorem gives us the following
###### Corollary 5.3.
The set $`N(,𝒜)`$ of normalizers of $``$ into $`𝒜`$ is a union of reflexive normalizing subspaces. More precisely,
$$N(,𝒜)=\{𝒰_\phi 𝒱_\psi ^{}:\phi (_1,_2),\psi (_2,_1)\}.$$
This corollary was proved in for the case of nest algebras.
Recall (Proposition 2.6) that w\*-closed normalizing subspaces are generated in norm by their partial isometries. Thus, the previous results have the following consequence:
###### Corollary 5.4.
Any seminormalizer (resp. normalizer) of $``$ into $`𝒜`$ is the norm-limit of linear combinations of partial isometries in $`SN(,𝒜)`$ (resp. in $`N(,𝒜)`$).
Having shown that the set of normalizers between reflexive algebras is a union of normalizing spaces, we now turn to the converse question: when does a normalizing reflexive space $`𝒰`$ consist of semi-normalizers or normalizers of two reflexive algebras $`𝒜`$ and $``$? For example, it is easily verified that every operator in $`𝒰`$ is a normalizer of the algebra $`Alg𝒮_{2\chi }`$ into the algebra $`Alg𝒮_{1\chi }`$ (where $`\chi =Map𝒰`$).
We write $`𝒜_d=𝒜𝒜^{}`$and $`_d=^{}`$. If $`𝒦_1=𝒰^{}(_2)=0_+^{}_1`$ and $`𝒦_2=𝒰(_1)=I_{}_2`$, we denote by $`𝒜_o`$ (resp. $`_o`$) the compression of $`𝒜`$ (resp. $``$) to $`𝒦_1`$ (resp. $`𝒦_2`$); we write $`𝒰_o(𝒦_1,𝒦_2)`$ for the restriction of $`𝒰`$ to $`𝒦_1`$ and $`\chi _o`$ for the map of $`𝒰_o`$.
###### Lemma 5.5.
The following are equivalent:
(1) $`𝒰^{}𝒰𝒜`$.
(2) For each $`LLat𝒜_d`$, $`\chi (L)\chi (L^{})`$.
(3) For each $`LLat𝒜`$, $`\chi (L)\chi (L^{})`$.
(4) With respect to the decomposition $`_1=𝒦_1𝒦_1^{}`$, each $`LLat𝒜`$ decomposes as $`L=L_1L_2`$, where $`L_1𝒮_{1\chi _o}`$.
###### Proof.
Since $`𝒰^{}𝒰𝒜`$ is equivalent to $`𝒰^{}𝒰𝒜_d`$ and hence to $`Lat𝒜_dLat𝒰^{}𝒰`$ and to $`Lat𝒜Lat𝒰^{}𝒰`$, the Lemma is an immediate application of Theorem 2.10, since $`Lat𝒰^{}𝒰=\{L𝒫_1:\chi (L)\chi (L^{})\}=\{L_1L_2:L_1𝒮_{1\chi _o}\}.`$ $`\mathrm{}`$
###### Theorem 5.6.
Let $`𝒰(_1,_2)`$ be a w\*-closed normalizing space. The following are equivalent:
(1) $`𝒰SN(,𝒜)`$.
(2) (i) $`\chi (L)\chi (L^{})`$ for each $`LLat𝒜`$ and (ii) $`\chi _o(Lat𝒜_o)Lat_o`$.
(3) (i) $`\chi (L)\chi (L^{})`$ for each $`LLat𝒜`$ and (ii) for each $`LLat𝒜`$, there exists a projection $`Q`$ such that $`\overline{[\chi (L)]}=\chi (L)|_{𝒦_1}Q`$.
###### Proof.
(1)$``$(3) Suppose (1) holds. Then $`𝒰^{}𝒰𝒜`$, and (i) holds by the Lemma. For all $`T,S𝒰`$, $`B`$ and $`LLat𝒜`$ we have
$$BSLx,TL^{}y=0(x,y_1).$$
Thus $`\overline{[\chi (L)]}\chi (L^{})^{}=(\chi (L)^{}\chi (I))^{}=\chi (L)\chi (I)^{}`$. Writing $`L=L_1L_2`$ where $`L_1𝒮_{1\chi _o}`$ we have $`\chi (L)=\chi _o(L_1)0`$. Since $`\chi (L)\overline{[\chi (L)]}`$ we obtain
$$\chi _o(L_1)0\overline{[\chi (L)]}\chi _o(L_1)I$$
which shows that $`\overline{[\chi (L)]}=\chi _o(L_1)Q`$, for some projection $`Q`$, as required.
These steps can clearly be reversed.
(1)$``$(2) Again (i) holds by the Lemma. For (ii), note first that $`𝒰_oSN(_o,𝒜_o)`$. Thus for each $`LLat𝒜_o`$, $`S,T𝒰_o`$ and $`B_o`$ we have $`L^{}T^{}BSL=0`$. As in the proof of (1)$``$(3), we conclude that $`\chi _o(L)\overline{[_o\chi _o(L)]}\chi _o(L^{})^{}=\chi _o(L)`$ since $`\chi _o(I)=I`$. Thus $`\chi _o(L)=\overline{[_o\chi _o(L)]}Lat_o`$.
(2)$``$(1) Given $`B`$ and $`T𝒰`$, we will prove that $`T^{}BT𝒜`$, equivalently that $`L^{}T^{}BTL=0`$ for each $`LLat𝒜`$. Since $`𝒰^{}𝒰𝒜`$, we have $`L=L_1L_2`$ where $`L_1𝒮_{1\chi _o}`$ by the Lemma. It is easily seen that $`L_1Lat𝒜_o`$.
Since $`T𝒰`$, we have $`T(L^{}_1)\chi (L^{})_2\chi (L)^{}_2`$ since $`\chi (L^{})\chi (L)`$ by the Lemma. Thus $`\chi (L)TL^{}=0`$. Since also $`\chi (L)^{}TL=0`$, it follows that $`TL=\chi (L)T`$ and $`TL^{}=\chi (L)^{}\chi (I)T`$ since the range of $`T`$ is contained in $`\chi (I)`$. Thus $`L^{}T^{}BTL=T^{}\chi (L)^{}\chi (I)B\chi (I)\chi (L)T`$. But $`\chi (I)B|_{𝒦_2}_o`$ and since $`\chi (L)|_{𝒦_2}=\chi _o(L_1)Lat_o`$ by assumption we have $`(\chi (I)B\chi (I))\chi (L)=\chi (L)(\chi (I)B\chi (I))\chi (L)`$ and so $`L^{}T^{}BTL=0`$. This shows that $`T^{}BT𝒜`$ and concludes the proof. $`\mathrm{}`$
###### Corollary 5.7.
Let $`𝒰(_1,_2)`$ be an essential normalizing ultraweakly closed subspace and $`\chi =Map𝒰`$. Then $`𝒰SN(,𝒜)`$ if and only if
*(i)* $`\chi (L)\chi (L^{})`$ for each $`LLat𝒜`$ and
*(ii)* $`\chi (Lat𝒜)Lat`$.
Proof. Immediate from Theorem 5.6. $`\mathrm{}`$
###### Corollary 5.8.
Let $`𝒰(_1,_2)`$ be a normalizing ultraweakly closed subspace and let $`\chi =Map𝒰`$ and $`\chi ^{}=Map𝒰^{}`$. Then $`𝒰N(,𝒜)`$ if and only if
*(i)* $`\chi (L)\chi (L^{})`$ for each $`LLat𝒜`$ while $`\chi ^{}(M)\chi ^{}(M^{})`$ for each $`MLat`$ and
*(ii)* $`\chi _o(Lat𝒜_o)=Lat_o`$.
Proof. If $`𝒰N(,𝒜)`$, then $`\chi _o(Lat𝒜_o)Lat_o`$ and $`\chi _o^{}(Lat_o)Lat𝒜_o`$ by Theorem 5.6. But $`𝒰_o^{}𝒰_o𝒜_o`$ and hence $`Lat𝒜_oLat(𝒰_o^{}𝒰_o)=𝒮_{1\chi _o}`$ (Theorem 2.10). It follows that $`\chi _o`$ is one-to-one on $`Lat𝒜_o`$ into $`Lat_o`$. Similarly, $`Lat_oLat(𝒰_o𝒰_o^{})=𝒮_{2\chi _o}`$ and hence the inverse of $`\chi _o`$, namely $`\chi _o^{}`$, maps $`Lat_o`$ 1-1 into $`Lat𝒜_o`$.
Suppose conversely that *(i)* and *(ii)* hold. By Theorem 5.6, $`𝒰SN(,𝒜)`$. As above, *(i)* gives $`Lat𝒜_o𝒮_{1\chi _o}`$ and hence $`\chi _o^{}\chi _o|_{Lat𝒜_o}=id|_{Lat𝒜_o}`$ from Theorem 2.10. Therefore $`\chi ^{}(Lat_o)Lat𝒜_o`$ and so $`𝒰^{}SN(𝒜,)`$ again from Theorem 5.6. $`\mathrm{}`$
###### Corollary 5.9.
Let $`𝒰(_1,_2)`$ be an essential normalizing ultraweakly closed subspace and let $`\chi =Map𝒰`$. Then $`𝒰N(,𝒜)`$ if and only if
*(i)* $`\chi (L)\chi (L^{})`$ for each $`LLat𝒜`$ and
*(ii)* $`\chi (Lat𝒜)=Lat`$.
Proof. From Corollary 5.8 it is sufficient to show that $`(i)`$ and $`(ii)`$ imply that $`\chi ^{}(M)\chi ^{}(M^{})`$ for each $`MLat_d`$. But this is immediate from Lemma 5.5 (applied to $`𝒰^{}`$) since from $`(ii)`$ we have $`Lat=\chi (Lat𝒜)𝒮_2=Lat(𝒰𝒰^{})`$ and so $`𝒰𝒰^{}`$. $`\mathrm{}`$
Note that Theorem 5.2 yields, for every $`TSN(,𝒜)`$, a certain normalizing subspace $`𝒰_\phi `$, such that $`T𝒰_\phi SN(,𝒜)`$. We show that this property extends from single operators to *linear spaces* consisting of seminormalizers. Recall that $`𝒰_\phi `$ is a $`_d,𝒜_d`$-bimodule, where $`𝒜_d`$ and $`_d`$ are the diagonals of $`𝒜`$ and $``$. It may be interesting to note that all right $`𝒜_d`$-modules $`𝒰`$ which consist of semi-normalizers of $``$ into $`𝒜`$ automatically satisfy a rather strong condition.
###### Proposition 5.10.
(1) Any *linear space* $`𝒰\mathrm{S}\mathrm{N}(,𝒜)`$ (resp. $`𝒰\mathrm{N}(,𝒜)`$) is contained in an ultraweakly closed normalizing space $`𝒰_\mathrm{A}\mathrm{S}\mathrm{N}(,𝒜)`$ (resp. $`𝒰_\mathrm{A}\mathrm{N}(,𝒜)`$) which is a right $`𝒜_\mathrm{d}`$-module.
(2) If $`𝒰\mathrm{S}\mathrm{N}(,𝒜)`$ is a right $`𝒜_\mathrm{d}`$-module, then $`𝒰`$ is automatically normalizing. Moreover, $`0_+=\mathrm{ker}𝒰`$ is in the centre of $`𝒜_\mathrm{d}`$ and the restriction $`𝒜_{\mathrm{o}\mathrm{d}}`$ of $`𝒜_\mathrm{d}`$ to $`𝒦_1=0_+_1`$ is the bicommutant of $`(𝒰^{}𝒰)|_{𝒦_1}`$. It follows that $`𝒮_{1\mathrm{\chi }}=\{\mathrm{L}\mathrm{I}:\mathrm{L}Lat𝒜_{\mathrm{o}\mathrm{d}}\}`$.
###### Proof.
(1) Suppose $`𝒰SN(,𝒜)`$ is a linear space. Define $`𝒰_a=[𝒰𝒜_d]`$ and $`𝒰_A=\overline{[𝒰𝒜_d]}^{uw}`$. If $`T_1,T_2𝒰`$ and $`A_1,A_2𝒜_d`$, then for each $`B`$ the fact that $`T_i^{}BT_j𝒜`$ and $`A_i,A_i^{}𝒜`$ yields $`(T_1A_1+T_2A_2)^{}B(T_1A_1+T_2A_2)𝒜`$. Thus $`𝒰_aSN(,𝒜)`$.
Now $`𝒰_a𝒜_d𝒰_a`$ by construction and $`𝒰_a^{}𝒰_a𝒜`$ since $`𝒰_aSN(,𝒜)`$. Therefore $`𝒰_a`$ is normalizing (Remark 2.2).
It follows from Theorem 2.4 that $`𝒰_A`$ is a reflexive normalizing space. Also, since $`𝒰_A`$ is the strong operator closure of $`𝒰_a`$ and $`SN(,𝒜)`$ is strongly closed, $`𝒰_ASN(,𝒜)`$. That the strong closure of a right $`𝒜_d`$-module is a right $`𝒜_d`$-module is obvious.
The case $`𝒰N(,𝒜)`$ is similar.
(2) That $`𝒰`$ is normalizing follows from Remark 2.2 as above. Let $`\chi =Map𝒰`$. Since $`𝒰𝒜_d𝒰`$ we have $`𝒮_{1\chi }Lat𝒜_d`$ (Lemma 1.1). On the other hand since $`𝒰SN(,𝒜)`$ we have $`𝒰^{}𝒰𝒜_d`$. Thus $`0_+`$ commutes with $`𝒜_d`$ and also $`0_+(𝒰^{}𝒰)^{\prime \prime }𝒜_d`$, so $`0_+`$ is in the centre of $`𝒜_d`$. Hence $`𝒜_d`$ can be written $`𝒜_{od}𝒜_{1d}`$ with respect to the decomposition $`_1=𝒦_1𝒦_1^{}`$. Using Theorem 2.10, we have
$$(𝒰^{}𝒰)^{\prime \prime }=\left(\begin{array}{cc}(𝒮_{1\chi }|_{𝒦_1})^{}& 0\\ 0& I\end{array}\right)𝒜_d𝒮_{1\chi }^{}=\left(\begin{array}{cc}(𝒮_{1\chi }|_{𝒦_1})^{}& 0\\ 0& (𝒦_1^{})\end{array}\right)$$
This shows that $`𝒜_{od}`$ equals $`(𝒮_{1\chi }|_{𝒦_1})^{}=(𝒰^{}𝒰|_{𝒦_1})^{\prime \prime }`$ and $`Lat𝒜_{od}`$ equals $`𝒮_{1\chi }|_{𝒦_1}=Lat(𝒰^{}𝒰|_{𝒦_1})`$. Thus $`𝒮_{1\chi }=\{LI:LLat(𝒰^{}𝒰|_{𝒦_1})\}=\{LI:LLat𝒜_{od}\}.\mathrm{}`$
It is clear that every linear space $`𝒰SN(,𝒜)`$ is also contained in a w\*-closed normalizing space which is a $`(_d,𝒜_d)`$-bimodule (just consider the w\*-closure of $`[_d𝒰𝒜_d]`$). This bimodule is not necessarily maximal with respect to being a linear space of seminormalizers.
For an example, consider the nest algebra $`𝒜`$ of all upper triangular $`2\times 2`$ matrices and let $`𝒰`$ be the space of all strictly lower triangular matrices. One easily checks that $`𝒰SN(𝒜,𝒜)`$ and that $`𝒰`$ is a bimodule over the diagonal algebra (which is a masa in this case). However, it is readily verified that the linear span of the matrix unit $`E_{11}`$ and $`𝒰`$ is also contained in $`SN(𝒜,𝒜)`$.
###### Remark 5.11.
Let $`𝒰SN(,𝒜)`$ be an essential w\*-closed normalizing space, which is a right $`𝒜_d`$-module. Then $`𝒰`$ is maximal with respect to being a linear space in $`SN(,𝒜).`$
###### Proof.
Let $`𝒱SN(,𝒜)`$ be a linear space containing $`𝒰`$ and let $`T𝒱`$. For each $`S𝒰`$, we have $`S^{}T𝒱^{}𝒱𝒜_d`$. Since $`𝒰𝒜_d𝒰`$, if $`\chi =Map𝒰`$ we have $`𝒮_{1\chi }Lat𝒜_d`$ (Lemma 1.1). Thus $`L^{}S^{}TL=0`$ for each $`L𝒮_{1\chi }`$ and hence $`TLx,SL^{}y=0`$ for all $`x,y`$. But the closure of $`\{SL^{}y:S𝒰,y_1\}`$ is $`\chi (L^{})_2`$ which equals $`\chi (L)^{}_2`$ since $`\chi `$ is essential; therefore $`TL(_1)\chi (L)^{}_2`$. We have shown that $`\chi (L)^{}TL=0`$ for all $`L𝒮_{1\chi }`$ and so $`T𝒰`$. $`\mathrm{}`$
Suppose now that $`𝒜`$ and $``$ are CSL algebras, that is, the respective invariant subspace lattices are commutative. In this case Proposition 3.4 and Theorem 4.1 immediately yield the next corollary. The last statement was proved by Coates for the case of nest algebras.
###### Corollary 5.12.
If $`𝒜`$ and $``$ are CSL algebras, the set $`SN(,𝒜)`$ ($`N(,𝒜)`$) of semi-normalizers (normalizers) of $``$ into $`𝒜`$ is a union of synthetic normalizing masa-bimodules. Each compact operator $`K`$ in $`SN(,𝒜)`$ ($`N(,𝒜)`$) can be approximated in norm by sums of rank one operators in $`SN(,𝒜)`$ ($`N(,𝒜)`$). Moreover, if $`K𝒞_p`$, then it can be approximated in the $`𝒞_p`$-norm by sums of rank one operators in $`SN(,𝒜)`$ ($`N(,𝒜)`$). Finally, if $`K`$ has finite rank, say $`n`$, it can be written as a sum of $`n`$ rank one operators in $`SN(,𝒜)`$ ($`N(,𝒜)`$).
We would like to conclude this paper with a discussion of the behaviour of the set of semi-normalizers (normalizers) with respect to addition. It is clear that, if $`T`$ and $`S`$ are semi-normalizers (normalizers) of an algebra into another, the sum $`T+S`$ is not necessarily a semi-normalizer (normalizer). Proposition 5.14 below gives a necessary and sufficient condition for this to happen, when the algebras in question are CSL algebras. First we prove a statement, concerning arbitrary masa-bimodules.
###### Lemma 5.13.
Let $`\mathrm{\Gamma }:(_1,_2)\{0,1\}`$ be such that, if $`\mathrm{\Gamma }(T)=1`$ for some operator $`T`$, then there is a reflexive masa-bimodule $`𝒰`$, containing $`T`$, such that $`\mathrm{\Gamma }(S)=1`$ for all $`S𝒰`$. Suppose that $`T_1`$ and $`T_2`$ are operators, such that $`\mathrm{\Gamma }(\lambda _1T_1+\lambda _2T_2)=1`$ for all real numbers $`\lambda _1`$ and $`\lambda _2`$. Then there exists a reflexive masa-bimodule $`𝒰`$, containing $`T_1`$ and $`T_2`$, such that $`\mathrm{\Gamma }(T)=1`$ for all $`T𝒰`$.
###### Proof.
Represent $`_1`$ as $`L^2(X,\mu )`$ and $`_2`$ as $`L^2(Y,\nu )`$, where $`(X,\mu )`$ and $`(Y,\nu )`$ are compact metric spaces with regular Borel measures, and let $`𝒟_1`$ and $`𝒟_2`$ be the respective multiplication algebras. By the support $`suppT`$ of an operator $`T`$ we will mean the (closed) support of the reflexive $`𝒟_2,𝒟_1`$-bimodule generated by $`T`$. Let $`suppT_1=\kappa _1`$, $`suppT_2=\kappa _2`$ and $`\kappa =\kappa _1\kappa _2`$. The sets $`\kappa _1,\kappa _2`$ and $`\kappa `$ are compact subsets of $`X\times Y`$. It suffices to find a real number $`\lambda `$ such that the operator $`T_1+\lambda T_2`$ has support $`\kappa `$. Indeed, in this case, there exists a reflexive $`𝒟_2,𝒟_1`$-bimodule $`𝒰`$ containing $`T_1+\lambda T_2`$ such that $`\mathrm{\Gamma }(T)=1`$ for all $`T𝒰`$. Since $`\kappa supp𝒰`$ it follows from , Theorem 4.6 that $`T_1,T_2𝒰`$.
Put $`\kappa _\lambda =supp(T_1+\lambda T_2)`$, $`\upsilon _\lambda =\kappa _\lambda ^c`$. It is clear that $`\upsilon _\lambda `$ is an open subset of $`X\times Y`$. We claim that, if $`\lambda \lambda ^{}`$ are nonzero, then $`\upsilon _\lambda \upsilon _\lambda ^{}\kappa ^c`$. Indeed whenever $`\alpha `$ and $`\beta `$ are open and such that $`\alpha \times \beta \upsilon _\lambda \upsilon _\lambda ^{}`$, then $`F(\beta )(T_1+\lambda T_2)E(\alpha )=0`$ and also $`F(\beta )(T_1+\lambda ^{}T_2)E(\alpha )=0`$. Thus $`(\lambda \lambda ^{})F(\beta )T_2E(\alpha )=0`$ and hence $`F(\beta )T_2E(\alpha )=0`$, so that $`\alpha \times \beta \kappa _2=\mathrm{}`$. Repeating the argument to the operators $`\frac{1}{\lambda }T_1+T_2`$ and $`\frac{1}{\lambda ^{}}T_1+T_2`$ gives $`\alpha \times \beta \kappa _1=\mathrm{}`$. It follows that $`\upsilon _\lambda \upsilon _\lambda ^{}\kappa _2=\mathrm{}`$ and $`\upsilon _\lambda \upsilon _\lambda ^{}\kappa _1=\mathrm{}`$, so that $`\upsilon _\lambda \upsilon _\lambda ^{}\kappa ^c`$.
Thus the union $`\{\upsilon _\lambda \kappa :\lambda \backslash \{0\}\}`$ is an uncountable union of (relatively) open disjoint subsets of $`\kappa `$. Since $`\kappa `$ is a compact metric space, no more than countably many of them can be nonempty. Thus there exists $`\lambda \backslash \{0\}`$ so that $`\upsilon _\lambda \kappa =\mathrm{}`$ or $`\kappa \kappa _\lambda `$. But $`\kappa _\lambda \kappa `$. Indeed if $`(\alpha \times \beta )\kappa =\mathrm{}`$ with $`\alpha `$ and $`\beta `$ open, then $`F(\beta )T_iE(\alpha )=0`$ for $`i=1,2`$ so $`F(\beta )(T_1+\lambda T_2)E(\alpha )=0`$ which means that $`(\alpha \times \beta )\kappa _\lambda =\mathrm{}`$. $`\mathrm{}`$
###### Proposition 5.14.
Let $`𝒜`$ and $``$ be CSL algebras and $`T,SSN(,𝒜)`$ (resp. $`T,SN(,𝒜)`$). Then $`T+SSN(,𝒜)`$ ($`T+SN(,𝒜)`$) if and only if there is a reflexive normalizing masa-bimodule $`𝒰SN(,𝒜)`$ ($`𝒰N(,𝒜)`$) such that $`T,S𝒰`$.
###### Proof.
We will consider the case of semi-normalizers, the case of normalizers is similar. It is clear that if there exists a reflexive normalizing masa-bimodule $`𝒰SN(,𝒜)`$ such that $`T,S𝒰`$, then $`T+S`$ is a semi-normalizer. Conversely, suppose that $`T,S`$ and $`T+S`$ are semi-normalizers of $``$ into $`𝒜`$. An elementary computation shows that then $`T^{}BS+S^{}BT𝒜`$ for each $`B`$. But then it is easy to see that $`\lambda _1T+\lambda _2S`$ is a semi-normalizer for all real numbers $`\lambda _1`$ and $`\lambda _2`$. From Theorem 5.2 we have that, if $`TSN(,𝒜)`$, then there exists a reflexive space $`𝒰SN(,𝒜)`$, containing $`T`$. The space $`𝒰`$ is a masa bimodule, since the semi-lattices of its map are commutative. Thus Lemma 5.13 applies with the property of being a semi-normalizer of $``$ into $`𝒜`$ in the place of $`\mathrm{\Gamma }`$. $`\mathrm{}`$
###### Corollary 5.15.
Let $`𝒜`$ and $``$ be CSL algebras and $`T,S`$ be semi-normalizers (normalizers) of $``$ into $`𝒜`$, such that $`T+S`$ is also a semi-normalizer (normalizer). Then $`B_1TA_1+B_2SA_2`$ is a semi-normalizer (normalizer) for every $`B_1,B_2^{}`$ and $`A_1,A_2𝒜𝒜^{}`$.
Addendum After this paper was completed, we were informed that what we call “normalizing spaces of operators” have been studied, from a different viewpoint, by other authors under the names “ternary rings of operators” or “triple systems”. See for example M. Neal and B. Russo, Contractive Projections and Operator Spaces (preprint, arXiv: math.OA/0201187) and its bibliography. We thank J. Arazy, D.P. Blecher and B. Russo for bringing the relevant literature to our attention. We have chosen to retain the name “normalizing spaces” to emphasize the relation with normalizers of operator algebras, which is one of the main points of our work.
We have also found out that our Proposition 2.6 is essentially contained in section 3 of the paper of L.A. Harris, ‘A generalization of C\*-algebras’, Proc. London Math. Soc. (3) 42 (1981) 331-361.
|
warning/0005/math0005177.html
|
ar5iv
|
text
|
# SOME ISOMORPHISMS FOR THE BRAUER GROUPS OF A HOPF ALGEBRA
## 1 Introduction
There have been given several generalizations of the Brauer group of a field $`k`$, due among others to Wall, Long, Van Oystaeyen, Caenepeel and Zhang. In particular, Caenepeel, Van Oystaeyen and Zhang defined in the Brauer group $`BQ(k,H)`$ of a Hopf algebra $`H`$ with bijective antipode. This is a special case of Brauer group of a braided monoidal category (see : the Brauer group of a symmetric monoidal category had been defined by B. Pareigis in ). Here the category is that of left modules of the Drinfel’d quantum double (see and ) of a finitely generated projective Hopf algebra $`H`$ over a commutative ring $`k`$.
$`BQ(k,H)`$ generalizes the Brauer-Long group of a commutative and cocommutative Hopf algebra over a commutative ring defined by Long in . In fact it is shown in that when $`H`$ is a commutative cocommutative Hopf algebra, the Brauer group of $`H`$ and the Brauer-Long group of $`H`$ are (anti-)isomorphic.
Another example of Brauer group of a braided monoidal category is the Brauer group $`BC(k,H,r)`$ where $`H`$ is a dual quasitriangular Hopf algebra with universal $`R`$-form $`r`$. In this case the category is that of right $`H`$-comodules.
By results in and results in (the dual version can be found in and the survey book in the context of quasi Hopf algebras) twisting the algebra structure of $`H`$ by a $`2`$-cocycle $`\sigma `$ provides a new dual quasitriangular Hopf algebra $`{}_{\sigma }{}^{}H_{\sigma ^1}^{}`$ whose comodule category is equivalent to that of $`H`$. Hence the two Brauer groups $`BC(k,H,r)`$ and $`BC(k,_\sigma H_{\sigma ^1},r_\sigma )`$ will be isomorphic. In this paper we use this result in order to show that although the universal $`R`$-forms $`r_t`$ for $`tk`$ of Sweedler’s four dimensional Hopf algebra $`H_4`$ are not all isomorphic (this fact was proved by Radford in ), the Brauer groups $`BC(k,H,r_t)`$ are all isomorphic for every $`tk`$. This result is achieved by finding a suitable cocycle which does not change the algebra structure of $`H_4`$ but changes the universal $`R`$-form $`r_t`$ into $`r_0`$. Since the Brauer group $`BC(k,H_4,r_0)`$ (or equivalently, $`BM(k,H_4,R_0)`$ the Brauer group of the braided category of $`H_4`$-modules with braiding given by the triangular $`R`$-matrix $`R_0`$) has been computed in , the computation of all $`BC`$’s for $`H_4`$ is accomplished. This group is the direct sum of the additive group $`(k,+)`$ and of another generalization of the Brauer group of $`k`$, the so-called Brauer-Wall group $`BW`$ of $`k`$. $`BW(k)`$ is the Brauer group of the category of $`𝐙_2`$-graded $`k`$-modules and it has been explicitely described by Wall in by means of an exact sequence having the usual Brauer group of $`k`$ as a kernel.
Cocycle twisting also provides an isomorphism between the groups $`BC(k,H,r)`$ and $`BC(k,H^{op},r\tau )`$ where $`\tau `$ is the usual flip and $`H^{op}`$ denotes the Hopf algebra with opposite product.
In the case of a commutative and cocommutative finitely generated and projective Hopf algebra $`H`$, it was shown by F. Tilborghs in that the Brauer-Long group of $`H`$ is (anti-)isomorphic to the Brauer-Long group of $`H^{}`$.
Hence for $`H`$ commutative and cocommutative, finitely generated and projective one has an isomorphism between the Brauer group $`BQ`$ of $`H`$ and the Brauer group $`BQ`$ of $`H^{}`$. We shall use an isomorphism of Radford involving $`D(H)`$ and $`D(H^{})`$ where $`D(H)`$ denotes the Drinfel’d quantum double of $`H`$, together with the results about the Brauer group $`BC`$ of the opposite Hopf algebra in order to generalize Tilborghs’ result to the case of finitely generated projective Hopf algebras with a bijective antipode. Everything boils down to the fact that there is an anti-equivalence of braided categories between the category of Yetter-Drinfel’d $`H`$-modules (or crossed bimodules, or Quantum Yang-Baxter modules) and that of Yetter-Drinfel’d $`H^{}`$-modules given by the usual duality functor mapping left $`H`$-modules to right $`H^{}`$-comodules and right $`H`$-comodules to left $`H^{}`$-modules. Equivalently, there is an anti-equivalence of braided categories between left $`D(H)`$-modules and left $`D(H^{})`$-modules.
At the end of the paper we relate the Brauer groups of $`H`$ and $`H^{op}`$. We show that there is always an isomorphism from $`BQ(k,H)`$ to $`BM(k,D(H^{op}),\tau _{D(H^{op})}^1)`$ where $`_{D(H^{op})}`$ denotes the standard $`R`$-matrix of $`D(H^{op})`$. If $`D(H)`$ were triangular with respect to its standard $`R`$-matrix, an isomorphism $`BQ(k,H)BQ(k,H^{op})`$ would follow because $`BQ(k,H)BM(k,D(H),)`$. However for nontrivial $`H`$ the quantum double $`D(H)`$ cannot be triangular with respect to its standard $`R`$-matrix.
The paper is organized as follows. In Section 2 the construction of the Brauer group of a braided monoidal category is recalled. The particular case of the Brauer group(s) of a Hopf algebra is included in a subsection. In Section 3 the theory of cocycle twists of a Hopf algebra is used in order to get isomorphisms between the groups $`BC`$ of Hopf algebras related by a twist. The main results of the paper are to be found in Sections 4 and 5. In Section 4 I apply the isomorphism described above to the particular example of Sweedler’s four dimensional Hopf algebra $`H_4`$ in order to obtain $`BC(k,H_4,r_t)`$ for every $`tk`$. In Section 5, I provide an isomorphism between the Brauer group $`BQ`$ of a Hopf algebra $`H`$ with the Brauer group $`BQ`$ of its dual and I relate them to the Brauer groups of its opposite Hopf algebra.
## 2 The Brauer group of a braided category
The Brauer group of a braided monoidal category was defined in and this definition contains all known Brauer groups. The case of a symmetric category had been already treated in and in the symmetric case the Brauer group is abelian. Here we give a short account of the general construction.
Let $`𝒞`$ denote a braided monoidal category with $``$, $`\psi `$ and $`I`$ respectively the tensor, braiding and identity object. For objects $`P`$ and $`Q`$ in $`𝒞`$ if the functor $`𝒞(P,Q)`$ is representable one denotes the representing object by $`[P,Q]`$. Similarly if the functor $`𝒞(P,Q)`$ is representable, the representing object is denoted by $`\{P,Q\}`$. An object $`A`$ in $`𝒞`$ is an algebra if there are morphisms $`m:AAA`$ (product) and $`\eta :IA`$ (unit) satisfying associativity an unitary conditions. The $`𝒞`$-opposite algebra $`\overline{A}`$ is then defined as $`A`$ as object but with product $`\overline{m}:=m\psi `$ and same unit. The tensor product of two algebras $`A`$ and $`B`$ in $`𝒞`$ becomes an algebra in $`𝒞`$ denoted by $`A\mathrm{\#}B`$ with product $`(m_Am_B)(\mathrm{id}\psi \mathrm{id})`$. For more details, see and references therein. In particular, $`[P,P]`$ and $`\{P,P\}`$ are algebras in $`𝒞`$.
For an algebra $`A`$ in $`𝒞`$ one has the two maps:
$`F:𝒞(X,A\mathrm{\#}\overline{A})`$ $`𝒞(X,[A,A])`$ (2.1)
$`F(a\mathrm{\#}\overline{b})<d>`$ $`=a\overline{m}(bd)𝒞(XY,A)`$ (2.2)
for every $`d𝒞(Y,A)`$ where $`X`$ and $`Y`$ are objects in $`𝒞`$ and $`<d>`$ stands for “evaluation” at $`d`$ and
$`G:𝒞(X,\overline{A}\mathrm{\#}A)`$ $`𝒞(X,\{A,A\})`$ (2.3)
$`<d>G(\overline{a}\mathrm{\#}b)`$ $`=\overline{m}(da)b𝒞(XY,A)`$ (2.4)
for every $`d𝒞(Y,A)`$ where $`X`$ and $`Y`$ are objects in $`𝒞`$ and $`<d>`$ stands for “evaluation” at $`d`$. An algebra in $`𝒞`$ is called $`𝒞`$-Azumaya if $`F`$ and $`G`$ are isomorphisms and $`A`$ is faithfully projective in $`𝒞`$ (see or for the definition of faithfully projective). It turns out that: the product of two $`𝒞`$-Azumaya algebras is $`𝒞`$-Azumaya; the opposite algebra of a $`𝒞`$-Azumaya algebra is $`𝒞`$-Azumaya. If $`P`$ is faithfully projective, then $`[P,P]`$ is also $`𝒞`$-Azumaya. One can define an equivalence relation on the set of $`𝒞`$-Azumaya algebras: $`AB`$ if there exist faithfully projective objects $`M`$ and $`N`$ such that
$$A\mathrm{\#}[M,M]B\mathrm{\#}[N,N].$$
(2.5)
It is proved in that this is indeed an equivalence relation and that the set of equivalence classes becomes a group $`Br(𝒞)`$ with product induced by $`\mathrm{\#}`$. The inverse of a class represented by an algebra $`A`$ will be the class represented by the algebra $`\overline{A}`$.
In the second Brauer group $`Br^{}(𝒞)`$ was also defined if $`𝒞`$ satisfies some extra conditions. Under general conditions there is a group homomorphism $`Br^{}(𝒞)Br(𝒞)`$ which is very often a monomorphism and it is the identity if the unit object $`I`$ is projective. The elements of $`Br^{}(𝒞)`$ are classes of separable $`𝒞`$-Azumaya algebras in the category (see or for the definition).
###### Remark 2.1
It is clear that if we replace a category $`𝒞`$ by its opposite category $`𝒞^{op}`$ with $`A^{op}B:=BA`$ for two objects in $`𝒞^{op}`$, $`f^{op}g:=gf`$ for two morphisms $`f`$ and $`g`$, and $`\psi _{AB}^{op}:=\psi _{BA}`$, then the map
$$\alpha :Br(𝒞)Br(𝒞^{op})$$
$$[A][\overline{A}]_{op}$$
where $`[A]`$ denotes the class of $`A`$ and $`[A]_{op}`$ denotes the equivalence class of $`A`$ in $`𝒞^{op}`$ is a well-defined isomorphisms of groups which induces also an isomorphism between $`Br^{}(𝒞)`$ and $`Br^{}(𝒞^{op})`$. In fact it is clear that $`A`$ is $`𝒞`$-Azumaya iff $`\overline{A}`$ is $`𝒞^{op}`$-Azumaya and $`A_𝒞B`$ iff $`\overline{A}_{𝒞^{op}}\overline{B}`$ iff $`A_{𝒞^{op}}B`$.
### 2.1 The Brauer groups of a Hopf algebra
From now on $`k`$ will denote a commutative ring. Every $`k`$-module $`M`$ will be assumed to be finitely generated, projective and faithful. This implies the existence of “dual bases” $`\{m_i\}M`$ and $`\{m_j^{}\}M^{}`$ for which $`m_j^{}(m)m_j=m`$ for every $`mM`$.
All tensor product will be intended to be over $`k`$.
All Hopf algebras will be assumed to be finitely generated, projective and with bijective antipode $`S`$. $`<,>`$ shall always denote evaluation between $`H^{}`$ and $`H`$. Since $`H`$ is finitely generated and projective, $`H^{}`$ is also a finitely generated projective Hopf algebra.
A $`k`$-module $`M`$ is called faithfully projective (progenerator) if there exist elements $`m_j^{}M^{}`$ and $`m_jM`$ for which $`m_j^{}(m_j)=1`$.
$`H^{op}`$ shall denote the Hopf algebra with opposite product and $`H^{cop}`$ shall denote the Hopf algebra with opposite coproduct.
The Brauer group $`BQ(k,H)`$ of a Hopf algebra $`H`$ over $`k`$ is a particular case of Brauer group of a braided monoidal category where the category is that of Yetter-Drinfel’d $`H`$-modules.
###### Definition 2.2
A Yetter-Drinfel’d $`H`$-module $`M`$ is a $`k`$-module which is a left $`H`$-module and a right $`H`$-comodule satisfying the compatibility condition
$$h_{(1)}m_{(0)}h_{(2)}m_{(1)}=(h_{(2)}m)_{(0)}(h_{(2)}m)_{(1)}h_{(1)}$$
(2.6)
for every $`hH`$ and $`mM`$. In the formula $``$ denotes the $`H`$-action on $`M`$, $`\chi (m)=m_{(0)}m_{(1)}MH`$ is the right comodule structure map and $`\mathrm{\Delta }(h)=h_{(1)}h_{(2)}`$ denotes the coproduct in $`H`$.
###### Remark 2.3
A Yetter-Drinfel’d module is also sometimes called a crossed bimodule (see ) or a Quantum Yang-Baxter $`H`$-module (see ).
It is a result of S. Majid in that the category of Yetter-Drinfel’d $`H`$-modules is equivalent to the category of left $`D(H)`$-modules where $`D(H)`$ is the Drinfel’d double of $`H`$ defined in .
In order to fix notation we recall that $`D(H)`$ is a quasitriangular Hopf algebra whose underlying coalgebra is $`H^{,cop}H`$, with product
$$(\xi a)(\eta b)=\xi \eta _{(2)}a_{(2)}b<\eta _{(1)},S^1(a_{(3)})><\eta _{(3)},a_{(3)}>$$
and with $`R`$-matrix $`(\epsilon h_i)(h_i^{}1)`$ where $`\{h_i\}`$ and $`\{h_i^{}\}`$ are dual bases in $`H`$ and $`H^{}`$.
In this notation the $`H`$-action $``$ and the $`H^{}`$-action $`h^{}m=(\mathrm{id}<h^{},>)\chi `$ together with the compatibility condition define a $`D(H)`$-module structure on a Yetter-Drinfel’d module $`M`$ and viceversa because $`H`$ and $`H^{}`$ are subalgebras of $`D(H)`$. We shall denote the $`D(H)`$-action by $``$.
It is well-known that the tensor product of two Yetter-Drinfel’d $`H`$-modules $`M`$ and $`N`$ can be naturally equipped of a Yetter-Drinfel’d $`H`$-module structure denoted by $`M\stackrel{~}{}N`$ as follows:
$$\chi (ab)=a_{(0)}b_{(0)}b_{(1)}a_{(1)}$$
(2.7)
and
$$h(ab)=h_{(1)}ah_{(2)}b.$$
(2.8)
The category $`𝒴𝒟(H)`$ of Yetter-Drinfel’d modules (with module and comodule morphisms) together with $`\stackrel{~}{}`$ becomes a monoidal category. Moreover, there is an isomorphism of Yetter-Drinfel’d $`H`$-modules between $`M\stackrel{~}{}N`$ and $`N\stackrel{~}{}N`$ given by
$$\varphi _{MN}(m\stackrel{~}{}n)=n_{(0)}\stackrel{~}{}n_{(1)}m$$
(2.9)
which makes of the category of Yetter-Drinfel’d $`H`$-modules a braided monoidal category. The mentioned equivalence of $`𝒴𝒟(H)`$ and $`{}_{D(H)}{}^{}`$ is an equivalence of braided monoidal categories. In terms of $`D(H)`$-modules, if $`=_1_2`$ is the standard $`R`$-matrix for $`D(H)`$ the braiding is nothing but
$$\varphi _{MN}(m\stackrel{~}{}n)=_2n\stackrel{~}{}_1m$$
(2.10)
In this setting for a Yetter-Drinfel’d module $`P`$, $`[P,P]`$ and $`\{P,P\}`$ are the usual $`End(P)`$ and $`End(P)^{op}`$ respectively, equipped with the Yetter-Drinfel’d module structures:
$$(hf)(m)=h_{(1)}f(S(h_{(2)}m));$$
(2.11)
$$\chi (f)(m)=f(m_{(0)})_{(0)}S^1(m_{(1)})f(m_{(0)})_{(1)}$$
(2.12)
for every $`mM`$ and $`fEnd(M)`$ for $`End(M)`$ and
$$(h^{}f)(m)=h_{(2)}f(S^1(h_{(1)})m);$$
(2.13)
$$\chi ^{}(f)(m)=f(m_{(0)})_{(0)}f(m_{(0)})_{(1)}S(m_{(1)})$$
(2.14)
for every $`mM`$ and $`fEnd(M)`$ for $`End(M)^{op}`$. Those constructions come from two possible natural Yetter-Drinfel’d module structures on $`[P,I]=P^{}`$.
An algebra in $`𝒴𝒟(H)`$ is called a Yetter-Drinfel’d $`H`$-module algebra and it corresponds to a $`D(H)`$-module algebra:
###### Definition 2.4
A Yetter-Drinfel’d $`H`$-module algebra $`M`$ is an algebra having the structure of a Yetter-Drinfel’d $`H`$-module and such that the module and comodule structure make of $`M`$ a left $`H`$-module algebra and a right $`H^{op}`$-comodule algebra.
The tensor product $`A\stackrel{~}{}B`$ of two Yetter-Drinfel’d $`H`$-module algebras $`A`$ and $`B`$ becomes a Yetter-Drinfel’d module algebra denoted by $`A\mathrm{\#}B`$ with product given by $`(a\mathrm{\#}c)(b\mathrm{\#}d):=a\varphi (cb)d=ab_{(0)}\mathrm{\#}(b_{(1)}c)d`$.
The maps $`F`$ and $`G`$ become then:
$`F:A\mathrm{\#}\overline{A}`$ $`End(A)`$ (2.15)
$`F(a\mathrm{\#}\overline{b})(c)`$ $`={\displaystyle ac_{(0)}(c_{(1)}b)}=m(a\varphi _{AA}(bc))`$ (2.16)
and
$`G:\overline{A}\mathrm{\#}A`$ $`End(A)^{op}`$ (2.17)
$`G(\overline{a}\mathrm{\#}b)(c)`$ $`={\displaystyle a_{(0)}(a_{(1)}c)b}=m(\varphi _{AA}\tau \mathrm{id})(\mathrm{id}\tau )(abc)`$ (2.18)
for every $`a,b`$ and $`c`$ in $`A`$. It had already been proved in that $`F`$ and $`G`$ are Yetter-Drinfel’d $`H`$-module algebra maps. A Yetter-Drinfel’d $`H`$-module algebra $`A`$ is $`H`$-Azumaya if $`F`$ and $`G`$ are isomorphisms. In this case $`AB`$ if there exist faithfully projective Yetter-Drinfel’d $`H`$-modules $`M`$ and $`N`$ such that $`A\mathrm{\#}End(M)B\mathrm{\#}End(N)`$ as Yetter-Drinfel’d $`H`$-module algebras. $`Br(𝒴𝒟(H))`$ in usually denoted by $`BQ(k,H)`$.
An $`H`$-Azumaya algebra is said to be strongly $`H`$-Azumaya if $`k`$ is a direct summand of $`A`$ as Yetter-Drinfel’d $`H`$-module. The second Brauer group $`Br^{}(𝒴𝒟(H))`$ is usually denoted $`BQS(k,H)`$ and it coincides with $`BQ(k,H)`$ if $`H`$ is semisimple-like and cosemisimple-like (see Prop. 2.27 in ). In general $`BQS(k,H)`$ is the subgroup whose elements are classes parametrized by strongly $`H`$-Azumaya algebras.
If $`H`$ is a quasitriangular Hopf algebra with universal $`R`$ matrix $`R=R_1R_2`$ it is well known that to every module (algebra) $`A`$ one can associate a right $`H^{op}`$-comodule (algebra) structure on $`A`$ given by
$$\chi _R(a)=R_2aR_1$$
(2.19)
obtaining a Yetter-Drinfel’d $`H`$-module algebra. In this case the category $`{}_{H}{}^{}`$ of left $`H`$-modules with $`H`$-module maps is a full subcategory of $`𝒴𝒟(H)`$. $`Br(_H)`$ is then a subgroup of $`BQ(k,H)`$ and it is usually denoted by $`BM(k,H,R)`$. It is the subgroup of $`BQ(k,H)`$ whose elements are represented by a Yetter-Drinfel’d module algebras whose comodule structure is defined by (2.19).
Dually, if $`H`$ is a dual quasitriangular Hopf algebra with universal $`R`$-form $`r`$, to a right $`H^{op}`$-module (algebra) $`A`$ one can associate a left $`H`$-module (algebra) structure on $`A`$ given by
$$ha=a_{(0)}r(ha_{(1)})$$
(2.20)
obtaining a Yetter-Drinfel’d module algebra. The category $`^H`$ of right $`H`$-comodules with $`H`$-comodule maps is a full subcategory of $`𝒴𝒟(H)`$. $`Br(^H)`$ is then a subgroup of $`BQ(k,H)`$ denoted by $`BC(k,H,r)`$. It is the subgroup of $`BQ(k,H)`$ whose elements are represented by a Yetter-Drinfel’d module algebras whose module structure is defined by (2.20).
It is well-known that $`BQ(k,H)BC(k,D(H)^{},r)BM(k,D(H),)`$ where $`D(H)`$ is the Drinfel’d double of $`H`$ and $``$ is its standard $`R`$-matrix. Hence it is enough to study $`BC(k,H,r)`$ for a dual quasitriangular Hopf algebra.
## 3 An equivalence of categories
In this section we show a few isomorphisms for the group $`BC`$ of a dual quasitriangular Hopf algebra. Everything can be dualized, considering quasitriangular Hopf algebras and $`BM`$. We leave this task to the reader.
Let $`H`$ be a bialgebra over $`k`$ and let $`B`$ be a left (resp. right) $`H`$-comodule algebra with comodule map $`\chi `$. A left (resp. right) $`2`$-cocycle $`\sigma `$ is a linear map $`\sigma :HHk`$ satisfying
* $`\sigma (k_{(1)}m_{(1)})\sigma (hk_{(2)}m_{(2)})=\sigma (h_{(1)}k_{(1)})\sigma (h_{(2)}k_{(2)}m)`$
$$\text{(resp. }\sigma (k_{(2)}m_{(2)})\sigma (hk_{(1)}m_{(1)})=\sigma (h_{(2)}k_{(2)})\sigma (h_{(1)}k_{(1)}m))$$
* $`\sigma (h1)=\sigma (1h)=\epsilon (h)`$ $``$ $`h,k,mH`$
Then, the $`\sigma `$-left (resp. $`\sigma `$-right) twisted comodule $`{}_{\sigma }{}^{}B`$ (resp. $`B_\sigma `$) is an algebra with the same underlying vector space as $`B`$, and product given by:
$$\overline{a}\overline{b}=\sigma (a_{(1)}b_{(1)})\overline{a_{(0)}b_{(0)}}$$
(3.1)
if $`\chi (a)=a_{(1)}a_{(0)}`$, and $`\chi (b)=b_{(1)}b_{(0)}HB`$ for $`a,bB`$;
$$(\mathrm{resp}.\overline{a}\overline{b}=\overline{a_{(0)}b_{(0)}}\sigma (a_{(1)}b_{(1)})$$
(3.2)
if $`\chi (a)=a_{(0)}a_{(1)}`$, and $`\chi (b)=b_{(0)}b_{(1)}BH`$), where $`a\overline{a}`$ denotes the identification of vector spaces (see for instance , Paragraph 7.5).
If $`B`$ is a bialgebra and $`\sigma `$ is a left 2-cocycle one can perform such a twist to $`B`$, viewed as a left (resp. right) $`B`$-comodule algebra. If $`\sigma `$ is a convolution invertible left 2-cocycle, then $`\sigma ^1`$ is a right 2-cocycle. It is well-known (see for instance , or ) that the double twist $`{}_{\sigma }{}^{}B_{\sigma ^1}^{}`$ with the same coproduct of $`B`$ is again a bialgebra and if $`B`$ is a Hopf algebra, then $`{}_{\sigma }{}^{}B_{\sigma ^1}^{}`$ is also a Hopf algebra with antipode $`S_\sigma `$ given by $`(uSu^1)(\mathrm{\Delta }\mathrm{id})\mathrm{\Delta }`$. Here $`u_\sigma B_{\sigma ^1}^{}B^{}`$ is the linear functional given by $`u=\sigma (\mathrm{id}S)\mathrm{\Delta }`$.
The following facts are also well-known (see for instance and references therein or, for a dual version involving the case of Drinfel’d quasi Hopf algebras, and references therein):
* The category of right $`H`$-comodules is equivalent to the category of right $`{}_{\sigma }{}^{}H_{\sigma ^1}^{}`$-comodules as monoidal category. In this case though we need the comodule structure of a tensor product to be
$$mnm_{(0)}n_{(0)}m_{(1)}n_{(1)}.$$
(3.3)
The monoidal functor $``$ is the identity on objects (the coproduct is unchanged) and the natural transformation $`\eta :(M)(N)(MN)`$
$$\eta (mn)=m_{(0)}n_{(0)}\sigma ^1(m_{(1)}n_{(1)}).$$
(3.4)
The compatibility conditions for $`\eta `$ follow by the cocycle condition on $`\sigma `$.
* If $`H`$ is dual quasitriangular with universal $`R`$-form $`r`$, then $`{}_{\sigma }{}^{}H_{\sigma ^1}^{}`$ is also dual quasitriangular with universal $`R`$-form given by $`r_\sigma =(\sigma \tau )r\sigma ^1`$.
* The categories $`^H`$ and $`^{{}_{\sigma }{}^{}H_{\sigma ^1}^{}}`$ are equivalent braided monoidal categories. In this case one uses the braiding
$$\varphi _{MN}(mn)n_{(0)}m_{(0)}r(m_{(1)}n_{(1)})$$
(3.5)
compatible with the different definition of comodule structure on $`MN`$. One can check that the braiding $`\varphi _{MN}`$ is respected by the functor $``$ and the natural transformation $`\eta `$ mentioned above.
* The universal $`R`$-forms $`r`$ and $`r^1\tau `$ of a dual quasitriangular Hopf algebra $`H`$ are 2-cocycles and $`{}_{r}{}^{}H_{r^1}^{}=H^{op}=_{r^1\tau }H_{r\tau }`$. The new universal $`R`$-form is in both cases $`r\tau `$.
It is a straightforward check that if $`H`$ is a bialgebra with coproduct $`\mathrm{\Delta }`$, and if $`\sigma `$ is a 2-cocycle for $`H`$, then $`\sigma \tau `$ is a 2-cocycle for $`H^{op}`$ and
$`(_\sigma H_{\sigma ^1})^{op}_{\sigma \tau }(H^{op})_{\sigma ^1\tau }`$. Hence, repeating the discussion above for $`H^{op}`$ the category of right $`H`$-comodules with tensor product structure
$$mnm_{(0)}n_{(0)}n_{(1)}m_{(1)}$$
and braiding
$$mnn_{(0)}m_{(0)}r(n_{(1)}m_{(1)})$$
is equivalent, as braided monoidal category, to that of $`{}_{\sigma }{}^{}H_{\sigma ^1}^{}`$-comodules with tensor product structure
$$mnm_{(0)}n_{(0)}\sigma (n_{(1)}m_{(1)})n_{(2)}m_{(2)}\sigma ^1(n_{(3)}m_{(3)})$$
and braiding
$$mnn_{(0)}m_{(0)}\sigma (m_{(1)}n_{(1)})r(n_{(2)}m_{(2)})\sigma ^1(n_{(3)}m_{(3)}).$$
The natural transformation $`\eta `$ is in this case:
$$mnm_{(0)}n_{(0)}\sigma ^1(n_{(1)}m_{(1)}).$$
(3.6)
An algebra $`A`$ in $`^H`$ is mapped by the functor to the algebra $`A_{\sigma ^1\tau }`$ with product $`m_A\eta _{A,A}`$ i.e. $`ab:=a_{(0)}b_{(0)}\sigma ^1(b_{(1)}a_{(1)})`$. Hence we have:
###### Proposition 3.1
Let $`H`$ be a (faithfully projective) dual quasitriangular Hopf algebra with universal $`R`$-form $`r`$. Let $`\sigma `$ be an invertible $`2`$-cocycle.
Then $`BC(k,H,r)BC(k,_\sigma H_{\sigma ^1},r_\sigma )`$. The class of an $`H`$-Azumaya algebra $`A`$ is mapped to the class of $`A_{\sigma ^1\tau }`$. In particular, $`BC(k,H,r)BC(k,H^{op},r\tau )`$.
Proof: It follows by the above observations. Observe that twisting $`H`$ by $`r^1\tau `$ implies that $`[A]`$ is mapped to $`[\overline{A^{op}}]`$.
###### Remark 3.2
The result about $`H^{op}`$ could be obtained also by checking that the categories of right $`H^{op}`$ comodules and of right $`H`$ comodules are anti-equivalent braided categories. Then the anti-isomorphism between the two Brauer groups is given on representatives by $`AA^{op}`$.
## 4 An example: $`BC(k,H_4,r_t)`$
Let $`k`$ be a field of characteristic different from $`2`$ and let $`H_4`$ denote Sweedler’s four dimensional Hopf algebra over $`k`$ generated by $`g`$ and $`h`$ such that $`g^2=1`$, $`h^2=0`$ and $`gh+hg=0`$. As far as the coproduct $`\mathrm{\Delta }`$ is concerned, $`g`$ is grouplike and $`h`$ is twisted-primitive with $`\mathrm{\Delta }(h)=hg+1h`$. The antipode $`S`$ is such that $`S(g)=g`$ and $`S(h)=gh`$. It is well-known that $`H_4^{}`$ is isomorphic to $`H_4`$. An isomorphism is obtained sending $`g`$ to $`f_1f_g`$ and $`h`$ to $`f_h+f_{gh}`$ where $`f_x`$ denotes the dual element of $`xH_4`$. It is also well-known that $`H_4`$ has a family of universal $`R`$-forms (and, dually of universal $`R`$-matrices) parametrized by the elements in $`k`$. The universal $`R`$-forms $`r_t`$ ($`tk`$) were firstly found by Radford in and are determined by the axioms of an $`R`$-form together with
$$r_t(1x)=r_t(x1)=\epsilon (x)$$
$$r_t(gg)=1r_t(gh)=r_t(hg)=r_t(ggh)=r_t(ghg)=0$$
$$r_t(ghh)=r_t(hh)=r_t(ghgh)=r_t(hgh)=t.$$
A first observation is that all those structures are cotriangular, i.e. $`r_t(r_t\tau )=\epsilon \epsilon =(r_t\tau )r_t`$. This means that for every $`a`$ and $`b`$ in $`H_4`$ one has:
$$r_t(a_{(1)}b_{(1)})r_t(b_{(2)}a_{(1)})=\epsilon (a)\epsilon (b)=r_t(b_{(1)}a_{(1)})r_t(a_{(2)}b_{(2)})$$
(4.1)
By the symmetry in the formula if we interchange $`a`$ and $`b`$ it is enough to check the left hand side condition on pairs of basis elements. We have various cases depending on the coproduct of $`a`$ and $`b`$:
* $`a`$ and $`b`$ are both grouplike elements of the basis (i.e. $`1`$ or $`g`$). The left hand side of $`(\text{4.1})`$ reads $`r_t(ab)r_t(ba)`$. This is $`1^2`$ for the pairs $`(1,g)`$, $`(g,1)`$ and $`(1,1)`$ and it is $`(1)^2`$ when $`a=b=g`$. In all cases this is equal to $`\epsilon (a)\epsilon (b)`$.
* One element in $`\{a,b\}`$ is grouplike and the other is twisted primitive (i.e. $`h`$ or $`gh`$). Then each summand of the left hand side of $`(\text{4.1})`$ will contain an expression of type $`r_t(xy)`$ with $`x`$ grouplike and $`y`$ twisted primitive. Therefore each summand is zero. For instance
$$r_tr_t\tau (hg)=r_t(hg)r_t(gg)+r_t(1g)r_t(gh)=0=\epsilon (g)\epsilon (h).$$
Hence $`r_t(a_{(1)}b_{(1)})r_t(b_{(2)}a_{(1)})=0=\epsilon (a)\epsilon (b)`$.
* $`a`$ and $`b`$ are both twisted primitives. Then the only nonzero terms in the sum on the left hand side of $`(\text{4.1})`$ appear when $`a_{(1)}`$ and $`b_{(1)}`$ are both twisted primitives, or when $`a_{(1)}`$ and $`b_{(1)}`$ are both grouplikes. If $`a=b=h`$ the expression becomes
$$r_tr_t\tau (hh)=r_t(hh)r_t(gg)+r_t(11)r_t(hh)=t+t=\epsilon (h)^2.$$
If $`a=b=gh`$ the expression becomes
$$r_t(ghgh)r_t(11)+r_t(gg)r_t(ghgh)=t+t=\epsilon (gh)\epsilon (gh).$$
If $`a=h`$ and $`b=gh`$ the expression becomes
$$r_t(hgh)r_t(1g)+r_t(1g)r_t(ghh)=tt=\epsilon (gh)\epsilon (h).$$
Finally if $`a=gh`$ and $`b=h`$ the expression is
$$r_tr_t\tau (ghh)=r_t(ghh)r_t(g1)+r_t(g1)r_t(hgh)=tt=\epsilon (h)\epsilon (gh).$$
Hence $`H_4`$ is cotriangular for every universal $`R`$-form $`r_t`$. Therefore $`BC(k,H_4,r_t)BC(k,H^{op},r_t^1)BC(k,H,r_t^1\tau )`$ and it is an abelian group.
Dually, one can check that $`(H_4,R_t)`$ is triangular for every universal $`R`$-matrix $`R_t`$. Although triangularity of $`R_t`$ follows by the previous result we sketch the proof for sake of completeness because we have not met this result in the literature before. The family of $`R`$-matrices (see or ) is given by
$$R_t=\frac{1}{2}(11+1g+g1gg)+\frac{t}{2}(hh+hgh+ghghghh)$$
for $`tk`$. $`R_0=\tau R_0`$ and $`R_0^2=11`$ because it corresponds to a Hopf involution $`f_R:H_4^{}H_4^{cop}`$ where $`f_R(\xi )=(<\xi ,>\mathrm{id})(R_0)`$ (see ). Hence $`(H_4,R_0)`$ is triangular. Put $`R_t=R_0+R_t^{}`$. $`(H_4,R_t)`$ is triangular if $`(\tau R_t)R_t=11=R_t(\tau R_t)`$. Since $`\tau `$ is an algebra isomorphism it is enough to check the relation for $`(\tau R_t)R_t`$. This expression is equal to
$$(\tau R_0)R_0+(\tau R_0)R_t^{}+(\tau R_t^{})R_0^{}+(\tau R_t^{})R_t^{}=11+(\tau R_0)R_t^{}+(\tau R_t^{})R_0^{}.$$
$`(\tau R_t^{})R_t^{}=0`$ because $`h^2=0`$ appears in every component. Checking that $`(\tau R_0)R_t^{}+(\tau R_t^{})R_0^{}=0`$ is a striaghtforward computation that we leave to the reader.
The group $`BM(k,H_4,R_0)`$ has been computed by Van Oystaeyen and Zhang in . This group is isomorphic to $`BC(k,H_4,r_0)`$ because the universal $`R`$-matrix $`R_0`$ goes over to the universal $`R`$-form $`r_0`$ under the isomorphism $`H_4H_4^{}`$ previously given.
We want to show here that $`BC(k,H_4,r_t)BC(k,H,r_0)`$ for every $`tk`$, hence that $`BM(k,H_4,R_0)BM(k,H_4,R_s)`$ for every $`sk`$. We shall do this by providing for every $`tk`$ there exists a suitable element $`\sigma _t(H_4H_4)^{}`$ such that
* $`\sigma _t`$ is a left 2-cocycle for $`H_4`$;
* $`\sigma _t`$ is invertible;
* the twisted product in $`{}_{\sigma _t}{}^{}(H_4)_{\sigma _t^1}^{}`$ coincides with the product in $`H_4`$;
* $`\sigma _t\tau r_t\sigma _t^1=r_0`$.
The functional $`\sigma _t`$ is defined on the basis elements of $`H_4H_4`$ as follows:
$$\sigma _t(x1)=\sigma _t(1x)=\epsilon (x)\text{for every }xH_4\text{.}$$
$$\sigma _t(gg)=1$$
$$\sigma _t(hh)=\sigma _t(ghh)=\sigma _t(hgh)=\sigma _t(ghgh)=\frac{t}{2}$$
and $`\sigma _t(xy)=0`$ whenever $`x`$ is grouplike and $`y`$ twisted primitive or the other way around.
It is a $`2`$-cocycle if for every triple of basis elements $`k,a,m`$ there holds:
$$\sigma _t(k_{(1)}m_{(1)})\sigma _t(ak_{(2)}m_{(2)})=\sigma _t(a_{(1)}k_{(1)})\sigma _t(a_{(2)}k_{(2)}m).$$
If one of the elements is $`1`$ then the condition is verified because $`\sigma _t`$ is unitary (i.e. it coincides with $`\epsilon `$ on $`1x`$ and $`x1`$). Hence we have to check the condition on all triples of elements in $`\{g,h,gh\}`$.
Then we have different cases depending on how often $`g`$ appears in the triple.
1. $`g`$ appears $`3`$ times.
Then we have
$$\sigma _t(gg)\sigma _t(g1)=1=\sigma _t(gg)\sigma _t(1g)$$
2. $`g`$ appears twice in the triple:
If $`g=a=m`$ and $`k`$ is twisted primitive the condition becomes
$$\sigma _t(k_{(1)}g)\sigma _t(gk_{(2)}g)=\sigma _t(gk_{(1)})\sigma _t(gk_{(2)}g).$$
Since $`k`$ is twisted primitive, in every summand $`k_{(1)}`$ and $`k_{(2)}`$ can never be both grouplikes, hence both sums are $`0`$. The cases $`g=a=k`$ and $`g=k=m`$ are checked similarly.
3. $`g`$ appears once in the triple:
If $`g=m`$ then the condition reads
$$\sigma _t(k_{(1)}g)\sigma _t(ak_{(2)}g)=\sigma _t(a_{(1)}k_{(1)})\sigma _t(a_{(2)}k_{(2)}g).$$
The only nonzero component of the left hand side appears when $`k_{(1)}`$ is grouplike and $`k_{(2)}=k`$, hence the left hand side is equal to $`\sigma _t(akg)`$. The only nonzero component of the right hand side is when $`a_{(2)}k_{(2)}`$ is grouplike, i.e. when both $`a_{(2)}`$ and $`k_{(2)}`$ are grouplikes, hence $`a_{(1)}=a`$, $`k_{(1)}=k`$ and the right hand side becomes $`\sigma _t(ak)`$. It is straightforward to check that for the twisted primitives $`a`$ and $`k`$ in $`H_4`$ there holds: $`\sigma _t(akg)=\sigma _t(ak)`$.
If $`g=a`$ by similar computations the left hand side becomes $`\sigma _t(km)`$, with $`k`$ and $`m`$ twisted primitives. The right hand side becomes $`\sigma _t(gkm)=\sigma _t(km)`$ for $`k`$ and $`m`$ twisted primitives.
If $`g=k`$ the right hand side is $`\sigma _t(agm)`$ and the left hand side is $`\sigma _t(agm)`$. Again they coincide for $`a`$ and $`m`$ twisted primitives.
4. $`g`$ does not appear in the triple.
Then $`\sigma _t(k_{(1)}m_{(1)})\sigma _t(ak_{(2)}m_{(2)})=0`$ because if $`k_{(2)}`$ and $`m_{(2)}`$ are both twisted primitives, their product involves an $`h^2`$ hence it is zero, if one is twisted primitive and the other grouplike, then $`\sigma _t(k_{(1)}m_{(1)})=0`$ and if they are both grouplikes $`\sigma _t(ak_{(2)}m_{(2)})=0`$. Similarly one shows that the right hand side is also equal to zero.
Hence $`\sigma _t`$ is a left $`2`$-cocycle. Similarly one can prove that $`\sigma _t`$ is also a right $`2`$-cocycle.
We claim that $`\nu _t(H_4H_4)^{}`$ is the convolution inverse of $`\sigma _t`$, where $`\nu _t`$ is defined on the basis elements as follows:
$$\nu _t(x1)=\nu _t(1x)=\epsilon (x)\text{for every }xH_4\text{.}$$
$$\nu _t(gg)=1$$
$$\nu _t(hh)=\nu _t(ghh)=\nu _t(hgh)=\nu _t(ghgh)=\frac{t}{2}$$
and $`\nu (xy)=0`$ whenever $`x`$ is grouplike and $`y`$ is twisted primitive or the other way around.
We show that $`\sigma _t\nu _t=\epsilon \epsilon =\nu _t\sigma _t`$. This means that we have to show that for every pair of basis elements $`a`$ and $`b`$ in $`H_4`$ one has
$$\sigma _t(a_{(1)}b_{(1)})\nu _t(a_{(2)}b_{(2)})=\epsilon (a)\epsilon (b)=\nu _t(a_{(1)}b_{(1)})\sigma _t(a_{(2)}b_{(2)}).$$
Again we divide the different cases:
* $`a`$ and $`b`$ are both grouplikes i.e. $`a,b\{1,g\}`$. Then the expression becomes
$$\sigma _t(ab)\nu _t(ab)=11=\epsilon (a)\epsilon (b)=\nu _t(ab)\sigma _t(ab)$$
* $`a`$ and $`b`$ are one grouplike and the other twisted primitive. Then the expressions are always zero because in each summand there will be either a $`\sigma _t(xy)`$ or a $`\nu _t(xy)`$ with one element grouplike and the other twisted primitive. Hence
$$\sigma _t(a_{(1)}b_{(1)})\nu _t(a_{(2)}b_{(2)})=0=\epsilon (a)\epsilon (b)=\nu _t(a_{(1)}b_{(1)})\sigma _t(a_{(2)}b_{(2)})$$
* $`a`$ and $`b`$ are both twisted primitive. The only nonzero terms in the sums will be those where both $`a_{(1)}`$ and $`b_{(1)}`$ are both grouplikes or both twisted primitives. We have for $`a=h=b`$:
$$\sigma _t\nu _t(hh)=\sigma _t(11)\nu _t(hh)+\sigma _t(hh)\nu _t(gg)+0=0=\epsilon (h)^2$$
$$\nu _t\sigma _t(hh)=\nu _t(11)\sigma _t(hh)+\nu _t(hh)\sigma _t(gg)=0=\epsilon (h)^2.$$
For $`a=h`$ and $`b=gh`$:
$$\sigma _t\nu _t(hgh)=\sigma _t(1g)\nu _t(hgh)+\sigma _t(hgh)\nu _t(g1)=0=\epsilon (h)\epsilon (gh)$$
$$\nu _t\sigma _t(hgh)=\nu _t(1g)\sigma _t(hgh)+\nu _t(hgh)\sigma _t(g1)=0=\epsilon (h)\epsilon (gh).$$
For $`a=gh`$ and $`b=h`$:
$$\sigma _t\nu _t(ghh)=\sigma _t(ghh)\nu _t(1g)+\sigma _t(1g)\nu _t(ghh)=0=\epsilon (gh)\epsilon (h)$$
$$\nu _t\sigma _t(ghh)=\nu _t(ghh)\sigma _t(1g)+\nu _t(1g)\sigma _t(ghh)=0=\epsilon (gh)\epsilon (h).$$
Finally for $`a=b=gh`$
$$\sigma _t\nu _t(ghgh)=\sigma _t(gg)\nu _t(ghgh)+\sigma _t(ghgh)\nu _t(11)=0=\epsilon (gh)^2$$
$$\nu _t\sigma _t(ghgh)=\nu _t(gg)\sigma _t(ghgh)+\nu _t(ghgh)\sigma _t(11)=0=\epsilon (gh)^2$$
So $`\nu _t=\sigma _t^1`$. Hence it makes sense to compute the product in $`{}_{\sigma _t}{}^{}H_{\sigma _t^1}^{}`$. We shall see that the product in $`{}_{\sigma _t}{}^{}H_{\sigma _t^1}^{}`$ coincides with the product in $`H_4`$. We check this fact on products of the the generators $`h`$ and $`g`$: the other products follow by associativity.
$$\overline{g}\overline{g}=\sigma _t(gg)\overline{g^2}\sigma _t^1(gg)=1;$$
$$\overline{g}\overline{h}=\sigma _t(g1)\overline{g}\sigma _t^1(gh)+\sigma _t(g1)\overline{gh}\sigma _t^1(gg)+$$
$$\sigma _t(gh)\overline{g^2}\sigma _t^1(gg)=\overline{gh};$$
$$\overline{h}\overline{g}=0+\sigma _t(1g)\overline{hg}\sigma _t^1(gg)=\overline{hg};$$
$$\overline{h}\overline{h}=\sigma _t(11)\overline{1}\sigma _t^1(hh)+\sigma _t(11)\overline{h}\sigma _t^1(hg)+$$
$$\sigma _t(1h)\overline{g}\sigma _t(hg)+\sigma _t(11)\overline{h}\sigma _t^1(gh)+\sigma _t(11)\overline{h^2}\sigma _t^1(gg)=$$
$$\sigma _t(1h)\overline{hg}\sigma _t^1(gg)+\sigma _t(h1)\overline{g}\sigma _t^1(gh)+\sigma _t(h1)\overline{gh}\sigma _t^1(gg)+$$
$$\sigma _t(hh)\overline{g^2}\sigma _t^1(gg)=\frac{t}{2}+0+\frac{t}{2}=0.$$
Hence the product in $`{}_{\sigma _t}{}^{}(H_4)_{\sigma _t^1}^{}`$ coincides with the product in $`H_4`$.
Now we compute how $`r_s`$ changes under twisting, i.e. $`\sigma _t\tau r_s\sigma _t^1`$ for every $`s`$ and $`t`$ in $`k`$. We shall prove that this is equal to $`r_{ts}`$.
It is known that $`\sigma _t\tau r_s\sigma _t^1`$ is a universal $`R`$-form for the twist of $`H_4`$, hence for $`H_4`$. By the properties of an $`R`$-form:
$$r(a1)=r(1a)=\epsilon (a)\text{and}$$
$$r(abc)=r(ac_{(1)})r(bc_{(2)})\text{for every }a,b\text{ and }cH_4$$
it is enough to check the equality when the first argument is $`h`$ or $`g`$. Moreover, since $`\mathrm{\Delta }(h)`$ and $`\mathrm{\Delta }(g)`$ can be expressed in terms of tensor products of $`h`$, $`g`$ and $`1`$, the property of any $`R`$-form $`r`$
$$r(abc)=r(a_{(2)}b)r(a_{(1)}c)\text{for every }a,b\text{ and }cH_4$$
implies that it is enough to check that the two forms coincide on $`hh`$, $`gg`$, $`gh`$ and $`hg`$. Then
$$\sigma _t\tau r_s\sigma _t^1(gg)=r_{ts}(gg)=1;$$
$$\sigma _t\tau r_s\sigma _t^1(gh)=\sigma _t\tau r_s\sigma _t^1(hg)=0=r_{ts}(hg)=r_{ts}(gh)$$
because every summand will involve one of the forms evaluated at a pair composed by the grouplike $`g`$ and the twisted primitive $`h`$.
$$\sigma _t\tau r_s\sigma _t^1(hh)=\sigma _t(hh)r_s(gg)\sigma _t^1(gg)+\sigma _t(1h)r_s(gh)\sigma _t^1(gg)+$$
$$\sigma _t(1h)r_s(g1)\sigma _t^1(gh)+\sigma _t(h1)r_s(hg)\sigma _t^1(gg)+$$
$$\sigma _t(11)r_s(hh)\sigma _t^1(gg)+\sigma _t(11)r_s(h1)\sigma _t^1(gh)+$$
$$\sigma _t(h1)r_s(1g)\sigma _t^1(hg)+\sigma _t(11)r_s(1h)\sigma _t^1(hg)+$$
$$\sigma _t(11)r_s(11)\sigma _t^1(hh)=\sigma _t(hh)+r_s(hh)+\sigma _t^1(hh)=$$
$$t+s=r_{ts}(hh).$$
In particular, for $`s=t`$ one has $`\sigma _s\tau r_s\sigma _s^1=r_0`$.
###### Remark 4.1
It is a result by Majid (see or ) that if two 2-cocycles are cohomologous then their corresponding twisted Hopf algebras are isomorphic. If the Hopf algebra involved is dual quasitriangular then the corresponding twisted Hopf algebras will be isomorphic as dual quasitriangular Hopf algebras. In particular if a $`2`$-cocycle is a coboundary (i.e. it is cohomologous to $`\epsilon \epsilon `$) one obtains the same dual quasitriangular Hopf algebra he started with. The computations above show that the converse is not true, i.e. there are $`2`$-cocycles which are not coboundaries for which the twist does not change the Hopf algebra structure. For $`t0`$ $`\sigma _t`$ is not a coboundary because otherwise $`(H_4,r_t)`$ would be isomorphic to $`(H_4,r_0)`$ as dual triangular Hopf algebras which is never true by the results in .
We have proved the following
###### Theorem 4.2
Let $`k`$ be a field of characteristic different from $`2`$. Let $`H_4`$ be Sweedler’s Hopf algebra, and $`r_t`$, $`tk`$ be its universal $`R`$-forms. Then for every $`tk`$, $`BC(k,H_4,r_t)BC(k,H_4,r_0)`$. The isomorphism maps the class of $`A`$ in $`BC(k,H_4,r_0)`$ to the class of $`A_{\sigma _t\tau }`$ in $`BC(k,H_4,r_t)`$.
###### Remark 4.3
If $`k`$ is algebraically closed, the isomorphism between $`BC(k,H_4,r_s)`$ and $`BC(k,H_4,r_t)`$ for $`st0`$ could be deduced a priori by the fact that all the dual quasitriangular structures are isomorphic for $`st0`$, see . But $`(H_4,r_s)\simeq ̸(H_4,r_0)`$ for $`s0`$. Still, we could show that the corresponding Brauer groups are isomorphic.
###### Remark 4.4
Theorem 4.2 combined with the results in and self-duality of $`H_4`$ provide a full descripition of $`BC(k,H_4,r_t)`$ and $`BM(k,H_4,R_t)`$. This group is the direct sum of $`(k,+)`$ and the Brauer-Wall group of $`k`$ (which is the Brauer group of the category of $`𝐙_2`$-graded $`k`$-modules). The computation of the full Brauer group $`BQ(k,H_4)`$ and the determination of how the different copies of $`BC`$ fit into $`BQ`$ in this case is still an open problem.
## 5 The Brauer groups of $`H^{}`$ and $`H^{op}`$
In this section we investigate the relation between $`BQ(k,H)`$ and $`BQ(k,H^{})`$ and between $`BQ(k,H)`$ and $`BQ(k,H^{op})`$. We start with a variation of a Lemma to be found in .
###### Lemma 5.1
For a finitely generated projective Hopf algebra $`H`$ over $`k`$ the usual flip $`\tau `$ defines an isomorphism between $`D(H)`$ and $`D(H^{op,cop,})^{op}`$. The standard $`R`$-matrix $``$ is mapped to $`(\tau _{13}\tau _{24})^{}`$ where $`^{}`$ is the standard $`R`$-matrix of $`D(H^{op,cop})`$. In particular, if the antipode $`S`$ of $`H`$ is bijective, then $`\tau (S^1S)`$ defines an isomorphism between $`D(H)`$ and $`D(H^{})^{op}`$ mapping the universal $`R`$-matrix $``$ to $`\tau _{D(H^{}),D(H^{})}^{}`$.
This Lemma implies that the categories of left modules of the pairs $`(D(H),)`$ and $`(D(H^{})^{op},^{})`$ are equivalent. Hence the categories of left modules of $`D(H)`$ and $`D(H^{})`$ are anti-equivalent braided monoidal categories. Dualizing the above result and using the discussion in the previous sections and Remark 3.2 we have:
###### Proposition 5.2
Let $`H`$ be a faithfully projective Hopf algebra with bijective antipode $`S`$. Then $`BQ(k,H)BQ(k,H^{})`$.
Proof: $`BQ(k,H)`$ is isomorphic to:
$$BC(k,D(H)^{},r)BC(k,D(H^{})^{op},r^{}\tau )BC(k,D(H^{})^{cop},r^{}\tau ).$$
Applying the antipode of $`D(H^{})^{}`$ and observing that the $`R`$-form $`r^{}\tau `$ remains unchanged one has:
$$BQ(k,H)BC(k,D(H^{})^{op},r^{}\tau )BC(k,D(H^{})^{},r^{})BQ(k,H^{}).$$
Observe that the above is in fact an anti-isomorphism. We describe it more explicitely generalizing the main result in . We give it in terms of Yetter-Drinfel’d modules and in terms of modules for the Drinfel’d double.
Suppose $`(M,,\chi )`$ is a Yetter-Drinfel’d $`H`$-module. Then $`(M,)`$ is a $`D(H)`$-module. The pull-back of this action along the map $`(\tau (S^1S))^1=(S^{}S^1)\tau `$ defines on $`M`$ a $`D(H^{})^{op}`$-module structure that is given by $`(a\epsilon ).m=S^1(a)m`$ for elements of the subalgebra $`H^{op}\epsilon `$ and by $`(1\xi ).m=S^{}(\xi )m`$ for elements of the sublagebra $`1H^{,op}`$. The antipode of $`D(H^{})^{op}`$ is $`S^1\mathrm{id}`$ on $`\epsilon H^{op}`$ and $`\mathrm{id}S^{}`$ on $`1H^{op}`$. Using the action defined in Remark 3.2 the $`H^{}`$-action on $`M`$ is given by $``$ and $``$. Therefore the map in Proposition 5.2 sends a Yetter-Drinfel’d $`H`$-module $`(M,,\chi )`$ to the Yetter-Drinfel’d $`H^{}`$-module $`(M,\rho ,)`$ where $`\rho (m)=m_{(0)}m_{(1)}MH^{}`$ is such that for every $`lHH^{}`$ one has $`(\mathrm{id}<,l>)\rho (m)=lm`$. This is possible because $`M`$ is rational.
In other words, the functor from objects in $`𝒴𝒟(H)`$ to objects in $`𝒴𝒟(H^{})`$ is given by the standard functors from the category of right $`H`$-comodules $`^H`$ to the category of left $`H^{}`$-modules $`{}_{H^{}}{}^{}`$ and from the category of left $`H`$-modules $`{}_{H}{}^{}`$ to the category of right $`H^{}`$-modules $`^H^{}`$ associating to a right comodule $`(M,\chi )`$ the left $`H^{}`$-module $`(M,)`$ where $`h^{}m=(\mathrm{id}<h^{},>)\chi (m)`$ and to the $`H`$-module $`(M,)`$ the right $`H^{}`$-comodule $`(M,\rho )`$ where $`\rho (m)=m_{(0)}m_{(1)}`$ as above. One can also check directly that this functor $`𝒟`$ maps Yetter-Drinfel’d modules of $`H`$ to Yetter-Drinfel’d modules of $`H^{}`$ and that it defines an anti-equivalence of braided categories. For instance, the compatibility condition between $`\rho `$ and $``$ for the $`H^{}`$ module $`(M,)`$ and comodule $`(M,\rho )`$ follows from
$$(\mathrm{id}<,l>)h_{(1)}^{}m_{(0)}h_{(2)}^{}m_{(1)}$$
$$=h_{(1)}^{}(m_{(0)}<m_{(1)},l_{(2)}>)<h_{(2)}^{},l_{(1)}>$$
$$=h_{(1)}^{}(l_{(2)}m)<h_{(2)}^{},l_{(1)}>$$
$$=(l_{(2)}m)_{(0)}<h_{(1)}^{},(l_{(2)}m)_{(1)}><h_{(2)}^{},l_{(1)}>$$
$$=(\mathrm{id}<h^{},>)(l_{(2)}m)_{(0)}(l_{(2)}m)_{(1)}l_{(1)}$$
for any $`h^{}H^{}`$, $`mM`$ and $`lHH^{}`$ together with the compatibility condition for the Yetter-Drinfel’d $`H`$-module $`M`$.
We have seen that $`𝒟`$ defines a monoidal functor from $`𝒴𝒟(H)`$ to $`𝒴𝒟(H^{})^{op}`$, the monoidal category whose tensor structure is given by $`\stackrel{~}{}^{op}`$, where $`M\stackrel{~}{}_{}^{op}N=N\stackrel{~}{}_{}M`$ is the opposite product of $`N`$ and $`M`$ viewed as Yetter-Drinfel’d $`H^{}`$-modules. The family of natural isomorphisms
$$\tau (M,N):𝒟(M)\stackrel{~}{}_{}^{op}𝒟(N)=𝒟(N)\stackrel{~}{}_{}𝒟(M)𝒟(M\stackrel{~}{}N)$$
compatible with $`𝒟`$ and the tensor products is given by the usual flips $`\tau `$. In fact one can also check directly that $`\tau `$ is a module and comodule isomorphism. For instance for $`M`$ and $`N`$ objects in $`𝒴𝒟(H)`$ the left $`H^{}`$-module structure on $`M\stackrel{~}{}N`$ is
$`h^{}\tau (n\stackrel{~}{}m)={\displaystyle m_{(0)}n_{(0)}}<h^{},n_{(1)}m_{(1)}>`$ (5.1)
$`={\displaystyle }m_{(0)}<h_{(2)}^{},m_{(1)}>n_{(0)}<h_{(1)}^{},n_{(1)}>`$ (5.2)
$`={\displaystyle (h_{(2)}^{}_Mm)(h_{(1)}^{}_Nn)}`$ (5.3)
where $`_M`$ and $`_N`$ denote the $`H^{}`$ left module structure associated to the right $`H`$-comodule structure of $`M`$ and $`N`$. The right $`H^{}`$-comodule structure $`\rho `$ of $`M\stackrel{~}{}N`$ is such that for every $`lH`$ one has
$$(\mathrm{id}^2<,l>)\rho (m\stackrel{~}{}n)=l(m\stackrel{~}{}n)=$$
$$(l_{(1)}m)\stackrel{~}{}(l_{(2)}n)=(\mathrm{id}l_{(1)})(\rho _M(m))(\mathrm{id}l_{(2)})(\rho _N(n))$$
$$=(\mathrm{id}^2<,l>)m_{(0)}n_{(0)}m_{(1)}n_{(1)}$$
where $`\rho _M(m)=m_{(0)}m_{(1)}`$ and $`\rho _N(n)=n_{(0)}n_{(1)}`$, hence $`\tau `$ is a comodule isomorphism.
One can also see directly that $`𝒟`$ and $`\tau `$ are compatible with the braidings $`\varphi `$ in $`𝒴𝒟(H)`$ and $`\psi _{MN}^{op}`$ in $`𝒴𝒟(H^{})^{op}`$ where $`\psi _{MN}^{op}`$ is $`\psi _{NM}:N\stackrel{~}{}_{}MM\stackrel{~}{}_{}N`$ as Yetter-Drinfel’d $`H^{}`$-modules. In fact for any pair of objects $`M`$ and $`N`$ in $`𝒴𝒟(H)`$ and for $`nm𝒟(M)\stackrel{~}{}^{op}𝒟(N)`$,
$$𝒟(\varphi _{MN})\tau _{NM}(nm)=\varphi _{MN}(mn)=n_{(0)}n_{(1)}m$$
coincides with
$$\tau _{MN}\psi _{MN}^{op}(nm)=\tau _{MN}m_{(0)}<m_{(1)},n_{(1)}>n_{(0)}=n_{(0)}n_{(1)}m.$$
In terms of $`𝒟`$ and $`\tau `$ one sees that if $`A`$ is a Yetter-Drinfel’d $`H`$-module algebra, the product in $`𝒟(A)`$ is given by $`𝒟(m_A)\tau _{A,A}`$, where $`m_A`$ is the product in $`A`$, i.e. $`A`$ equipped with the opposite product is a Yetter-Drinfel’d $`H^{}`$-module algebra as we had seen before. We have that $`\tau `$ defines an algebra isomorphism between $`𝒟(B\mathrm{\#}A)=(B\mathrm{\#}A)^{op}`$ and $`𝒟(B)^{op}\mathrm{\#}_{}^{op}𝒟(A)A^{op}\mathrm{\#}_{}B^{op}`$ where $`\mathrm{\#}_{}`$ the tensor product of Yetter-Drinfel’d $`H^{}`$-module algebras. This can be checked directly for every $`a,cA`$ and $`b,dB`$:
$$m_{(B\mathrm{\#}A)^{op}}(\tau \tau )((a\mathrm{\#}_{}b)(c\mathrm{\#}_{}d))=d\varphi _{AB}(cb)a=$$
$$\tau \left(a\tau \varphi _{AB}(cb)d\right)=\tau \left(a(\psi _{BA}\tau (cb))d\right)$$
$$=\tau \left(a(\psi _{BA}(bc))d\right)=\tau m_{A^{op}\mathrm{\#}_{}B^{op}}((a\mathrm{\#}_{}b)(c\mathrm{\#}_{}d))$$
where $``$ denotes the opposite product in $`A`$ and $`B`$ and the interchange between $`\tau \varphi _{AB}`$ and $`\varphi _{BA}\tau `$ follows by the compatibility of $`𝒟`$ and $`\tau `$ with the braidings.
###### Proposition 5.3
Let $`H`$ be a finitely generated projective Hopf algebra over $`k`$ with bijective antipode. Then the isomorphism between $`BQ(k,H)`$ and $`BQ(k,H^{})`$ of Proposition 5.2 maps $`[A][\overline{𝒟(A)}]_{}`$. Here $`[B]_{}`$ denotes the equivalence class of $`B`$ as $`H^{}`$-Azumaya algebra.
For sake of completeness we show also the (functorial) reciprocity for the opposite algebra, $`F`$ and $`G`$ with respect to $`𝒟`$. $`\overline{𝒟(A)}`$ has product $`m_{𝒟(A)}\psi _{AA}^{op}=m_A\tau \psi _{AA}=m_A\varphi _{AA}\tau =m_{𝒟(\overline{A})}`$. Hence $`𝒟(\overline{A})\overline{𝒟(A)}`$. As far as the canonical maps in $`𝒴𝒟(H^{})`$ $`F_{}`$ and $`G_{}`$ are concerned, $`F_{}`$ is the composite:
$$𝒟(A)\mathrm{\#}_{}^{op}\overline{𝒟(A)}=𝒟(A)\mathrm{\#}_{}𝒟(\overline{A})\stackrel{\tau }{}𝒟(A\mathrm{\#}\overline{A})\stackrel{𝒟\left(F\right)}{}𝒟(End(A))=End(𝒟(A))^{op}$$
and similarly for $`G_{}`$
$$\overline{𝒟(A)}\mathrm{\#}_{}^{op}𝒟(A)=𝒟(\overline{A})\mathrm{\#}_{}𝒟(A)\stackrel{\tau }{}𝒟(\overline{A}\mathrm{\#}A)\stackrel{𝒟\left(G\right)}{}𝒟(End(A)^{op})=End(𝒟(A)).$$
It is a small exercise in sigma notation to check that $`F_{}(a\mathrm{\#}_{}b)(c)=G(\overline{b}\mathrm{\#}a)(c)`$, $`G_{}(\overline{b}\mathrm{\#}_{}a)(c)=F(a\mathrm{\#}\overline{b})(c)`$ and that $`𝒟(End(A))`$ is indeed $`End(A)^{op}`$ with the natural Yetter-Drinfel’d $`H^{}`$-module structure.
###### Corollary 5.4
For a finitely generated projective Hopf algebra $`H`$ over a commutative ring $`k`$, with bijective antipode the isomorphism in Proposition 5.3 induces an isomorphism between $`BQS(k,H)`$ and $`BQS(k,H^{})`$.
Proof: It is immediate to check that $`k`$ is a direct summand of $`A`$ as a Yetter-Drinfel’d $`H`$-module if and only if $`k`$ is a direct summand of $`A`$ as a Yetter-Drinfel’d $`H^{}`$-module.
###### Corollary 5.5
If $`H`$ is commutative or cocommutative then all the “associated” Hopf algebras (opposite, co-opposite, opposite of the dual, co-opposite of the dual, etcetera) have the same Brauer group.
It is an interesting question whether an isomorphism between $`BQ(k,H)`$ and $`BQ(k,H^{op})`$ holds in general. Here follow some observations concerning this question.
###### Lemma 5.6
The linear map
$$S_H^{}\mathrm{id}:D(H^{cop})D(H)^{cop}$$
is a Hopf algebra isomorphism. If for $`D(H)^{cop}`$ we have the $`R`$-matrix $`\tau _{D(H)}`$ then the corresponding $`R`$-matrix in $`D(H^{cop})`$ is $`\tau _{D(H^{cop})}^1`$.
Proof: Direct computation.
###### Proposition 5.7
For any finitely generated projective Hopf algebra $`H`$, there is an isomorphism between $`BQ(k,H)`$ and $`BM(k,D(H^{cop}),\tau _{D(H^{cop})}^1)`$.
Proof: One has
$$BQ(k,H)BM(k,D(H),R)BM(k,D(H)^{cop},\tau )$$
where the second isomorphism is clear because the algebra structure is unchanged. By Lemma 5.6
$$BQ(k,H)BM(k,D(H^{cop}),\tau _{D(H^{cop})}^1)BM(k,D(H^{op}),\tau _{D(H^{op})}^1)$$
where the second isomorphism is induced by the isomorphism $`S_H^1S_H^{}`$ from $`D(H^{op})`$ to $`D(H^{cop})`$ that leaves the $`R`$-matrix invariant.
Under the above series of maps (coming from equivalences of categories) the Yetter-Drinfel’d $`H`$-module $`(M,,\chi )`$ is mapped to the Yetter-Drinfel’d $`H^{op}`$-module $`(M,,\chi )`$ with $`hm:=S(h)m`$ where $`S`$ is the antipode of $`H`$. In fact
$$\chi (hm)=\chi (Shm)=(Sh)_{(2)}m_{(0)}(Sh)_{(3)}m_{(1)}S^1(Sh)_{(1)}=$$
$$S(h)_{(2)}m_{(0)}S(h_{(1)})m_{(1)}h_{(3)}=S(h)_{(2)}m_{(0)}h_{(3)}m_{(1)}S_{H^{op}}^1(h_{(1)})$$
where $``$ denotes the opposite product. This construction defines a functor which, together with the natural transformation $`\tau `$ gives an anti-equivalence of monoidal categories. The natural isomorphism $`\tau `$ would respect the braiding if for every Yetter-Drinfel’d module $`M`$ and $`N`$, for every $`mM`$ and $`nN`$ one had
$$m_{(0)}m_{(1)}n=(Sn_{(1)})mn_{(0)}.$$
In terms of modules over Drinfel’d quantum double of $`H`$,
$$m_{(0)}m_{(1)}n=\tau ((nm))$$
if $`=_1_2`$ is the canonical $``$-matrix for $`D(H)`$. If $`P=P_1P_2`$ is $`\tau _{D(H),D(H)}^1=\tau _{D(H),D(H)}(\mathrm{id}S_{D(H)}^1)`$, then
$$(Sn_{(1)})mn_{(0)}=\tau (P(nm)).$$
Hence $`\tau `$ would respect the braidings if $`D(H)`$ were triangular. Then $`BQ(k,H)`$ would be isomorphic to $`BQ(k,H^{op})`$. However, for $`k`$ a field and $`H`$ nontrivial, $`(D(H),)`$ is never triangular. An easy way to see this is given by the theory of the exponent of a finite-dimensional Hopf algebra introduced by P. Etingof and S. Gelaki in . The authors show the so-called exponent of a Hopf algebra $`H`$ is equal to the order of $`(\tau )`$ in $`D(H)^2`$ and also equal to the order of the element $`u=m(S_{D(H)}\mathrm{id}^2)\tau ()=S^1(h_j^{})h_j`$ in $`D(H)`$. Here $`\{h_j^{}\}_{jJ}`$ and $`\{h_j\}_{jJ}`$ are dual bases and $`S^{}`$ is the antipode of $`H^{}`$. It is clear then that $`(\tau )\epsilon 1\epsilon 1`$ because $`u\epsilon 1`$.
Observe that triangularity of $``$ is not a necessary condition for $`BQ(k,H)BQ(k,H^{op})`$ to hold. This is clear by looking at the case of $`H`$ commutative. In particular Proposition 5.7 implies that if $`H`$ is commutative then there is an isomorphism between $`BC(k,D(H),)`$ and $`BC(k,D(H),\tau ^1)`$.
ACKNOWLEDGEMENTS
This work has been carried out at the University of Antwerp where the author had a post-doc position financed by the EC network “Algebraic Lie Representations”contract ERB-FMRX-CT97-0100. The author wishes to thank University of Antwerpen for the warm hospitality and Professor Fred Van Oystaeyen, Dr. Juan Cuadra and Dr. Yinhuo Zhang for valuable comments and useful discussions.
|
warning/0005/nucl-th0005048.html
|
ar5iv
|
text
|
# Algebraic Model for Quantum Scattering. Reformulation, Analysis and Numerical Strategies
## 1 Introduction
In the quest for solving the Schrödinger equation for both bound and continuum states, square-integrable bases have been repeatedly used. For bound states this turns out to be a traditional way to obtain the spectral properties of quantum systems. It has been shown however that a single representation of the Schrödinger equation in terms of an $`L^2`$ basis can be formulated that allows for a description of both bound and continuum states .
A version of such a formulation is called the Algebraic Model of the Resonating Group Method (hereafter referred to as the Algebraic Model or AM). It was originally tailored to treat clusterized problems, but can be used for all kinds of quantum mechanical many-particle configurations without major modifications. The AM has been formulated in terms of different types of bases, depending on the more specific features of the quantum system considered. One very important feature of the AM is the fact that the boundary conditions of the system are translated from a coordinate space context to the context of expansion coefficients, and are explicitly incorporated in the dynamical equations .
We will consider a specific AM formulation (i.e. a specific $`L^2`$ basis) to elucidate the analysis of the method. The methodology used for this analysis is of a general nature however, and can be repeated for other bases. The specific AM version chosen in this paper features an oscillator basis, and is in particular very suitable for obtaining both the bound and continuum spectra of nuclear systems with very different configurational properties; where appropriate, nuclear spectroscopic units will therefore be used. The choice of this AM version is mainly due to the background of the authors. Results of the AM approach considered here have already been reported on , and show in particular how the coupling of cluster and collective configurations can be treated seamlessly in such a description.
Where necessary, a specific form for the potential operator will be used. We will consider a Gaussian form in this work. Although this again is a popular potential form in nuclear spectroscopic calculations, it is also an interesting functional form to approximate a large variety of potentials by discrete and continuous superpositions.
We will concentrate our attention to the solutions of the dynamical equations for scattering situations only, as these are much more involved than the bound state problems. Indeed, the latter can be well approximated by a simple diagonalization of the energy matrix, as is well known.
We will specifically discuss how strongly the convergence of the solutions of an AM system depend on the parameters of the problem. In particular the dependence on the precision and the convergence properties of the solutions of both the oscillator radius of the basis (the parameter unambiguously fixing the square-integrable basis) and the form of the potential energy contribution, will be treated in detail.
The convergence is crucial for obtaining stable approximations to the solutions of problems expressed in terms of an (infinite) set of basis functions, in our case an $`L^2`$ basis. This problem was repeatedly investigated, mainly for bound-state solutions. As applications to scattering problems, expressed in an $`L^2`$ basis, appeared, several algorithms were suggested to accelerate the convergence of the results within a restricted subset of the basis. For instance, Heller and Yamani used “Kato correction”, and Revai et al. introduced the “Lanczos factor”. A more intuitive approximation was proposed by Fillipov et al. .
The analysis of the AM equations presented in this work will be shown to lead to (1) an optimal choice for the basis, given the potential, yielding well-converging and stable solutions of the AM system, or, if the optimal basis cannot be used, to (2) algorithmic procedures for solving the AM system in an acceptable and controlled approximation. These algorithmic procedures will depend on the specific choice made for the basis (i.e. the oscillator radius chosen), but also on information on the asymptotic behaviour of the expansion coefficients, as well as on the dynamical equations themselves. In this way the algorithm used will depend on the physical properties of the system considered.
In a forthcoming paper the different strategies for solving AM equations will be applied explicitly to a number of problems from nuclear, atomic and molecular physics.
## 2 The AM in an Oscillator Basis
Choosing an oscillator basis to describe some specific Hilbert subspace in which to solve the Schrödinger equation leads to the following form of the latter in terms of a system of linear equations:
$$\underset{m=0}{\overset{\mathrm{}}{}}<n\widehat{H}Em>c_m=0$$
(1)
where the coefficients $`c_m`$ are the expansion coefficients in the oscillator basis of the wave function corresponding to the energy E
$$\mathrm{\Psi }_E>=\underset{n=0}{\overset{\mathrm{}}{}}c_nn>,$$
(2)
and subject to a typical boundary condition. For simplicity we have omitted the angular momentum quantum number as well as the energy dependence of the coefficients $`c_n`$.
The boundary conditions of quantum systems are traditionally expressed in coordinate space, but can also be formulated in terms of the expansion coefficients $`c_n`$. Indeed, for very large $`n`$, the dynamical equations are reduced to a simple (in an oscillator basis a three-term recurrence) form containing the kinetic energy operator solely. The equations can therefore be solved analytically for very large $`n`$ . The above mentioned reduction of the equations only involves the supposition that the potential energy matrix elements vanish for very large $`n`$, which has been shown to be an acceptable approximation for relatively short-range interactions.
One obtains the following asymptotic behaviour for bound states:
$`c_n^{(as)}`$ $``$ $`\sqrt{R_n}\mathrm{exp}(\kappa R_n)/R_n`$ (3)
$`(\kappa `$ $`=`$ $`\sqrt{2mE/\mathrm{}^2})`$
and for continuum states:
$`c_n^{(as)}`$ $``$ $`\sqrt{R_n}(j_L(kR_n)+\mathrm{tan}(\delta )n_L(kR_n))`$ (4)
$`(k`$ $`=`$ $`\sqrt{2mE/\mathrm{}^2})`$
where $`j_L`$ and $`n_L`$ are the traditional Bessel and Neumann special functions.
A striking resemblance with the asymptotic forms of a wave function in coordinate representation is observed, by replacing the radial coordinate in the latter by the discrete value $`R_n=\sqrt{4n+2L+3}`$. A heuristic argument for this observation is that $`R_n`$ corresponds to the turning point of the oscillator in state $`n>`$ with angular momentum $`L`$.
## 3 A Simple Scheme for Solving the AM System of Equations
In the previous section we introduced the infinite dimensional system of linear equations obtained from the AM formulation, to be solved subject to a proper boundary condition. As indicated earlier we will concentrate on the scattering situation, and therefore only consider the asymptotic behaviour of the $`c_n`$ corresponding to the continuum boundary condition.
A scheme for solving the linear system of equations quite naturally presents itself. Under the assumption that the matrix elements potential energy $`<iVj>`$ vanish for sufficiently large values of one of the basis state indices $`i`$ or $`j`$, one chooses a limiting value $`N`$ to set this vanishing point in terms of the basis states. In the region where the potential matrix elements are neglected, the expansion coefficients $`c_n`$ are then given by (4), and can be written as
$$c_n=c_n^{(+)}+\mathrm{tan}(\delta )c_n^{()}(nN)$$
(5)
assuming $`N`$ to be sufficiently large, with
$`c_n^{(+)}`$ $`=`$ $`\sqrt{R_n}j_L(kR_n)(nN)`$
$`c_n^{()}`$ $`=`$ $`\sqrt{R_n}n_L(kR_n)(nN)`$ (6)
The choice for $`N`$ divides the linear system in three different regions:
* a finite number of N equations with $`0<n<N`$, in which the potential energy matrix elements are fully taken into account. These equations correspond to the “internal region” in terms of the basis states
* an infinite number of equations corresponding to $`n>N`$, in which the potential energy matrix elements are neglected. These equations correspond to the “asymptotic region”, and are trivially fulfilled due to the boundary condition
* the equation with $`n=N`$, in which the potential matrix elements are neglected. This equation corresponds to the “matching condition”, as it couples the internal region through coefficient $`c_{N1}`$ with the asymptotic region through the phase-shift $`\delta `$
This scheme amounts to solving the following $`N+1`$ dimensional system of linear equations for the $`N`$ coefficients $`c_n`$ with $`n=0,1,..,N1`$, and the phase-shift $`\delta `$:
$`\left(\begin{array}{cccc}H_{00}E& \mathrm{}& H_{0,N1}& 0\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ H_{N1,0}& \mathrm{}& H_{N1,N1}E& T_{N1,N}c_N^{()}\\ 0& \mathrm{}& T_{N,N1}& 𝒯_N^{()}\end{array}\right)\times \left(\begin{array}{c}c_0\\ \mathrm{}\\ c_{N1}\\ \mathrm{tan}(\delta )\end{array}\right)`$ (15)
$`=\left(\begin{array}{c}0\\ \mathrm{}\\ T_{N1,N}c_N^{(+)}\\ 𝒯_N^{(+)}\end{array}\right)`$ (20)
where
$`𝒯_N^{(+)}`$ $`=`$ $`(T_{N,N}E)c_N^{(+)}+T_{N,N+1}c_{N+1}^{(+)}`$
$`𝒯_N^{()}`$ $`=`$ $`(T_{N,N}E)c_N^{()}+T_{N,N+1}c_{N+1}^{()}`$ (21)
and $`T`$ stands for the kinetic energy operator.
As the potential matrix elements do not actually drop to zero exactly for $`nN`$, one should vary the value of $`N`$ to test the stability of the solution. It turns out that this stability strongly depends on the specific problem considered. As an example, the solution of the linear system of equations for a nuclear two-cluster problem, in which the distance of the clusters is considered as the degree of freedom, shows a rapid convergence in terms of $`N`$. A solution for a monopole description of the nucleus, in which the radius of the nucleus is considered to be the prominent degree of freedom, shows a very slow convergence in terms of $`N`$. These results indicate that one should be very careful when omitting potential energy matrix elements, and that a proper study of the form of the equations is necessary.
## 4 An Analysis of the AM Equations
### 4.1 A Reformulation of the AM Equations
To study the properties of the (in principle infinite dimensional) linear system (1) to be solved, we will rewrite this set of equations using the following substitution for the expansion coefficients $`c_n`$:
$$c_n=c_n^{(as)}+c_n^{(0)}$$
(22)
By this substitution, the coefficients $`c_n`$ are considered to represent a deviation from the asymptotic behaviour, i.e. the coefficients $`c_n^{(as)}`$. The coefficient $`c_n^{(0)}`$ then quantifies this deviation, which in particular will be zero in the true asymptotic region (i.e. for very large $`n`$). The first term in (22) is responsible for the long-range behaviour of the system. The second term corresponds to the short-range correction caused by the potential; in other words, in coordinate representation the coefficients $`\{c_n^{(0)}\}`$ would represent that part of the wave-function that is dominated by, and within the range of, the potential. The knowledge of $`c_n^{(0)}`$ as a function of $`n`$ thus provides a key element for determining a proper index $`N`$ distinguishing the internal from the asymptotic region in terms of the basis functions.
Rewriting the original AM linear system of equations (1) in the unknowns $`(\{c_n\})`$, yields an equivalent linear system in the unknowns $`(\{c_n^{(0)}\},\delta )`$, where $`\delta `$ is defined by the following form of the asymptotic expansion coefficients $`c_n^{(as)}`$:
$$c_n^{(as)}=c_n^{(+)}+\mathrm{tan}(\delta )c_n^{()}$$
(23)
Because the asymptotic coefficients now appear for all $`n`$ in this representation, they should be properly defined. In order to do so, we consider the coordinate representation of the outgoing asymptotic wave-functions $`\mathrm{\Psi }^{(+)}`$ and $`\mathrm{\Psi }^{()}`$, which are originally defined as the two linearly independent solutions of
$$(\widehat{T}E)\mathrm{\Psi }=0$$
(24)
where $`\widehat{T}`$ is the kinetic energy operator. $`\mathrm{\Psi }^{(+)}`$ is commonly called the “regular” solution, and behaves properly for all $`r`$. $`\mathrm{\Psi }^{()}`$ is the “irregular” solution, and has an irregular (infinite) behaviour near the origin $`r=0`$. To provide a regular character at the origin for both $`\mathrm{\Psi }^{(+)}`$ and $`\mathrm{\Psi }^{()}`$, we redefine their equations in the following way, as was suggested earlier by Heller and Yamani :
$`(\widehat{T}E)\mathrm{\Psi }^{(+)}`$ $`=`$ $`0`$
$`(\widehat{T}E)\mathrm{\Psi }^{()}`$ $`=`$ $`\beta _0\mathrm{\Phi }_0`$ (25)
The non-zero right-hand side in (25), in which $`\mathrm{\Phi }_0(r)=<r0>`$ represents the zero-quanta oscillator state, and where $`\beta _0`$ equals
$$\beta _0=\mathrm{}\omega \frac{\mathrm{exp}(k^2/2)}{k^{L+1}}\sqrt{\frac{2}{\mathrm{\Gamma }(L+1/2)}}\frac{1}{\mathrm{\Gamma }(L+1/2)}$$
(26)
accounts for a regular behaviour near the origin for the modified $`\mathrm{\Psi }^{()}`$:
$`\mathrm{\Psi }^{()}(kr)\{\begin{array}{cc}j_L(kr)\hfill & \text{for }r1\text{ }\hfill \\ n_L(kr))\hfill & \text{for }r1\hfill \end{array}`$ (29)
In terms of a Fourier representation, using oscillator states as a basis, one then has the following well-defined expansion:
$`\mathrm{\Psi }^{(+)}(r)={\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}<rm>c_m^{(+)}`$
$`\mathrm{\Psi }^{()}(r)={\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}<rm>c_m^{()}`$ (30)
This provides a proper definition for the coefficients $`c_n^{(+)}`$ and $`c_n^{()}`$ in (23), the “regular”, resp. “irregular” asymptotic coefficients, the explicit form of which can be found in . The equations for the asymptotic coefficients in Fourier space are then:
$`{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}<n\widehat{T}Em>c_m^{(+)}`$ $`=`$ $`0,`$
$`{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}<n\widehat{T}Em>c_m^{()}`$ $`=`$ $`\beta _0\delta _{n,0}`$ (31)
One notices an identical form for all equations, except for the single one with $`n=0`$.
Substituting the solutions of (31) for the asymptotic coefficients in the set of equations (1), taking into account the substitutions (22) and (23), then leads to the following linear system:
$`{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}<n\widehat{H}Em>c_m^{(0)}`$ $`+`$ $`\mathrm{tan}(\delta )[\beta _0\delta _{n,0}+{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}<n\widehat{V}m>c_m^{()}]=`$ (32)
$``$ $`{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}<n\widehat{V}m>c_m^{(+)}`$
Equation (32) now shows the influence of the potential energy matrix elements on the behaviour of the system in a clear way.
Let us introduce the “Dynamical Regular and Irregular Coefficients” $`V_n^{(+)}`$ and $`V_n^{()}`$ as follows:
$`V_n^{(+)}`$ $`=`$ $`{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}<n\widehat{V}m>c_m^{(+)}`$
$`V_n^{()}`$ $`=`$ $`{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}<n\widehat{V}m>c_m^{()}`$ (33)
or, in an alternative integral representation in three-dimensional coordinate space:
$`V_n^{(+)}`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}𝑑rr^2\mathrm{\Phi }_n(r)\widehat{V}(r)\mathrm{\Psi }^{(+)}(kr)`$
$`V_n^{()}`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}𝑑rr^2\mathrm{\Phi }_n(r)\widehat{V}(r)\mathrm{\Psi }^{()}(kr)`$ (34)
where we use the coordinate representation of $`\mathrm{\Phi }_n(r)`$, the n-quanta oscillator function $`<rn>`$:
$`\mathrm{\Phi }_n(r)`$ $`=`$ $`(1)^nN_{nL}\rho ^L\mathrm{exp}({\displaystyle \frac{1}{2}}\rho ^2)L_n^{L+\frac{1}{2}}(\rho ^2)(\rho ={\displaystyle \frac{r}{b}})`$ (35)
$`(N_{nL}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2(n!)}{\mathrm{\Gamma }(n+L+3/2)}}}{\displaystyle \frac{1}{b^{3/2}}})`$
Substitution of the Dynamical Coefficients (33) in (32) leads to
$$\underset{m=0}{\overset{\mathrm{}}{}}<n\widehat{H}Em>c_m^{(0)}+\mathrm{tan}(\delta )\left[\beta _0\delta _{n,0}+V_n^{()}\right]=V_n^{(+)}$$
(36)
No approximations have been made so far to obtain this representation of the AM dynamical equations. We have taken into account the main (asymptotic) behaviour of the solutions in the equations, and used a regularization scheme to achieve this in a well-defined way.
To reveal how the AM linear system can be solved in a numerically optimal, or at least acceptable, way, an analysis of the Dynamical Coeeficients $`V_n^{(+)}`$ and $`V_n^{()}`$ seems imperative. It is indeed clear that, if e.g. $`V_n^{(+)}`$ and $`V_n^{()}`$, from a given index $`n`$ on, are sufficiently zero when compared to the other terms in (36), the equations are reduced in a controlled and secure way.
There is another important reason to investigate the behaviour of $`V_n^{(+)}`$. It is indeed well-known from the integral equation formulation of quantum mechanics that:
$$\mathrm{tan}\delta _L=\frac{mk}{\mathrm{}^2}_0^{\mathrm{}}𝑑r\mathrm{\Psi }^{(+)}(kr)\widehat{V}(r)\mathrm{\Psi }(k,r)$$
(37)
where $`\mathrm{\Psi }(k,r)`$ is the exact solution of the Schrödinger equation obtained with potential $`\widehat{V}`$. In an oscillator representation, (37) reads as:
$$\mathrm{tan}\delta _L=\frac{mk}{\mathrm{}^2}\underset{n=0}{\overset{\mathrm{}}{}}V_n^{(+)}c_n$$
(38)
from which we immediately recognize the importance of $`V_n^{(+)}`$ in the convergence of the solutions of the continuous spectrum.
The analysis of $`V_n^{(+)}`$, $`V_n^{()}`$ must be done in terms of the basis index $`n`$, of the oscillator parameter $`b`$ fixing the basis, and of the potential energy parameters. Results emerging from such an analysis can then be checked by studying the solutions $`c_n^{(0)}`$ and $`\mathrm{tan}(\delta )`$ as a function of the same parameters.
### 4.2 A Model Analysis using the Gaussian Potential
To analyze the behaviour of the AM set of linear equations and its defining quantities in a more or less general way, we consider a simple gaussian potential of the form
$$\widehat{V}(r)=V_0\mathrm{exp}((\frac{r}{a})^2)$$
(39)
There are a number of reasons to justify such a choice: (1) an operator with gaussian functional form is easily handled in a harmonic oscillator basis, as matrix elements can be calculated using closed expressions, or simple recurrence formulae; (2) (semi-) realistic potentials are often expressed explicitly as a finite sum of gaussians, each with specific amplitude and width parameters, and are of common use e.g. in nuclear physics calculations; (3) a very large class of potentials can be expressed in terms of a gaussian transform, such as e.g. a Yukawa or a Coulomb potential.
In general the matrix elements of the potential energy operator, due to the $`r/a`$ dependence of the latter, will depend on the ratio $`b/a`$. In the specific case of a gaussian potential one has an explicit dependence on
$$\gamma =(\frac{b}{a})^2$$
(40)
This means that matrix elements of the gaussian potential energy operator in an oscillator basis are invariant under all scale transformations of $`a`$ and $`b`$ which preserve $`\gamma `$, and different physical situations will lead to the same matrix $`<n\widehat{V}m>`$. In particular, small values of $`\gamma `$ are realised by small values of $`b`$ relative to $`a`$, or large values of $`a`$ relative to $`b`$; this situation corresponds to a “long range” potential. Large values of $`\gamma `$ correspond likewise to a “short range” potential.
The potential is certainly not the only parametrized quantity characterizing the set of AM equations. The oscillator radius of the oscillator basis is another and equally important parameter, because of the repercussion on the values of both the kinetic and potential matrix elements appearing in the equations. In this section we will therefore study the behaviour of the AM equations as a function of both the oscillator radius of the basis, and the width of the gaussian potential. Actually, as indicated by (40) only the ratio of these quantities is required, without any loss of generality for the current analysis. We therefore make a specific choice of potential parameters, namely $`V_0=8MeV`$ and $`a=1fm`$, so that the potential has both a discrete and continuous spectrum. The $`\gamma `$ ratio will then be varied by adapting the oscillator radius $`b`$.
A central theme in the analysis is the characterization for a rapidly converging, and stabilized, solution of the AM equations. This is most naturally done by searching for a maximal, limiting, value for the number of basis states involved in a specific solution, depending on the problem parameters used. As the asymptotic behaviour of the solution is governed essentially by the kinetic energy term, one should therefore take into account how both potential and kinetic matrix elements behave with respect to one another. In other words, it is not sufficient to know about the insignificance of potential matrix elements in absolute terms to decide whether one has reached the asymptotic region, but rather consider some relative insignificance with respect to the value of the kinetic matrix elements.
When working in a coordinate representation, one can obtain a lot of information concerning the wave-function by analyzing the Hamiltonian only, and this both in terms of coordinate and energy ranges. The main reason for this is that the Hamiltonian is globally defined in the whole coordinate space. In a $`L^2`$ representation, in our case using an oscillator basis, this is not such a straightforward matter. The main parameters determining the rate of convergence, as well as the behaviour of the solution (the wave-function in an oscillator representation), are of a “global” nature, such as $`_m<n\widehat{V}m>c_m=<n\widehat{V}\mathrm{\Psi }>`$, $`<n\widehat{V}\mathrm{\Psi }^{(+)}>`$ and so on. The matrix elements of the kinetic, respectively potential energy operators are “local” quantities (cfr. the value of these operators in a single point in coordinate space). The study of these elements however reveals the peculiarities of the “global” quantities, and will help to understand their behaviour.
Concerning the kinetic energy term, two remarks are very important in the context of an oscillator basis. The kinetic energy contribution has a very simple representation in terms of the oscillator basis: (1) it is proportional to the inverse square of the oscillator radius $`b`$, and (2) the kinetic energy matrix has a tri-diagonal form, i.e. has non-zero matrix elements only along the major diagonal and the first subdiagonals.
#### 4.2.1 Analysis of the Hamiltonian matrix elements
Fig. 1 displays the overall (qualitative) behaviour of a typical gaussian potential matrix in some specific oscillator basis. The main properties to be noted are the comportment of (1) the main diagonal which falls off monotonically for large $`n`$, and of (2) the non-diagonal matrix elements with non-zero values concentrated symmetrically around the main diagonal. This structure of the potential contribution is certainly of a promising nature, regarding the earlier remarks concerning the kinetic energy contribution.
A more quantitative view of the diagonal behaviour of the potential matrix is displayed in Fig. 2, and this for a number of choices of oscillator bases. Three different views are presented in this figure: (a) the pure diagonal matrix elements; (b) the diagonal matrix elements multiplied by the square of the oscillator radius (eliminating the kinetic energy dependence on the choice of basis), and (c) the diagonal matrix elements normalized with respect to the first one; both figures 2 (a) and 2 (b) lead to Fig. 2 (c) through this normalization. From these figures it should be clear that one has to be careful when drawing conclusions for a proper choice of basis. Fig. 2 (a) might suggest that very large values of the oscillator radius $`b`$ are optimal. Fig. 2 (b) on the contrary might suggest that very small values of $`b`$ are optimal. The normalized Fig. 2 (c) finally suggests that, for the current potential parameters, the interval for $`b`$ between $`0.25fm`$ and $`2.0fm`$ would be optimal. Fig. 2 at least reveals that there are important differences in behaviour of the potential energy contribution for different choices of the oscillator radius. Later on it will be shown in a more quantitative way that the qualitative conclusion for an optimal $`b`$ interval as suggested by Fig. 2 (c) is correct. An interesting conclusion that is apparent from the figure is that a (coordinate space defined) short-range potential turns out to have a long-range character in an oscillator representation. As the oscillator basis is used as a Fourier basis to portray the solution, this is a well-known effect in terms of Fourier representation theory.
A quantitative view of the non-diagonal behaviour of the potential matrix for the same bases displayed in Fig. 2 can be found in Fig. 3, and this by showing a typical row for a fixed column index ($`n=50`$ in this case). The same 3 views (pure, multiplied by $`b^2`$, and normalized with respect to the diagonal matrix element with $`n=50`$) are shown respectively in figures 3 (a), (b) and (c). Again a strong dependence on the choice of the oscillator radius is remarked, but it is much more difficult to draw conclusions for an optimally converging basis from these figures. As this figure indicates that the potential energy contribution is concentrated around the main diagonal of the matrix, but with a relatively important distribution, it should already warn against carelessly applying the simple but straightforward solution scheme presented in section 2 of this paper!
The previous figures thus only allow us to initiate some general discussion, but not to draw final and well-formed conclusions on the convergence problem, let alone on the determination of an optimal basis and the number of states involved in a stable solution.
#### 4.2.2 Analysis of the Dynamical Coefficients
Up to now only local views of the potential contribution, i.e. matrix elements and their behaviour, have been considered to provide possible hints for a properly converging solution. The new representation of the AM equations presented earlier in this section introduced some new, global quantities which might be of interest in the convergence analysis. These are the Dynamical Coefficients $`V_n^{(+)}`$ and $`V_n^{()}`$, and they combine all matrix elements of row $`n`$ as indicated above. If, and when, these quantities would be (sufficiently) zero from a given value $`N`$ on, this would certainly determine the maximal number of oscillator states to be considered for a converging solution. As $`V_n^{(+)}`$ and $`V_n^{()}`$ depend on both the potential and the basis parameters, they will be analyzed as a function of the ratio of potential width to oscillator radius. As $`V_n^{(+)}`$ and $`V_n^{()}`$ also depend explicitly on energy, this will be an additional parameter to consider.
A qualitative view of $`V_n^{(+)}`$ is shown in Fig. 4 (a) as a function of both the index $`n`$ and the ratio $`b/a`$; the quantity was normalized with respect to $`V_0^{(+)}`$, and the scale is a logarithmic one. From this figure one immediately notices that a well-defined value of the ratio of the potential to oscillator radius zeroes the quantities for already very small values of $`n`$. Fig. 4 (b) depicts the same view for $`V_n^{()}`$, carrying an analogous conclusion. Although both optimal values do not coincide completely, they are sufficiently close to define a narrow interval of optimal values for a swiftly converging solution. Indeed, if we consider a value $`N`$ for which $`V_N^{(+)}/V_0^{(+)}<ϵ`$ and $`V_N^{()}/V_0^{()}<ϵ`$ (with e.g. $`ϵ<10^6`$), we can assume to have reached the asymptotic region. The value $`N`$ then determines the number of basis functions needed to obtain a well-converged solution. Indeed, the solutions of (32) are deviations with respect to the (known) asymptotic solutions, and become zero when both $`V_n^{(+)}`$ and $`V_n^{()}`$ are (sufficiently) zero. In other words, the value $`N`$ from which on $`V_n^{(+)}`$ and $`V_n^{()}`$, and thus the solutions $`c_n^{(0)}`$, are zero determines what would be called in RGM terminology the matching point between the internal and the asymptotic region. It is important to remark that $`N`$ is determined prior to solving the AM equations, and is obtained from the simple knowledge of the potential energy matrix elements.
To corroborate the fact for an optimal choice of basis for a given gaussian potential, a quantitative view of $`V_n^{(+)}`$ and $`V_n^{()}`$ as a function of the radius parameter $`b`$ is shown in figures 5 (a) and (b) for a selected number of $`n`$ values. From this figure one can determine for which values of the radius parameter one obtains a properly converged solution, by considering the intersection of both optimal intervals for $`V_n^{(+)}`$ and $`V_n^{()}`$. One notices that for a potential consisting of a single gaussian term, a very limited number of basis states is necessary for a proper convergence. For more intricate potential forms this will not necessarily be the case, although an optimal value for the radius parameter, be it associated with a relatively larger value of $`N`$, will still be available.
To check the dependence of the so-called “optimal” $`b`$ values (or regions) as a function of energy, Fig. 6 displays the values of both $`V_n^{(+)}`$ and $`V_n^{()}`$, relative to respectively $`V_0^{(+)}`$ and $`V_0^{()}`$ for varying $`b`$ and energy for a fixed $`n`$. It is important to notice that, at least for the gaussian potential, the optimal value for the oscillator radius is independent of energy.
Finally Fig. 7 shows the quantitative behaviour of $`V_n^{(+)}`$ and $`V_n^{()}`$, again relative to respectively $`V_0^{(+)}`$ and $`V_0^{()}`$, as a function of $`n`$. From these figures one easily obtains an optimal number of basis functions, given the specific parameters of the problem and the precision considered. As a hands on example, Fig. 7 indicates that, for a precision of about $`10^8`$, less than 12 basis functions are needed for all $`b`$ values between $`0.5fm`$ and $`1fm`$.
The conclusions above indicate the importance of the $`V_n^{(+)}`$ and $`V_n^{()}`$ quantities, and a closer investigation imposes itself. A closed expression for $`V_n^{(+)}`$ can be obtained in a straightforward way as:
$`V_n^{(+)}=V_0{\displaystyle \frac{(12\gamma )^n}{(1+2\gamma )^{n+L+3/2}}}k^L\mathrm{exp}({\displaystyle \frac{1}{2}}k^2{\displaystyle \frac{1}{1+2\gamma }})N_{nL}L_n^{L+1/2}({\displaystyle \frac{k^2}{14\gamma ^2}})`$ (41)
for which the full calculation is reproduced in appendix A.
This expression features a single minimum in terms of $`\gamma `$, namely for $`\gamma =1/2`$ or $`b=a/\sqrt{2}`$, with a value
$$V_n^{(+)}=V_0N_{nL}\frac{1}{4}\frac{1}{n!}(\frac{k}{2})^{2n+L+1/2}\mathrm{exp}(\frac{1}{4}k^2)$$
(42)
which drops off to zero very fast in terms of $`n`$ because of its following behaviour:
$$V_n^{(+)}\frac{1}{n!R_n^{L+1/2}}$$
(43)
Both position and behaviour of $`V_n^{(+)}`$ around this minimum were already apparent from the foregoing figures.
The asymptotic behaviour of $`V_n^{(+)}`$ for both small and large values of $`\gamma `$ is also readily obtained. For small values of $`\gamma `$ (i.e. small values of the oscillator radius $`b`$ compared to the potential width $`a`$) equation (41) yields the following asymptotic form, valid for large $`n`$:
$`V_n^{(+)}`$ $``$ $`V_0\mathrm{exp}\{\gamma R_n^2\}c_n^{(+)}`$ (44)
$``$ $`V_0\mathrm{exp}\{\gamma R_n^2\}\sqrt{2R_n}j_L(kbR_n)`$
This expression consists of two factors in coordinate representation, the gaussian potential $`\widehat{V}(r)`$ and the (asymptotic) wave function $`\mathrm{\Psi }^{(+)}(k,r)`$, both evaluated in one and the same discrete point $`r=bR_n`$ ($`R_n`$ is the classical turning point, cfr. (3) and (4)). For $`\gamma `$ approaching zero the expression shows a slowly decreasing behaviour of $`V_n^{(+)}`$ as a function of $`n`$.
For large values of $`\gamma `$ one obtains the following asymptotic form valid for large $`n`$
$`V_n^{(+)}`$ $`(1)^n`$ $`V_0{\displaystyle \frac{1}{2\gamma }}\sqrt{{\displaystyle \frac{2}{k}}}\mathrm{exp}\left\{{\displaystyle \frac{R_n^2+k^2}{4\gamma }}\right\}I_{L+1/2}({\displaystyle \frac{k}{2\gamma }}R_n)`$ (45)
which for the limiting case of $`\gamma `$ now approaching infinity again displays a slowly decreasing behaviour as a function of $`n`$.
Both very large and very small values of $`\gamma `$ will thus lead to rather badly converging solutions, which essentially means that a large number of basis functions will have to be considered when $`\gamma `$ reaches towards limiting values.
This discussion on asymptotic behaviour confirms conclusions already apparent from the numerically calculated figures of $`V_n^{(+)}`$.
Fig. 8 indicates that the asymptotic formulae are quite valid for a broad range of $`b`$ values, and as such, can often be used to obtain reasonable values for $`N`$ for properly converged solutions. Indeed, as the main factor which defines the decreasing behaviour of $`V_n^{(+)}`$ for small, respectively large values of $`\gamma `$, is the potential term, which has the value $`\mathrm{exp}\{\gamma R_n^2\}`$, respectively $`\mathrm{exp}\{\frac{R_n^2}{4\gamma }\}`$, as can be seen in equations (44) and (45). So, if one considers e.g. $`10^6`$ to be a reasonable measure for precision, one can write approximately
$`\left|{\displaystyle \frac{V_n^{(+)}}{V_0^{(+)}}}\right|`$ $``$ $`\mathrm{exp}\{\gamma R_n^2\}=10^6\text{(small}\gamma )`$
$`\left|{\displaystyle \frac{V_n^{(+)}}{V_0^{(+)}}}\right|`$ $``$ $`\mathrm{exp}\{{\displaystyle \frac{R_n^2}{4\gamma }}\}=10^6\text{(large}\gamma )`$ (46)
which immediately leads to the respective values $`N14/(4\gamma )`$ (for small $`\gamma `$) and $`N14\gamma )`$ (for large $`\gamma `$); for a value of $`b/a=0.2`$ one obtains $`N90`$, and for $`b/a=3.0`$ $`N125`$.
The asymptotic behaviour of $`V_N^{(+)}`$ can be used to an even better extent, by taking it into account when solving the AM system of equations. This will be pursued in full detail in a forthcoming section.
Although more intricate to develop, an analogous analysis can be made for $`V_n^{()}`$, leading to corresponding conclusions.
#### 4.2.3 Analysis of the phase shifts
In Fig. 9 we show the phase shifts obtained at an energy of $`E=1MeV`$ for different values of $`\gamma `$ (actually in all further results $`a=1.0fm`$, so that $`b`$ and $`\gamma `$ coincide), as a function of the number of basis states involved in the calculation. One notices an important gain in convergence speed (and thus precision) when using the reformulated version of the AM equations (36), although small and large $`b`$ (or $`\gamma `$) values still require an important number of basis states.
On the same figure the normalized $`c_n^{(0)}`$ solutions are shown for different values of $`b`$ ($`\gamma `$), for a calculation involving 100 basis states. These results are normalized with respect to the norm $`W`$
$$W=\frac{_{n=0}^{99}c_n^{(0)}^2}{_{n=0}^{99}c_n^2}$$
(47)
which is also include in the figures. One notices from this value that in the “optimal $`b`$ ($`\gamma `$)” region very few, very small $`c_n^{(0)}`$ contribute to the solution.
The results displayed on Fig. 9 confirm our suggestion made by analysing the behaviour of $`V^{(+)}`$ and $`V^{()}`$, i.e. that for optimal values of $`b`$ ($`\gamma `$) convergence is achieved with less than 12 basis functions. They also confirm that for $`b=0.2`$ one can use approximately 75, and for $`b=3.0`$ approximately 100 basis states for reasonable convergence; the latter were indeed overestimated by the numbers obtained from equation (46).
To corroborate the fact that the reformulated AM equations (36) provide faster converging solutions, we show in Fig. 10 both the $`c_n`$ and $`c_n^{(0)}`$ coefficients obtained in a calculation at $`E=1MeV`$ and $`b=3.0`$, with a total number of 100 basis states. This picture confirms the fact that the true solution indeed deviates only by a small amount from the asymptotic one.
To show the dependence of the results on energy, we show in Fig. 11 the phase shift obtained with 5, 10 and 15 basis states compared to the exact results, and this with the original (“simple”) version (20) and the reformulated version (36) of the AM equations. For $`b`$ values deviating reasonably from the optimal value, the latter form of the equations is seen to be highly superior to the original one. For very small and very large $`b`$ values, the convergence is still problematic.
## 5 Stable Solutions for the AM Equations
In the previous section we obtained interesting results concerning the convergence behaviour of the solutions of the AM equations in terms of the parameters defining the basis and the potential energy. It was shown for a gaussian potential that the rate of convergence was reliably predictable.
The parameter $`\gamma `$ or, for fixed potential width $`a`$, the oscillator radius $`b`$ determines the rate of convergence of the expansion $`\mathrm{\Psi }=_nc_nn>`$. It was indeed seen that an optimal value (or a restricted range of optimal values) for $`b`$ leads to very fast convergence. However, values deviating more or less strongly from the optimal $`b`$ lead to slowly converging and numerically unprecise results.
As it is not always possible in a realistic calculation to choose the optimal $`b`$, for physical as well as for numerical reasons, it is important to develop strategies for stable results even in non-optimal $`b`$-regions. This problem was already recognized in , were a coupled channels calculated for $`{}_{}{}^{4}He`$ was performed with the AM, in which both cluster and collective configurations were taken into account: the $`b`$ value was fixed by physical arguments (essentially optimizing the cluster channel results), and very non-optimal for the collective channels (in particular the Monopole Mode).
The following subsections indicate how such problems can be solved by considering novel strategies to solve the AM equations, without modifying the original physical problem (e.g. by tampering with $`b`$).
### 5.1 A General Analysis of the Asymptotics
We attempt to evaluate the asymptotic behaviour (i.e. for $`n1`$) for the expansion coefficients $`c_n=<n\mathrm{\Psi }>`$ and the matrix elements $`<n\widehat{V}\mathrm{\Psi }>`$, using only very general information on the (unknown) wave function $`\mathrm{\Psi }`$. Based on this formulation we will then introduce strategies to overcome the slow-convergence situations. All assumptions made, and results obtained, will be checked against those produced with a gaussian potential.
To derive a general view on the asymptotic behaviour of the Fourier coefficients $`c_n`$, we use a Generator Coordinate (GC) representation for the wave function $`\mathrm{\Psi }`$:
$`\mathrm{\Psi }(r)={\displaystyle _0^{\mathrm{}}}𝑑\beta g(\beta )r^L\mathrm{exp}\{\beta ^2r^2\}`$ (48)
using a scaled gaussian $`<r\beta >=r^L\mathrm{exp}\{\beta ^2r^2\}`$ as a kernel of the integral transformation. It is well known that this kernel provides an alternative, continuous, basis to the traditional, discrete, oscillator basis. A tacit assumption in (48) is that the continuous spectrum wave functions are representable by such a representation, (in particular, that a “reasonable” form for $`g(\beta )`$ exists, leading to an integrable result).
#### 5.1.1 Asymptotics of $`c_n`$
The gaussian kernel of (48) is easily expanded in terms of the oscillator basis used throughout this paper. The basis functions $`n>`$ are explicited in a coordinate representation by (35).
By substituting $`\rho b`$ for $`r`$ in the integral kernel $`r^L\mathrm{exp}\{\beta ^2r^2`$}, one can recognizes the generating function for the oscillator functions:
$`(1+\epsilon )^{L3/2}\rho ^L\mathrm{exp}\{{\displaystyle \frac{1}{2}}{\displaystyle \frac{1\epsilon }{1+\epsilon }}\rho ^2\}={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\epsilon ^n}{N_n}}n>`$ (49)
where
$`\beta ^2b^2={\displaystyle \frac{1}{2}}{\displaystyle \frac{1\epsilon }{1+\epsilon }}\text{or}\epsilon ={\displaystyle \frac{12\beta ^2b^2}{1+2\beta ^2b^2}}`$ (50)
The expansion coefficients $`<n\beta >`$ of $`\mathrm{exp}\{\beta ^2r^2\}`$ in the oscillator basis are then
$`<n\beta >`$ $`=`$ $`(1+\epsilon )^{L+3/2}{\displaystyle \frac{\epsilon ^n}{N_n}}b^L`$ (51)
$`=`$ $`{\displaystyle \frac{(12\beta ^2b^2)^n}{(1+2\beta ^2b^2)^{n+L+3/2}}}{\displaystyle \frac{2^{L+3/2}}{N_n}}b^L`$
from which one obtains the expansion coefficients $`c_n=<n\mathrm{\Psi }>`$ of $`\mathrm{\Psi }`$ in the oscillator basis as
$`<n\mathrm{\Psi }>={\displaystyle _0^{\mathrm{}}}𝑑\beta g(\beta )<n\beta >`$
$`={\displaystyle _0^{\mathrm{}}}𝑑\beta g(\beta ){\displaystyle \frac{(12\beta ^2b^2)^n}{(1+2\beta ^2b^2)^{n+L+3/2}}}{\displaystyle \frac{2^{L+3/2}}{N_n}}`$ (52)
If, and when, $`g(\beta )`$ is concentrated in a small vicinity of $`\beta \beta _0`$, one can expect a highly convergent expansion for an oscillator length $`b1/\sqrt{2}\beta _0`$.
To study the asymptotic behaviour of the expansion coefficients (52), we consider two limiting regions for $`b`$, i.e. “small $`b`$” and “large $`b`$”. To this end, we rewrite (52) as follows
$`<n\mathrm{\Psi }>={\displaystyle _0^{\beta _0}}𝑑\beta g(\beta )<n\beta >+{\displaystyle _{\beta _0}^{\mathrm{}}}𝑑\beta g(\beta )<n\beta >`$ (53)
where $`\beta _0=1/(\sqrt{2}b)`$.
##### Small $`b`$ values:
for $`0\beta <1/(\sqrt{2}b)`$ and large values of $`n`$, one can use the following approximate formula
$`{\displaystyle \frac{(12\beta ^2b^2)^n}{(1+2\beta ^2b^2)^{n+L+3/2}}}\mathrm{exp}\{R_n^2\beta ^2b^2\}`$ (54)
For large values of $`n`$ ($`n1`$) one also has
$`{\displaystyle \frac{1}{N_n}}=\sqrt{{\displaystyle \frac{\mathrm{\Gamma }(n+L+3/2)}{2\mathrm{\Gamma }(n+1)}}}R_n^{L+1/2}/2^{L+1}`$ (55)
where again $`R_n`$ is the classical turning point (cfr. eqs (3) and (4)).
For very small values of $`b`$, one can consider only the first term in (53); using the approximations (54) and (55) one then obtains the following asymptotic form of (52)
$`<n\mathrm{\Psi }>\sqrt{2}\sqrt{R_n}{\displaystyle _0^{\mathrm{}}}𝑑\beta g(\beta )(R_nb)^L\mathrm{exp}\left\{R_n^2\beta ^2b^2\right\}`$ (56)
which, considering (48), leads to the final approximation:
$`c_n=<n\mathrm{\Psi }>\sqrt{2}\sqrt{R_n}\mathrm{\Psi }(bR_n)`$ (57)
For small values of the oscillator length $`b`$, the expansion coefficients of the wave function $`\mathrm{\Psi }(r)`$ on an oscillator basis are thus proportional to the wave function itself taken at the discrete argument values $`bR_n`$.
Such a relationship between the wave function in coordinate space and its oscillator representation was already obtained long ago , for the regular and irregular asymptotic solutions. Later, this correspondence has been used as a heuristic principle for solving e.g. the Coulomb problem in an oscillator representation .
##### Large $`b`$ values:
for $`1/(\sqrt{2}b)\beta <\mathrm{}`$ and large values of $`n`$, one can use the following approximate formula
$`{\displaystyle \frac{(12\beta ^2b^2)^n}{(1+2\beta ^2b^2)^{n+L+3/2}}}`$ $`=`$ $`(1)^n{\displaystyle \frac{1}{(2\beta ^2b^2)^{L+3/2}}}{\displaystyle \frac{(11/2\beta ^2b^2)^n}{(1+1/2\beta ^2b^2)^{n+L+3/2}}}`$ (58)
$``$ $`(1)^n{\displaystyle \frac{1}{(2\beta ^2b^2)^{L+3/2}}}\mathrm{exp}\{{\displaystyle \frac{R_n^2}{4\beta ^2b^2}}\}`$
The expansion coefficients can then be approximated by the second term in (53), which for very large values of $`b`$ leads to the following asymptotic form of (52)
$`<n\mathrm{\Psi }>(1)^n\sqrt{2}R_n^{L+1/2}b^L{\displaystyle _0^{\mathrm{}}}𝑑\beta g(\beta ){\displaystyle \frac{\mathrm{exp}\{R_n^2/4\beta ^2b^2\}}{(2\beta ^2b^2)^{3/2}}}`$ (59)
Using an integral transformation \[12, 11.4.29\] the asymptotic form of the $`c_n`$ for large $`b`$ becomes
$`c_n=<n\mathrm{\Psi }>`$ $``$ $`(1)^n{\displaystyle \frac{2}{\sqrt{\pi }}}\sqrt{R_n}{\displaystyle _0^{\mathrm{}}}𝑑rr^2j_L\left({\displaystyle \frac{R_n}{b}}r\right)\mathrm{\Psi }(r)`$ (60)
$`=`$ $`(1)^n{\displaystyle \frac{2}{\sqrt{\pi }}}\sqrt{R_n}\mathrm{\Phi }({\displaystyle \frac{R_n}{b}})`$ (61)
where $`\mathrm{\Phi }`$ is the wave function in momentum representation.
For large values of the oscillator length $`b`$, the expansion coefficients of the wave function $`\mathrm{\Psi }(r)`$ on an oscillator basis are thus proportional to the wave function in momentum space taken at discrete arguments which are the momentum values $`R_n/b`$. As oscillator functions in coordinate and momentum space only differ by a phase $`(1)^n`$ and the argument $`\rho `$ (which in coordinate space equals $`r/b`$ and in momentum space equals $`kb`$), both wave functions $`\mathrm{\Psi }(r)`$ in coordinate representation and $`\mathrm{\Phi }(p)`$ in momentum representation have essentially the same expansion coefficients.
By integrating (60) by parts, assuming that $`\mathrm{\Psi }(r)`$ is smooth and without singular points, and taking the leading asymptotic term for $`c_n`$ (i.e. for $`R_n1`$), one obtains
$`<n\mathrm{\Psi }>(1)^n{\displaystyle \frac{1}{R_n^{3/2}}}\mathrm{\Psi }(0)`$ (62)
The same result can be obtained by considering that the wave functions $`\mathrm{\Phi }(p)`$ of discrete and continuous spectrum states in momemtum space decrease at least as $`1/p^2`$. Substitution of this limiting factor in (61) again leads to (62)
Summing up one notes that the asymptotics of $`c_n`$ are related to the wave function behaviour in coordinate space: for small values of $`b`$ the asymptotics are proportional to the wave function for large $`r`$, and for large values of $`b`$ the asymptotics are proportional to the wave function in the vicinity of the origin.
#### 5.1.2 Asymptotics of $`<n\widehat{V}\mathrm{\Psi }>`$
Asymptotic expressions for the matrix elements $`<n\widehat{V}\mathrm{\Psi }>`$ can be obtained in an analogous way as for $`c_n=<n\mathrm{\Psi }>`$ by using the GC representation of the solution (48). We omit the details of the calculation, and only present the final results.
##### Small $`b`$ values:
for small $`b`$ values (or better, small $`\gamma `$ values), the matrix elements factorize as follows:
$`<n\widehat{V}\mathrm{\Psi }>`$ $``$ $`\widehat{V}(bR_n)<n\mathrm{\Psi }>`$ (63)
$``$ $`\widehat{V}(bR_n)\sqrt{2}\sqrt{R_n}\mathrm{\Psi }(bR_n)`$
One notices that the behaviour of the matrix elements coincides with the one of the $`\mathrm{\Psi }(r)`$, up to a trivial factor which is potential dependent.
These results obtained for a general form of a short-range potential, are confirmed by the asymptotic form for $`V_n^{(+)}`$, evaluated earlier (44) for a gaussian potential.
##### Large $`b`$ values:
for large values of $`b`$ (or $`\gamma `$), the asymptotic form of the matrix elements $`<n\widehat{V}\mathrm{\Psi }>`$ reduces to the integral
$`<n\widehat{V}\mathrm{\Psi }>(1)^n{\displaystyle \frac{2}{\sqrt{\pi }}}\sqrt{R_n}{\displaystyle _0^{\mathrm{}}}𝑑rr^2j_L({\displaystyle \frac{R_n}{b}}r)\widehat{V}(r)\mathrm{\Psi }(r)`$ (64)
or, in other words, to the convolution of the wave function with the potential in momentum representation
$`<n\widehat{V}\mathrm{\Psi }>(1)^n{\displaystyle \frac{2}{\sqrt{\pi }}}\sqrt{R_n}{\displaystyle _0^{\mathrm{}}}𝑑\stackrel{~}{k}\stackrel{~}{k}^2\widehat{v}(k_n\stackrel{~}{k})\mathrm{\Phi }(\stackrel{~}{k})`$ (65)
where $`\widehat{v}(k)`$ and $`\mathrm{\Phi }_L(k)`$ are the Fourier transforms of the potential $`\widehat{V}(r)`$ and the wave function $`\mathrm{\Psi }(r)`$ respectively.
These results for a general potential can again be checked by considering the corresponding form for $`V_n^{(+)}`$ obtained earlier with a gaussian potential. Indeed, by calculating the integral
$`V_0{\displaystyle _0^{\mathrm{}}}𝑑rr^2j_L({\displaystyle \frac{R_n}{b}}r)\mathrm{exp}\left\{\gamma r^2\right\}j_L(kr)`$ (66)
where the unknown solution $`\mathrm{\Psi }(r)`$ is substituted by $`\mathrm{\Psi }^{(+)}(k,r)=\sqrt{2/\pi }j_L(kr)`$ (the “free particle” solution), one obtains (45), which is valid for both large $`b`$ and $`n`$.
The matrix elements $`<n\widehat{V}\mathrm{\Psi }>`$ can be evaluated in the asymptotic region by making the same assumptions concerning the behaviour of $`\mathrm{\Psi }(r)`$ as in the previous section to obtain the asymptotic form of $`c_n=<n\mathrm{\Psi }>`$. Expanding $`\mathrm{\Psi }(r)`$ in a power series of $`r`$, keeping the first term $`\mathrm{\Psi }(0)r^L`$ and integrating term by term in (64) leads to the following asymptotic form:
$$<n\widehat{V}\mathrm{\Psi }>(1)^n\frac{2}{\sqrt{\pi }}\sqrt{R_n}\widehat{v}(\frac{R_n}{b})\mathrm{\Psi }(0)$$
(67)
where the Fourier transform $`\widehat{v}(R_n/b)`$ of the potential determines how fast the matrix elements decrease in terms of $`n`$.
### 5.2 New Strategies for Solving the AM Equations
Based on the results obtained above, we can now suggest new strategies for solving the dynamical equations of the AM, in those cases where the oscillator length $`b`$, or more exactly the ratio $`\gamma =(b/a)^2`$, is relatively large or small compared to the optimal value.
##### Small $`b`$ values:
we start from the Schrödinger equation in (1)
$`{\displaystyle \underset{m}{}}<n\widehat{T}Em>c_m+<n\widehat{V}\mathrm{\Psi }>=0`$ (68)
$`{\displaystyle \underset{\mu }{}}<\nu \widehat{T}E\mu >c_\mu +<\nu \widehat{V}\mathrm{\Psi }>=0`$ (69)
where the index $`n`$ is connected with the internal region and $`\nu `$ with the asymptotic region, and $`N`$ delimits the internal region. In the asymptotic region, we can use the asymptotic form (63) for the matrix elements
$$<\nu \widehat{V}\mathrm{\Psi }>=\widehat{V}(bR_\nu )c_\nu $$
(70)
which yields
$`{\displaystyle \underset{m}{}}<n\widehat{T}Em>c_m+<n\widehat{V}\mathrm{\Psi }>=0`$ (71)
$`{\displaystyle \underset{\mu }{}}<\nu \widehat{T}+\widehat{V}(bR_\nu )E\mu >c_\mu =0`$ (72)
In the original version of the AM the asymptotic coefficients $`c_\nu `$ were given by (6), and obtained from a simple three term recurrence relation, due to the very selective coupling induced by the kinetic energy operator between oscillator states. Instead of the original form, and based on (72), we propose a modified three term recurrence form, including the asymptotic behaviour of the potential matrix elements in the diagonal term
$`{\displaystyle \underset{\mu }{}}<\nu \widehat{T}+\widehat{V}(bR_\nu )E\mu >c_\mu =0`$ (73)
or
$`[<\nu \widehat{T}\nu >+\widehat{V}(bR_\nu )E]c_\nu `$
$`+<\nu \widehat{T}\nu 1>c_{\nu 1}+<\nu \widehat{T}\nu +1>c_{\nu +1}=0`$ (74)
In the asymptotic region $`c_\nu =c_\nu ^{(+)}+\mathrm{tan}\delta c_\nu ^{()}`$, hence the system (74) should be solved independently for regular ($`c_\nu ^{(+)}`$) and irregular ($`c_\nu ^{()}`$) coefficients, with ”boundary” conditions at starting remote points $`\nu =N_a,N_a+1`$
$`c_{\nu _0}^{(+)}=\sqrt{2R_\nu }j_L(kR_\nu )`$ (75)
$`c_{\nu _0}^{()}=\sqrt{2R_\nu }n_L(kR_\nu )`$ (76)
These modified asymptotic coefficients $`c_\nu ^{(+)}`$ and $`c_\nu ^{()}`$ should be considered when calculating the correspondingly modified $`V_n^{(+)}`$ and $`V_n^{()}`$ in the dynamical equations (36). The modified equations should then be solved for the phase shift $`\delta `$ and the coefficients of the internal region $`c_n^{(0)}`$.
##### Large $`b`$ values:
for large values of $`b`$, we consider the dynamical equations in the form (32), containing the $`V_n^{(+)}`$ and $`V_n^{()}`$ terms
$`{\displaystyle \underset{m=0}{}}<n\widehat{H}Em>c_m^{(0)}+V_n^{()}\mathrm{tan}(\delta )=V_n^{(+)}`$ (77)
for the internal region ($`nN`$), and
$$\underset{\mu }{}<\nu \widehat{T}E\mu >c_\mu +<\nu \widehat{V}\mathrm{\Psi }>=0$$
(78)
for $`\nu >N`$.
Substituting $`c_\mu ^{(0)}+c_\mu ^{(+)}+\mathrm{tan}(\delta )c_\mu ^{()}`$ for $`c_\mu `$ in (78), an assuming that $`c_\mu ^{(+)}`$ and $`c_\mu ^{()}`$ satisfy the three-term recurrence relation (31), one obtains
$$\underset{\mu }{}<\nu \widehat{T}E\mu >c_\mu ^{(0)}+<\nu \widehat{V}\mathrm{\Psi }>=0$$
(79)
Taking the asymptotic form (67) into account, one obtains the following equation for $`c_\mu ^{(a)}`$
$$\underset{\mu }{}<\nu \widehat{T}E\mu >c_\mu ^{(a)}+(1)^n\sqrt{R_\nu }v(\frac{R_\nu }{b})=0$$
(80)
where $`c_\nu ^{(0)}`$ was substituted by
$$c_\nu ^{(0)}=c_\nu ^{(r)}+Wc_\nu ^{(a)}$$
(81)
introducing the ”asymptotic” coefficients $`c_\nu ^{(a)}`$ and the ”residual” coefficients $`c_\nu ^{(r)}`$. All constants in the asymptotic form (67) were omitted and replaced by the global factor W to be determined. Equation (80) should now be solved subject to the boundary condition
$$c_\nu ^{(a)}=0$$
(82)
for $`\nu =N_a,N_a+1`$. Having obtained $`c_\nu ^{(a)}`$, the equations for the internal region should be modified to take the substitution (81) into account
$`{\displaystyle \underset{m=0}{}}<n\widehat{H}Em>c_m^{(r)}+\left[V_n^{(a)}Ec_n^{(a)}\right]W+V_n^{()}\mathrm{tan}(\delta )=V_n^{(+)}`$ (83)
were $`V_n^{(a)}`$ now stands for the sum
$$V_n^{(a)}=\underset{m=0}{}<n\widehat{H}m>c_m^{(a)}$$
(84)
The meaning of the asymptotic coefficients $`c_n^{(a)}`$ can be most easily understood by considering the major part of the asymptotic behaviour already in an intermediate $`n`$ region. Indeed, for large $`b`$ values, there is a very slow convergence of the results. This implies that the internal region extends towards very large $`n`$, with corresponding $`c_n^{(0)}`$ solutions probably very close (i.e. to first order) to the $`c_n^{(a)}`$.
##### Numerical application:
To demonstrate how the new (“small $`b`$” and “large $`b`$”) strategies accelerate the convergence, we again consider a gaussian potential in the two limiting cases where (i) the oscillator radius is a factor of 5 less than the potential range, and (ii) the oscillator radius is a factor of 5 larger than the radius of the potential.
Fig. 12 compares the phase shifts, obtained at an energy of $`1MeV`$, obtained in the original, reformulated and asymptotic approaches.
In Fig. 13 we display the exact phase shifts and those obtained with the new strategies using 5, respectively 10, basis states. One notices that for the “small $`b`$” case, 5 basis functions yield almost exact results (within a precision less than $`0.01\%`$); by even considering only one basis function, the phase shift is obtained within a precision of $`1\%`$! In the original form the latter precision could only be reached by using more than 50 basis functions.
For the “large $`b`$ strategy” ($`b/a=5`$), the convergence speed is remarkably increased, although less spectacular than in the “small $`b`$” case. The number of states needed for an identical result in the original formulation (36) is about 5 times higher.
The two strategies suggested are thus seen to significantly improve the convergence of the results and to reduce the computational efforts to obtain a desired precision.
By modifying the form of the asymptotic AM equations through inclusion of dynamical features as proposed and realized above for both small and large values of $`b`$, one is naturally led to the introduction of an “intermediate region”. This region distinguishes itself (i) from the “internal region” where the solutions are governed by the potential, and (ii) from the “asymptotic region” where the kinetic energy dominates the equations. In the intermediate region, the solutions are easily (i.e. in a numerically simple way) obtained. There are then actually two variational parameters to consider, $`N`$ marking the border between the internal and intermediate region, and $`N_a`$ marking the transition from the intermediate to the true asymptotic region. The larger or smaller is $`\gamma `$, the larger is the size (i.e. the difference $`N_aN`$) of the intermediate region. The strategies proposed above for small and large values of $`\gamma `$ have shown that, by considering a sufficiently large intermediate region (e.g. $`N_a=500`$), significantly reduces the range of the internal region, and dramatically decreases the computational effort to obtain converged results.
The intermediate region is charactarised by a noticeable but non-dominant presence of dynamical (i.e. potential) effects. These effects can be well approximated by asymptotic forms of both the expansion coefficients $`c_n`$ and potential matrix elements $`<n\widehat{V}\mathrm{\Psi }>`$. For small $`\gamma `$ (long-range correlations), this leads to a redefinition of the equations of $`c_n^{(+)}`$ and $`c_n^{()}`$ by incorporation a potential term in the three term recurrence relation governing the true asymptotic behaviour. For large $`\gamma `$ (short-range correlations) one approximates part of the $`c_n^{(0)}`$ solution by its asymptotic behaviour $`c_n^{(a)}`$. The latter are solutions of an inhomogeneous set of linear equations containing the Fourier transform of the potential, and as such accumulate the potential effects in the intermediate region.
In the region of $`\gamma `$ near to the optimal value, where fast converging solutions are obtained, there is no need for an intermediate region, and thus for an approximation of the dynamical equations; the values of $`N`$ and $`N_a`$ coincide in this case.
## 6 Conclusions
By introducing the “Dynamical Coefficients” $`V_n^{(+)}`$ and $`V_n^{()}`$ we were able to suggest a new form of the dynamical equations of the Algebraic Version of the Resonating Group Method (AM). The dynamical coefficients allow one to determine oscillator basis parameters (oscillator length and number of basis functions) for obtaining otimally converging and stable solutions of these equations. It was shown that for an optimal parameter set 5 to 10 functions were sufficient to obtain results with a precision of more than $`99.9\%`$.
The new form of equations have also been shown to be a proper starting point for obtaining new solution strategies in parameter regions were slow convergence would occur. Indeed, in parameter regions were the oscillator length is much larger, respectively much smaller, than the optimal one asymptotic approximations could be formulated by analysing the AM equations. The latter can be easily implemented numerically, and have shown to dramatically improve the convergence of the solutions.
It was shown also that, if the oscillator radius is much smaller than the width of the potential, the expansion coefficients $`c_n`$ of the wave function on the oscillator basis coincide with the wave function in coordinate space up to a simple factor. In the other limiting case, when the oscillator radius is much larger than the width of the potential, the expansion coefficients $`c_n`$ were found to be proportional to the wave function in momentum space.
## 7 Acknowledgments
The authors are gratefull to Profs. G. F. Fillipov and P. Van Leuven for valuable discussions. They would also like to thank INTAS for financial support for part of this work under the grant contract INTAS 93/755. One of us (V. S. V.) gratefully acknowledges the kind hospitality of the members of the research group “Computational Qauntum Physics” of the Department of “Mathematics and Computer Sciences”, University of Antwerp, RUCA, Belgium, and the “Departement voor Wetenschappelijke, Technische en Culturele aangelegenheden, Deelregering Vlaanderen” for a scholarship.
## Appendix A appendix
In this appendix the explicit form, and governing recurrence relations, of the dynamical coefficients $`V_n^{(+)}`$ and $`V_n^{()}`$ are calculated for a gaussian potential.
The following integrals should be evaluated:
$`V_n^{(+)}`$ $`=`$ $`V_0{\displaystyle _0^{\mathrm{}}}𝑑rr^2\mathrm{\Phi }_n(r)\mathrm{exp}\left(r^2/a^2\right)\mathrm{\Psi }^{(+)}(kr)`$ (85)
$`V_n^{()}`$ $`=`$ $`V_0{\displaystyle _0^{\mathrm{}}}𝑑rr^2\mathrm{\Phi }_n(r)\mathrm{exp}\left(r^2/a^2\right)\mathrm{\Psi }^{()}(kr)`$ (86)
where $`\mathrm{\Psi }^{(+)}=\sqrt{\frac{2}{\pi }}j_L(kr)`$ and $`\mathrm{\Psi }^{()}`$ is a solution of the inhomogeneous differential equation:
$`(\widehat{T}E)\mathrm{\Psi }^{()}=\beta _0\mathrm{\Phi }_0(r)`$ (87)
($`\beta _0`$ is defined in eq. (26) ), which can be represented in the integral form
$`\mathrm{\Psi }^{()}`$ $`=`$ $`\beta _0{\displaystyle 𝑑\stackrel{~}{r}\stackrel{~}{r}^2G(r,\stackrel{~}{r})\mathrm{\Phi }_0(\stackrel{~}{r})}`$ (88)
$`=`$ $`\beta _0{\displaystyle \frac{2}{\pi }}{\displaystyle 𝑑\stackrel{~}{k}𝑑\stackrel{~}{r}\stackrel{~}{r}^2\frac{j_L(\stackrel{~}{k}r)j_L(\stackrel{~}{k}\stackrel{~}{r})}{k^2\stackrel{~}{k}^2}\mathrm{\Phi }_0(\stackrel{~}{r})}`$
$`=`$ $`\beta _0\sqrt{{\displaystyle \frac{2}{\pi }}}{\displaystyle 𝑑\stackrel{~}{k}\frac{j_L(\stackrel{~}{k}r)c_0^{(+)}(\stackrel{~}{k})}{k^2\stackrel{~}{k}^2}}`$
Integral (85) is easily obtained by using formula 7.421.4 from yielding
$`V_n^{(+)}`$ $`=`$ $`(1)^nV_0{\displaystyle \frac{(12\gamma )^n}{(1+2\gamma )^{n+L+3/2}}}q^L\mathrm{exp}({\displaystyle \frac{1}{2}}k^2{\displaystyle \frac{1}{1+2\gamma }})N_{nL}L_n^{L+1/2}({\displaystyle \frac{k^2}{14\gamma ^2}})`$ (89)
$`=`$ $`V_0{\displaystyle \frac{(12\gamma )^{n+L/2}}{(1+2\gamma )^{n+(L+3)/2}}}\mathrm{exp}({\displaystyle \frac{1}{2}}k^2{\displaystyle \frac{\gamma }{1+2\gamma }})c_n^{(+)}(k/\sqrt{14\gamma ^2})`$
If $`\gamma =\frac{1}{2}`$, then
$`V_n^{(+)}=V_0N_{nL}{\displaystyle \frac{1}{4}}{\displaystyle \frac{1}{n!}}({\displaystyle \frac{k}{2}})^{2n+L+1/2}\mathrm{exp}({\displaystyle \frac{1}{4}}k^2)`$ (90)
The second integral (86), reduced to the form
$`V_n^{()}=\beta _0{\displaystyle 𝑑\stackrel{~}{k}\frac{V_n^{(+)}(\stackrel{~}{k})c_0^{(+)}(\stackrel{~}{k})}{k^2\stackrel{~}{k}^2}}`$ (91)
can be expressed through the $`c_n^{()}`$ coefficients
$`V_n^{()}=V_0{\displaystyle \frac{\mathrm{exp}(\frac{1}{2}q^2\frac{\gamma }{2\gamma +1})}{(1+2\gamma )^{n+(L+3)/2}(1+\gamma )^{n+L/2}}}`$
$`{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \frac{\mathrm{\Gamma }(n+L+3/2)}{m!\mathrm{\Gamma }(nm+L+3/2)}}{\displaystyle \frac{N_{nL}}{N_{nm,L}}}[\gamma (1+2\gamma )]^mc_{nm}^{()}(q)`$ (92)
where
$`q=k\sqrt{{\displaystyle \frac{1+\gamma }{1+2\gamma }}}`$ (93)
New variables $`k^{}`$ and $`\stackrel{~}{k}^{}`$
$`k^{}=\sqrt{{\displaystyle \frac{1+\gamma }{1+2\gamma }}}k,\stackrel{~}{k}^{}=\sqrt{{\displaystyle \frac{1+\gamma }{1+2\gamma }}}\stackrel{~}{k},`$ (94)
and the following relation for Laguerre polynomials (see , vol. 2, formula 10.12(40))
$`L_n^\alpha (\lambda x)={\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \frac{\mathrm{\Gamma }(n+\alpha +1)}{m!\mathrm{\Gamma }(nm+\alpha +1)}}\lambda ^{nm}(1\lambda )^mL_{nm}^\alpha (x)`$ (95)
were introduced in (88) to obtain this result.
To obtain the recurrence relations for $`V_n^{(+)}`$ and $`V_n^{()}`$, we start from the equations for the functions $`\mathrm{\Psi }^{(+)}`$ and $`\mathrm{\Psi }^{()}`$
$`(\widehat{T}E)\mathrm{\Psi }^{(+)}`$ $`=`$ $`0,`$ (96)
$`(\widehat{T}E)\mathrm{\Psi }^{()}`$ $`=`$ $`\beta _0\mathrm{\Phi }_0(r)`$ (97)
Multiplying both sides of these equations by $`\mathrm{\Phi }_n(r)V(r)`$ and integrating, one obtains
$`<n|\widehat{V}\widehat{T}|\mathrm{\Psi }^{(+)}>EV_n^{(+)}`$ $`=`$ $`0,`$ (98)
$`<n|\widehat{V}\widehat{T}|\mathrm{\Psi }^{()}>EV_n^{()}`$ $`=`$ $`\beta _0<n\widehat{V}0>`$ (99)
On notices the similarity of the left hand side of both equations, leading to identical recurrence relations for both type of coefficients. The matrix element $`<n\widehat{V}0>`$ for the gaussian potential is
$`<n\widehat{V}0>=(1)^nV_0(1z)^nz^{L+3/2}\sqrt{{\displaystyle \frac{\mathrm{\Gamma }(n+L+3/2)}{n!\mathrm{\Gamma }(L+3/2)}}}`$ (100)
where
$`z=(\gamma +1)^1`$ (101)
Using the commutation relations of the operators $`\widehat{T}`$ and $`\widehat{V}`$, or the explicit form of $`V_n^{(+)}`$, one finally obtains
$`(1+2\gamma )^2T_{n,n+1}V_{n+1}^{(+)}+[(14\gamma ^2)T_{n,n}E]V_n^{(+)}`$ $`+`$ $`(12\gamma )^2T_{n,n1}V_{n1}^{(+)}`$ (102)
$`=`$ $`0`$
$`(1+2\gamma )^2T_{n,n+1}V_{n+1}^{()}+[(14\gamma ^2)T_{n,n}E]V_n^{()}`$ $`+`$ $`(12\gamma )^2T_{n,n1}V_{n1}^{()}`$ (103)
$`=`$ $`\beta _0<n|V|0>`$
If in (103) the potential is switched off ($`V_0=1`$ and $`z=1`$ (or $`\gamma =0`$)), the recurrence relations for $`V_n^{(+)}`$ and $`V_n^{()}`$ revert to these of $`c_n^{(+)}`$ and $`c_n^{()}`$, as is to be expected.
|
warning/0005/hep-ph0005104.html
|
ar5iv
|
text
|
# Effects of neutrino mixing on high-energy cosmic neutrino flux
## I Introduction
High-energy neutrino ($`E10^6`$ GeV) astroparticle physics is now a rapidly developing field impelled by the need for improved flux estimates as well as a good understanding of detector capabilities for all neutrino flavors, particularly in light of recently growing experimental support for flavor oscillations .
Currently envisaged astrophysical sources of high-energy cosmic neutrinos include, for instance, Active Galactic Nuclei (AGN) and Gamma Ray Burst fireballs (GRB). Production of high-energy cosmic neutrinos other than the AGNs and GRBs may also be possible .
High-energy neutrino production in cosmologically distant astrophysical systems such as AGNs and GRBs follow mainly from the production and decay of relevant unstable hadrons. These unstable hadrons may be produced mainly when the accelerated protons in these environments interact with the ambient photon field ($`p\gamma `$) and/or protons ($`pp`$) present there. The electron and muon neutrinos (and corresponding anti neutrinos) are mainly produced in the decay chain of charged pions whereas the tau neutrinos (and anti neutrinos) are mainly produced in the decay chain of charmed mesons in the same collisions at a suppressed level .
Previously the effects of vacuum neutrino mixing on the intrinsic ratios of high-energy cosmic neutrinos, in the context of three flavors, are briefly considered in . Here we update the description and consider three and four neutrino schemes with the most up-to-date constraints from the terrestrial, solar and atmospheric neutrino experiments. We also discuss the case where the intrinsic high-energy cosmic neutrino flux has nonstandard ratio which may not be obtained in charged pion decays.
The present study is particularly useful as the new underice/water Čerenkov light neutrino telescopes namely AMANDA and Baikal (also the ANTARES and NESTOR) as well as the large (horizontal) shower array detectors will have energy, angle and possibly flavor resolutions for high-energy neutrinos originating at cosmological distances . Several discussions are now available pointing towards the possibility of flavor identification for high-energy cosmic neutrinos .
The plan of this paper is as follows. In section II, we discuss the effects of vacuum neutrino mixing in three as well as four flavor schemes and numerically estimate the final (downward going) ratios of flux for high-energy cosmic neutrinos using the ranges of neutrino mixing parameters implied by recent searches for neutrino oscillations. In section III, we discuss the consequences of a hypothesis in which the intrinsic cosmic neutrino flux does not have the standard ratio \[$`F^0(\nu _e)`$: $`F^0(\nu _\mu )`$: $`F^0(\nu _\tau )=\mathrm{\hspace{0.17em}1}`$: $`2`$: 0\]. In section IV, we summarize our results.
## II Effects of neutrino mixing on high-energy cosmic neutrino flux
In $`p\gamma `$ and in $`pp`$ collisions, typically one obtains the following ratio of intrinsic high-energy cosmic neutrinos flux: $`F^0(\nu _e)`$: $`F^0(\nu _\mu )`$: $`F^0(\nu _\tau )=\mathrm{\hspace{0.17em}1}`$: $`2`$: $`<10^5`$. For simplicity, here we take these ratios as $`F^0(\nu _e)`$: $`F^0(\nu _\mu )`$: $`F^0(\nu _\tau )=\mathrm{\hspace{0.17em}1}`$: $`2`$: 0. We also discuss the effects of vacuum neutrino mixing by varying first two of these ratios from their above quoted values as this might be the case under some circumstances . We count both neutrinos and anti neutrinos in the symbol for neutrinos.
We consider an order of magnitude energy interval essentially around $`10^6`$ GeV since the currently available models for high-energy cosmic neutrinos give neutrino flux above the relevant atmospheric neutrino background specifically from AGN cores within this energy interval. Also in this energy range the flavor identification for high-energy cosmic neutrinos may be conceivable in new km<sup>2</sup> surface area neutrino telescopes . We have checked that for high-energy neutrinos originating from cosmologically distant sources, the change in the flavor composition of the high-energy cosmic neutrinos due to vacuum mixing is essentially energy independent for the entire energy range relevant for observations as the energy effects are averaged out in the relevant neutrino flavor oscillation probability expressions.
It has been pointed out that there are no matter effects on vacuum neutrino oscillations for relevant mass squared difference values \[$`𝒪(10^{10})\mathrm{\Delta }m^2/\text{eV}^2𝒪(1)`$\] for high-energy cosmic neutrinos scattering over the matter particles in the vicinity of sources . Nevertheless, for high-energy cosmic neutrinos scattering over the relic neutrinos, in halos or other wise, during their propagation, there may be relatively small (at most, of the order of few percent) matter enhanced flavor oscillation effects under the assumption of rather strong CP asymmetric neutrino background . However, given the current status of high-energy cosmic neutrino flux measurements, vacuum flavor oscillations remain the dominant effect and therefore from now on we consider here the effects of vacuum flavor oscillations only. For some other possible mixing effects, see . The typical distance to these astrophysical sources is taken as $`L\mathrm{\hspace{0.17em}10}^2`$ Mpc (where 1 pc $`\mathrm{\hspace{0.17em}3}10^{18}`$ cm). It is relevant here to mention that our following analysis is not necessarily restricted to high-energy cosmic neutrinos originating only from AGNs or GRBs as the above mentioned ratios of intrinsic neutrino flux can in principle be conceivable from some other cosmologically distant astrophysical sources as well . For simplicity, we assume the absence of relatively dense objects between the cosmologically distant high-energy neutrino sources and the prospective neutrino telescopes so as not to change significantly the vacuum oscillation pattern.
### A Three neutrino scheme
It has been known in the two flavor framework that the solar and the atmospheric neutrino problems are accounted for by neutrino oscillations with $`\mathrm{\Delta }m_{\mathrm{smaller}}^210^5`$eV<sup>2</sup> or $`10^{10}`$eV<sup>2</sup> and $`\mathrm{\Delta }m_{\mathrm{larger}}^210^{2.5}`$eV<sup>2</sup>, respectively. Without loss of generality we assume that $`|\mathrm{\Delta }m_{21}^2|<|\mathrm{\Delta }m_{32}^2|<|\mathrm{\Delta }m_{31}^2|`$ where $`\mathrm{\Delta }m_{ij}^2m_i^2m_j^2`$ ($`ij`$; $`i,j=1,2,3`$). If both the solar and the atmospheric neutrino problems are to be solved by energy dependent solutions, we have to have $`\mathrm{\Delta }m_{21}^2\mathrm{\Delta }m_{}^2`$ and $`\mathrm{\Delta }m_{32}^2\mathrm{\Delta }m_{\text{atm}}^2`$, i.e., we have mass hierarchy in this case. Here, the subscripts $``$ and atm stand for the mixing angles for the solar and atmospheric neutrino oscillations, respectively.
In the three flavor framework, the flavor and mass eigenstates are related by the MNS matrix U :
$`\left(\begin{array}{c}\nu _e\\ \nu _\mu \\ \nu _\tau \end{array}\right)=U\left(\begin{array}{c}\nu _1\\ \nu _2\\ \nu _3\end{array}\right),U`$ $``$ $`\left(\begin{array}{ccc}U_{e1}& U_{e2}& U_{e3}\\ U_{\mu 1}& U_{\mu 2}& U_{\mu 3}\\ U_{\tau 1}& U_{\tau 2}& U_{\tau 3}\end{array}\right).`$ (10)
In the present convention for the mass pattern, the disappearance probability for the CHOOZ experiment is given by
$`P(\overline{\nu }_e\overline{\nu }_e)=14|U_{e3}|^2(1|U_{e3}|^2)\mathrm{sin}^2\left({\displaystyle \frac{\mathrm{\Delta }m_{32}^2L}{4E}}\right),`$ (11)
and combining the negative result by the CHOOZ experiment with the atmospheric neutrino data, we have (for quantitative discussions, see ):
$`|U_{e3}|^21.`$ (12)
As was mentioned in the Introduction, the matter effect is irrelevant in our discussions, so let us consider the vacuum oscillation probability in the case where the oscillation length is very small as compared to the distance between the source and the detector \[hereafter referred to as far distance (approximation)\]. In vacuum, it is well known that the flavor oscillation probability is given by
$`P(\nu _\alpha \nu _\beta ;L)=\delta _{\alpha \beta }{\displaystyle \underset{jk}{}}U_{\alpha j}^{}U_{\beta j}U_{\alpha k}U_{\beta k}^{}\left(1e^{i\mathrm{\Delta }E_{jk}L}\right).`$ (13)
In the limit $`L\mathrm{}`$, we have
$`P(\nu _\alpha \nu _\beta ;L=\mathrm{})=\delta _{\alpha \beta }{\displaystyle \underset{jk}{}}U_{\alpha j}^{}U_{\beta j}U_{\alpha k}U_{\beta k}^{}`$ $`=`$ $`{\displaystyle \underset{j}{}}|U_{\alpha j}|^2|U_{\beta j}|^2,`$ (14)
where we have averaged over rapid oscillations. Thus, we can represent the oscillation probability as a symmetric matrix $`P`$ and $`P`$ can be written as a product of a matrix $`A`$:
$`P\left(\begin{array}{ccc}P_{ee}& P_{e\mu }& P_{e\tau }\\ P_{e\mu }& P_{\mu \mu }& P_{\mu \tau }\\ P_{e\tau }& P_{\mu \tau }& P_{\tau \tau }\end{array}\right)AA^T,`$ (18)
with
$`A\left(\begin{array}{ccc}|U_{e1}|^2& |U_{e2}|^2& |U_{e3}|^2\\ |U_{\mu 1}|^2& |U_{\mu 2}|^2& |U_{\mu 3}|^2\\ |U_{\tau 1}|^2& |U_{\tau 2}|^2& |U_{\tau 3}|^2\end{array}\right).`$ (22)
Now, the cosmic neutrino flux in the far distance can be expressed as a product of $`P`$ and the intrinsic flux $`F^0(\nu _\alpha )(\alpha =e,\mu ,\tau )`$:
$`\left(\begin{array}{c}F(\nu _e)\\ F(\nu _\mu )\\ F(\nu _\tau )\end{array}\right)=P\left(\begin{array}{c}F^0(\nu _e)\\ F^0(\nu _\mu )\\ F^0(\nu _\tau )\end{array}\right)=AA^T\left(\begin{array}{c}F^0(\nu _e)\\ F^0(\nu _\mu )\\ F^0(\nu _\tau )\end{array}\right).`$ (32)
Throughout this section we assume the standard ratio of the intrinsic cosmic neutrino flux, i.e., $`F^0(\nu _e)`$: $`F^0(\nu _\mu )`$: $`F^0(\nu _\tau )=\mathrm{\hspace{0.17em}1}`$: $`2`$: 0, so that, we get
$`A^T\left(\begin{array}{c}F^0(\nu _e)\\ F^0(\nu _\mu )\\ F^0(\nu _\tau )\end{array}\right)`$ $`=`$ $`\left(\begin{array}{ccc}|U_{e1}|^2& |U_{\mu 1}|^2& |U_{\tau 1}|^2\\ |U_{e2}|^2& |U_{\mu 2}|^2& |U_{\tau 2}|^2\\ |U_{e3}|^2& |U_{\mu 3}|^2& |U_{\tau 3}|^2\end{array}\right)\left(\begin{array}{c}1\\ 2\\ 0\end{array}\right)F^0(\nu _e),`$ (42)
$`=`$ $`\left(\begin{array}{c}1\\ 1\\ 1\end{array}\right)F^0(\nu _e)+\left(\begin{array}{c}|U_{\mu 1}|^2|U_{\tau 1}|^2\\ |U_{\mu 2}|^2|U_{\tau 2}|^2\\ |U_{\mu 3}|^2|U_{\tau 3}|^2\end{array}\right)F^0(\nu _e),`$ (49)
where we have used the unitarity condition, i.e., $`_j|U_{\alpha j}|^2=`$1. When $`|U_{e3}|^21`$ and $`|U_{\mu 3}||U_{\tau 3}|`$, we have $`||U_{\mu j}|^2|U_{\tau j}|^2|1(j=1,2,3)`$, so the second term in Eq. (49) is small. Hence with the constraints of the solar and atmospheric neutrino and the reactor data, we obtain
$`\left(\begin{array}{c}F(\nu _e)\\ F(\nu _\mu )\\ F(\nu _\tau )\end{array}\right)A\left(\begin{array}{c}1\\ 1\\ 1\end{array}\right)F^0(\nu _e)\left(\begin{array}{c}1\\ 1\\ 1\end{array}\right)F^0(\nu _e),`$ (59)
where we have used the unitarity condition again. Therefore, we conclude that the ratio of the cosmic high-energy neutrino fluxes in the far distance is 1: 1: 1, irrespective of which solar solution is chosen. This is because the MNS matrix elements satisfy $`|U_{e3}|^21`$, $`|U_{\mu 3}||U_{\tau 3}|`$ and $`F^0(\nu _e)`$: $`F^0(\nu _\mu )`$: $`F^0(\nu _\tau )=\mathrm{\hspace{0.17em}1}`$: $`2`$: 0. This is the feature that is expected to be observed at the new km<sup>2</sup> surface area neutrino telescopes in the case of the three neutrino oscillation scheme.
Using the allowed region for the MNS matrix elements given in , we have evaluated the final ratio of the cosmic neutrino flux numerically. To plot the ratio of the three neutrino flavors, we introduce the triangle representation, which is introduced by Fogli, Lisi, and Scioscia in a different context. In Fig. 1, a unit regular triangle is drawn and the position of the point gives the ratio of the final high-energy neutrino flux with $`F_\alpha F(\nu _\alpha )`$. One reason that we adopt this representation is because currently we do not know the precise total cosmic neutrino flux, while we may observe the ratio of flux of different cosmic neutrino flavors experimentally to a certain extent. Using this triangle representation, the allowed region in the three flavor framework with the constraints from the terrestrial, solar and atmospheric neutrino data, is given in Figs. 2a and 2b. The allowed region is a small area around the midpoint $`F(\nu _e)=F(\nu _\mu )=F(\nu _\tau )=1/3`$ and the small deviation from the midpoint indicates the smallness of $`|U_{e3}|`$ and $`||U_{\mu j}|^2|U_{\tau j}|^2|`$.
### B Four neutrino scheme
Let us now consider the case with three active and one sterile neutrino. Schemes with sterile neutrinos have been studied by a number of authors . Here, we are interested in the four neutrino scheme in which the solar, atmospheric and LSND experiment are all explained by neutrino oscillations. By generalizing the discussion on Big-Bang Nucleosynthesis (BBN) from the two neutrino scheme to four neutrino case, it has been shown that the neutrino mixing angles are strongly constrained not only by the reactor data but also by BBN if one demands that the number $`N_\nu `$ of effective neutrinos be less than four. In this case, the $`4\times 4`$ MNS matrix splits approximately into two $`2\times 2`$ block diagonal matrices.
On the other hand, some authors have given conservative estimate for $`N_\nu `$ and if their estimate is correct then we no longer have strong constraints on the neutrino mixing angles. In this section, we discuss the final ratio of the cosmic neutrino flux components with and without BBN constraints separately. Throughout this section we assume the following mass pattern
$`m_1^2m_2^2m_3^2m_4^2,`$ (60)
with corresponding $`\mathrm{\Delta }m_{21}^2`$, $`\mathrm{\Delta }m_{43}^2`$ and $`\mathrm{\Delta }m_{32}^2`$ stand for the mass squared differences suggested by the solar, atmospheric neutrino deficits and the LSND experiment, respectively.
#### 1 Four neutrino scheme with BBN constraints
Here, we adopt the notation in for the $`4\times 4`$ MNS matrix:
$`U`$ $``$ $`\left(\begin{array}{cccc}U_{e1}& U_{e2}& U_{e3}& U_{e4}\\ U_{\mu 1}& U_{\mu 2}& U_{\mu 3}& U_{\mu 4}\\ U_{\tau 1}& U_{\tau 2}& U_{\tau 3}& U_{\tau 4}\\ U_{s1}& U_{s2}& U_{s3}& U_{s4}\end{array}\right),`$ (65)
$``$ $`R_{34}({\displaystyle \frac{\pi }{2}}\theta _{34})R_{24}(\theta _{24})R_{23}({\displaystyle \frac{\pi }{2}})e^{2i\delta _1\lambda _3}R_{23}(\theta _{23})e^{2i\delta _1\lambda _3}e^{\sqrt{6}i\delta _3\lambda _{15}/2}`$ (66)
$`\times `$ $`R_{14}(\theta _{14})e^{\sqrt{6}i\delta _3\lambda _{15}/2}e^{2i\delta _2\lambda _8/\sqrt{3}}R_{13}(\theta _{13})e^{2i\delta _2\lambda _8/\sqrt{3}}R_{12}(\theta _{12}),`$ (67)
where $`c_{ij}\mathrm{cos}\theta _{ij}`$, $`s_{ij}\mathrm{sin}\theta _{ij}`$ and
$`R_{jk}(\theta )\mathrm{exp}\left(iT_{jk}\theta \right),`$ (68)
is a $`4\times 4`$ orthogonal matrix with
$`\left(T_{jk}\right)_\mathrm{}m=i\left(\delta _j\mathrm{}\delta _{km}\delta _{jm}\delta _k\mathrm{}\right),`$ (69)
and $`2\lambda _3\mathrm{diag}(1,1,0,0)`$, $`2\sqrt{3}\lambda _8\mathrm{diag}(1,1,2,0)`$, $`2\sqrt{6}\lambda _{15}\mathrm{diag}(1,1,1,3)`$ are diagonal elements of the $`SU(4)`$ generators. From the constraint of the reactor data of Bugey , which strongly constrain the disappearance probability $`P(\overline{\nu }_e\overline{\nu }_e)`$ for the entire region of mass squared difference implied by the data of LSND and E776 , we have
$`|U_{e3}|^2+|U_{e4}|^21.`$ (70)
On the other hand, from the BBN constraint that sterile neutrino be not in thermal equilibrium, we get
$`|U_{s3}|^2+|U_{s4}|^21.`$ (71)
In this case, the oscillation probabilities in the far distance are given essentially by the following two neutrino flavor formulae:
$`P(\nu _e\nu _e;L=\mathrm{})`$ $`=`$ $`1{\displaystyle \frac{1}{2}}|U_{e1}|^2|U_{e2}|^2=1{\displaystyle \frac{1}{2}}\mathrm{sin}^22\theta _{12},`$ (72)
$`P(\nu _e\nu _s;L=\mathrm{})`$ $`=`$ $`{\displaystyle \frac{1}{2}}|U_{e1}|^2|U_{e2}|^2={\displaystyle \frac{1}{2}}\mathrm{sin}^22\theta _{12},`$ (73)
$`P(\nu _\mu \nu _\mu ;L=\mathrm{})`$ $`=`$ $`1{\displaystyle \frac{1}{2}}|U_{\mu 3}|^2|U_{\mu 4}|^2=1{\displaystyle \frac{1}{2}}\mathrm{sin}^22\theta _{24},`$ (74)
$`P(\nu _\mu \nu _\tau ;L=\mathrm{})`$ $`=`$ $`{\displaystyle \frac{1}{2}}|U_{\mu 3}|^2|U_{\mu 4}|^2={\displaystyle \frac{1}{2}}\mathrm{sin}^22\theta _{24},`$ (75)
where, we have substituted $`\theta _{13}=\theta _{14}=\theta _{23}=\theta _{34}=0`$ in Eq. (67) and $`\theta _{12}`$, $`\theta _{24}`$ correspond to the mixing angles of the solar and atmospheric neutrino oscillations, respectively. It has been known that only the Small Mixing Angle (SMA) MSW solution is allowed in the $`\nu _e\nu _s`$ solar oscillation scheme, so in the present case, we take $`|\theta _{12}|1`$. Thus, we have the following ratio of the final cosmic high-energy neutrino flux:
$`\left(\begin{array}{c}F(\nu _e)\\ F(\nu _\mu )\\ F(\nu _\tau )\\ F(\nu _s)\end{array}\right)`$ $`=`$ $`\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& c_{\text{atm}}^2& s_{\text{atm}}^2\\ 0& 0& s_{\text{atm}}^2& c_{\text{atm}}^2\\ 0& 1& 0& 0\end{array}\right)\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& 0& 1\\ 0& c_{\text{atm}}^2& s_{\text{atm}}^2& 0\\ 0& s_{\text{atm}}^2& c_{\text{atm}}^2& 0\end{array}\right)\left(\begin{array}{c}1\\ 2\\ 0\\ 0\end{array}\right)F^0(\nu _e),`$ (92)
$`=`$ $`\left(\begin{array}{c}1\\ 2\mathrm{sin}^22\theta _{\text{atm}}\\ \mathrm{sin}^22\theta _{\text{atm}}\\ 0\end{array}\right)F^0(\nu _e),`$ (97)
where, we have taken $`\theta _{24}\theta _{\text{atm}}`$ which satisfies
$`0.88\mathrm{sin}^22\theta _{\text{atm}}1.`$ (98)
In this case, therefore, we again have $`F(\nu _e)`$: $`F(\nu _\mu )`$: $`F(\nu _\tau )\mathrm{\hspace{0.17em}1}`$: $`1`$: 1.
We have also numerically obtained the allowed region by letting $`\theta _{12}`$, $`\theta _{13}`$, $`\theta _{14}`$, $`\theta _{23}`$, $`\theta _{34}`$ and $`\theta _{24}`$ to vary within the constraints obtained from the reactor, solar and atmospheric neutrino data. In the four neutrino scheme, the total flux of active neutrino is in general not the same as that at the production point, and in principle we have to use a unit tetrahedron to express the ratio of the four neutrino flux. In practice, however, we may not observe the cosmic sterile neutrino events nor do we know the precise total flux of the cosmic high-energy neutrinos, so it is useful to normalize the flux of each active neutrino by the total flux of active neutrinos:
$`\left(\begin{array}{c}\stackrel{~}{F}(\nu _e)\\ \stackrel{~}{F}(\nu _\mu )\\ \stackrel{~}{F}(\nu _\tau )\end{array}\right){\displaystyle \frac{1}{F(\nu _e)+F(\nu _\mu )+F(\nu _\tau )}}\left(\begin{array}{c}F(\nu _e)\\ F(\nu _\mu )\\ F(\nu _\tau )\end{array}\right).`$ (105)
After redefining the flux, we can plot the ratio of each active neutrino with the same triangle graph as in the three neutrino case, and the result is depicted in Fig. 3. The allowed region is again small and the reason that the region lies horizontally is because of possible deviation of $`\theta _{24}\theta _{\text{atm}}`$ from $`\pi /4`$ \[see Eq. (98)\].
#### 2 Four neutrino scheme without BBN constraints
In this subsection, we discuss what happens to the ratio of the final cosmic high-energy neutrino flux if we lift BBN constraints. Without BBN constraints, the only conditions we have to take into account are the solar and atmospheric neutrino deficit data and the results of experiment of LSND (the appearance experiment of $`\overline{\nu }_\mu \overline{\nu }_e`$) , Bugey (the disappearance experiment of $`\overline{\nu }_e\overline{\nu }_e`$) and CDHSW (the disappearance experiment of $`\overline{\nu }_\mu \overline{\nu }_\mu `$) . For the range of the $`\mathrm{\Delta }m^2`$ suggested by the LSND data, which is given by 0.2$`\mathrm{\Delta }m_{\text{LSND}}^2/\text{eV}^2`$ 2 when combined with the data of Bugey and E776 , the constraint by the Bugey data is very stringent and
$`|U_{e3}|^2+|U_{e4}|^210^2,`$ (106)
has to be satisfied . Therefore, for simplicity, we take $`U_{e3}=U_{e4}=0`$, in the following discussion.
The analysis of the solar neutrino data in the four neutrino scheme with ansatz $`U_{e3}=U_{e4}=0`$ has been done recently by Giunti, Gonzalez-Garcia and Peña-Garay . They have shown that the scheme is reduced to the two neutrino framework in which only one free parameter $`c_s|U_{s1}|^2+|U_{s2}|^2`$ appears in the analysis<sup>*</sup><sup>*</sup>*In the notation of this parameter is given by $`c_s=c_{23}^2c_{24}^2`$. We adopt different notation from for the parametrization of the mixing angles, however, will use $`c_s`$ for $`|U_{s1}|^2+|U_{s2}|^2`$.. Their conclusion is that the SMA (MSW) solution exists for the entire region of $`0c_s1`$, while the Large Mixing Angle (LMA) and Vacuum Oscillation (VO) solutions survive only for $`0c_s0.2`$ and $`0c_s0.4`$, respectively.
The analysis of the atmospheric neutrino data in the four neutrino framework has been done by one of the authors more recently again with ansatz $`U_{e3}=U_{e4}=0`$. The conclusion of is that the region $`\pi /2\theta _{34}\pi /2`$ and $`0\theta _{23}\pi /6`$ as well as $`0\delta _1\pi `$ is consistent with the Superkamiokande atmospheric neutrino data of the contained and upward going $`\mu `$ events, where $`\theta _{34}`$ and $`\theta _{23}`$ stand for, roughly speaking, the ratio of $`\nu _\mu \nu _\tau `$ versus $`\nu _\mu \nu _s`$ and the ratio of the contributions of $`\mathrm{sin}^2\left(\mathrm{\Delta }m_{\text{atm}}^2L/4E\right)`$ versus $`\mathrm{sin}^2\left(\mathrm{\Delta }m_{\text{LSND}}^2L/4E\right)`$ in the flavor oscillation probability, respectively, and $`\delta _1`$ is the only CP phase left in this scheme. Notice that the recent claim by the Superkamiokande group that $`\nu _\mu \nu _s`$ is almost completely excluded is based on the hypothesis of two neutrino $`\nu _\mu \nu _s`$ oscillations with only one mass squared difference $`\mathrm{\Delta }m_{\text{atm}}^2`$, and their claim is completely consistent with the results in , where the region $`\theta _{34}\pm \pi /2`$ and $`\theta _{23}0`$, which would lead to pure $`\nu _\mu \nu _s`$ oscillations, is excluded. For generic mixing angles of $`\theta _{34}`$ and $`\theta _{23}`$, the ansatz in implies hybrid of $`\nu _\mu \nu _\tau `$ and $`\nu _\mu \nu _s`$ oscillations in general, and one has to take into account the constraint of the CDHSW experiment also. Following , we take $`\mathrm{\Delta }m_{32}^2\mathrm{\Delta }m_{\text{LSND}}^2=`$ 0.3 eV<sup>2</sup>, which is consistent with the negative result of the disappearance experiment of CDHSW for the entire region of the mixing angles obtained in .
Combining the results of on solar neutrinos and on atmospheric neutrinos, we have evaluated the final ratio of the cosmic high-energy neutrino flux and the result is given in Figs. 4a and 4b, where the allowed region with the SMA (MSW), LMA (MSW) and VO solutions are shown separately. Almost the entire allowed region with the SMA (MSW) solution lies above the line $`\stackrel{~}{F}(\nu _e)=1/3`$, and this is because the total normalization $`F(\nu _e)+F(\nu _\mu )+F(\nu _\tau )=1F(\nu _s)`$ becomes less than 1 while $`F(\nu _e)=(1\mathrm{sin}^22\theta _{12})F^0(\nu _e)`$ hardly changes due to smallness of $`|\theta _{12}|=|\theta _{}|`$. On the other hand, for most of the region with LMA (MSW) and VO solutions, $`F(\nu _e)/F^0(\nu _e)`$ becomes smaller than $`1F(\nu _s)`$ and the region lies below the line $`\stackrel{~}{F}(\nu _e)=1/3`$. This four neutrino scheme without BBN constraint gives us clearly a distinctive pattern for the final ratio of the cosmic high-energy neutrino flux, and observationally, if we have good precision then it may be even possible to distinguish the SMA (MSW) solution from the LMA (MSW) or VO solutions. The allowed regions in Figs. 4a and 4b are plotted for $`\delta _1=0`$ and $`\delta _1=\pi /2`$. We observe that most of the allowed regions for $`\delta _1=0`$ and $`\delta _1=\pi /2`$ overlap with each other and it implies that distinction between $`\delta _1=0`$ and $`\delta _1=\pi /2`$ is difficult in this scheme of four neutrinos.
We note in passing that the allowed region of the LOW solution to the solar netrino problem is contained in that of the LMA solution as far as $`\mathrm{sin}^22\theta _{}`$ is concerned, and therefore the LOW solution gives us only a subset of the allowed region of the LMA solution in Figs. 4a and 4b.
## III Nonstandard ratio of the high-energy neutrino flux
In section II, we have seen that the schemes of three neutrinos and of four neutrinos with BBN constraints give us the ratio $`F(\nu _e)`$: $`F(\nu _\mu )`$: $`F(\nu _\tau )\mathrm{\hspace{0.17em}1}`$: $`1`$: 1, irrespective of which solar solution is chosen. This is due to the fact that the intrinsic high-energy cosmic neutrino flux has the ratio $`F^0(\nu _e)`$: $`F^0(\nu _\mu )`$: $`F^0(\nu _\tau )=\mathrm{\hspace{0.17em}1}`$: $`2`$: 0 and the oscillation of atmospheric neutrinos is with maximal mixing while $`|U_{e3}|^21`$. It has been pointed out that the intrinsic flux of the cosmic high-energy neutrinos may not have the standard ratio $`F^0(\nu _e)`$: $`F^0(\nu _\mu )`$: $`F^0(\nu _\tau )=\mathrm{\hspace{0.17em}1}`$: $`2`$: 0, mainly because some of muons may lose their energy in a magnetic field. Here, we discuss in a model independent way the consequences of a generic scenario which is characterized by $`F^0(\nu _e)`$: $`F^0(\nu _\mu )`$: $`F^0(\nu _\tau )=\lambda /3`$: $`1\lambda /3`$: 0, where $`\lambda `$ is a free parameter $`0\lambda 1`$ (the standard ratio is obtained for $`\lambda =1`$).
We have plotted the allowed region in Figs. 5a and 5b for $`\lambda =1/2`$ and $`\lambda =0`$, respectively. We observe that the cosmic high-energy neutrino flux with the nonstandard ratio gives relatively lower value of $`\stackrel{~}{F}(\nu _e)`$. If relatively lower value of cosmic $`\nu _e`$ flux is observed in the future experiments, then it may imply that the possible oscillation scenario is the four neutrino scheme without BBN constraints and the intrinsic cosmic neutrino flux with the nonstandard ratio. Independent information on neutrino mixing parameters as well as on relevant astrophysical inputs may be needed here to arrive at a more definite conclusion.
## IV Results and Discussion
In general, the final flux of high-energy cosmic neutrinos is expected to be almost equally distributed among the three (active) cosmic neutrino flavors because of vacuum flavor mixing/oscillations provided the astrophysical sources for these high-energy cosmic neutrinos are cosmologically distant, essentially irrespective of the neutrino flavors (three or four). Nevetheless, this may not be the case if the intrinsic high-energy cosmic neutrino flux ratios differ from the standard one, namely from $`F^0(\nu _e)`$: $`F^0(\nu _\mu )`$: $`F^0(\nu _\tau )=\mathrm{\hspace{0.17em}1}`$: $`2`$: 0. In the examples considered in this work for the nonstandard intrinsic high-energy cosmic neutrino flux, a relatively lower final flux for cosmic $`\nu _e`$ is obtained. The situation of nonstandard intrinsic cosmic neutrino flux may arise, for instance, if some of the muons lose their energy in the relatively intense magnetic field in the vicinity of the source.
A simultaneous measurement of the three cosmic neutrino flavors may be useful to obtain information about a particular neutrino (mass and) mixing scheme depending on the relevant achievable resolutions for typical km<sup>2</sup> surface area neutrino telescopes.
Irrespective of the numbers of neutrino flavors, in each of the neutrino (mass and) mixing scheme discussed in this work, the final flux of high-energy cosmic tau neutrinos is essentially comparable to that of non tau (active) neutrinos, even if it is intrinsically negligible. This may, at least in principle, be useful to constrain the relevant nonstandard particle physics/astro physics scenarios.
In this work, we have considered the effects of vacuum neutrino flavor oscillations on high-energy cosmic neutrino flux in the context of three as well as four flavors. These oscillations result in an energy independent ratio, $`R_{\alpha \beta }N_\alpha /N_\beta (\alpha \beta ;\alpha =\beta =\nu _e,\nu _\mu ,\nu _\tau )`$ of the number of events detected for the neutrino flavors $`\alpha `$ and $`\beta `$. It is so because the various neutrino flavor precession probabilities given in section II \[see Eq. (5) and Eq. (17)\] are essentially independent of neutrino energy. In the following paragraph, we briefly describe the prospects offered by the typical km<sup>2</sup> surface area under ice/water neutrino telescopes which are currently under construction/planning to possibly identify the cosmic neutrino flavor and hence to determine the ratio, $`R_{\alpha \beta }`$ .
We ignore the possible observational difference between cosmic neutrinos and anti neutrinos for simplicity in the following discussion and assume that the flavor content in the cosmic neutrino flux is equally distributed because of vacuum flavor oscillations. Several of the recent discussions suggest that the absorption of high-energy cosmic neutrino flux by earth is neutrino flavor dependent . The upward going electron and muon neutrino fluxes are significantly attenuated typically for $`E_0510^4`$ GeV, whereas the upward going tau neutrinos with $`E>E_0`$ may reach the detector with $`EE_0`$ because of the short life time of the associated tau lepton and may appear as a rather small pile up with fairly flat zenith angle dependence. For $`EE_0`$, the upward going cosmic neutrino event rates range typically as: $`N_{\nu _\mu }𝒪(10^1)`$ whereas $`N_{\nu _\tau }𝒪(10^0)`$ in units of per year per steradian for typical km<sup>2</sup> surface area neutrino telescopes, if one uses the current upper high-energy cosmic neutrino flux limits . Let us note that for $`EE_0`$, presently the high-enegy cosmic neutrino fluxes from AGNs dominate above the atmospheric neutrino background. For downward going high-enegy cosmic neutrino flux, the event rate ranges typically as: $`N_{\nu _e}𝒪(10^{1.5}),N_{\nu _\mu }𝒪(10^2)`$, whereas $`N_{\nu _\tau }𝒪(10^1)`$ in units of per year per steradian for the same high-energy cosmic neutrino fluxes. For $`E>E_0`$, the downward going cosmic tau neutrinos typically produce a two bang event topology such that the two bangs are connected by a $`\mu `$-like track. The size of the second bang being on the average a factor of two larger than the first bang. The downward going electron neutrinos produce a single bang at these energies whereas the muon neutrinos typically produce a single shower alongwith a zipping $`\mu `$-like track in km<sup>2</sup> surface area neutrino telescopes. Based on this rather distinct event topologies, cosmic neutrino flavor identification may be conceivable. The above order of magnitude estimates indicate that the typical km<sup>2</sup> surface area neutrino telescopes do offer some prospects for observations of high-energy cosmic neutrino flavor ratio, $`R_{\alpha \beta }`$, or at least may constrain it meaningfully.
##### Acknowledgments
The work of H. A. is supported by a Japan Society for the Promotion of Science fellowship. The authors thank the hospitality of Summer Institute 99 at Yamanashi, Japan where this work was started. This research was supported in part by a Grant-in-Aid for Scientific Research of the Ministry of Education, Science and Culture, Japan (#12047222, #10640280).
Figures
The representation of the ratio of the final flux of (downward going) high-energy cosmic neutrinos on earth with a unit regular triangle. The $`x`$ and $`y`$ coordinates of the points are given by $`x=(2F_\tau +F_e)/\sqrt{3}`$, $`y=F_e`$ and the inside region of the triangle is given by $`0x2/\sqrt{3}`$, $`0y1`$. The point with an asterisk stands for the ratio without oscillations $`F(\nu _e)`$: $`F(\nu _\mu )`$: $`F(\nu _\tau )=\mathrm{\hspace{0.17em}1}`$: $`2`$: 0 and is given by the coordinate $`(1/3\sqrt{3},1/3)`$.
The ratio of the final flux of high-energy cosmic neutrinos in the far distance in the three neutrino scheme. The allowed region is the inside of each contour; (b) is an enlarged figure of (a). The allowed region lies near the mid point $`(1/\sqrt{3},1/3)`$.
The ratio of the final flux of high-energy cosmic neutrinos in the far distance in the four neutrino scheme with BBN constraints. The allowed region lies near the mid point $`(1/\sqrt{3},1/3)`$.
The ratio of the final flux of high-energy cosmic neutrinos in the far distance in the four neutrino scheme without BBN constraints; (b) is an enlarged figure of (a). The allowed region of the LOW solution is a subset of that of the LMA solution.
The ratio of the final flux of high-energy cosmic neutrinos in the far distance in a nonstandard scenario characterized by $`F^0(\nu _e)`$: $`F^0(\nu _\mu )`$: $`F^0(\nu _\tau )=\lambda /3`$: $`1\lambda /3`$: 0. In Fig. 5(a), $`\lambda =1/2`$, the cases of $`N_\nu =3`$ with SMA and $`N_\nu =4`$ with BBN constraints have small region near the point $`F(\nu _e)`$: $`F(\nu _\mu )`$: $`F(\nu _\tau )=`$ 2: 5: 5, the region for the cases of $`N_\nu =3`$ with LMA and VO lie above this point, whereas most of the region for the case of $`N_\nu =4`$ without BBN constraints lie to the left of this point. As in Fig.4, the allowed region of the LOW solution is a subset of that of the LMA solution. In Fig. 5(b), $`\lambda =0`$, the cases of $`N_\nu =3`$ with SMA and $`N_\nu =4`$ with BBN constraints have small region near the point $`F(\nu _e)`$: $`F(\nu _\mu )`$: $`F(\nu _\tau )=\mathrm{\hspace{0.17em}0}`$: $`1`$: 1, the region for the cases of $`N_\nu =3`$ with LMA and VO lie above this point, whereas most of the region for the case of $`N_\nu =4`$ without BBN constraints lie to the left of this point.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.